IN-HAND OBJECT POSE TRACKING

Abstract

Apparatuses, systems, and techniques are described that estimate the pose of an object while the object is being manipulated by a robotic appendage. In at least one embodiment, a sample-based optimization algorithm tracks in-hand object poses during manipulation via contact feedback and a GPU-accelerated robotic simulation is developed. In at least one embodiment, parallel simulations concurrently model object pose changes that may be caused by complex contact dynamics. In at least one embodiment, the optimization algorithm tunes simulation parameters during object pose tracking to further improve tracking performance. In various embodiments, real-world contact sensing may be improved by utilizing vision in-the-loop.

Description

TECHNICAL FIELD

At least one embodiment pertains to training and simulating robots to perform and facilitate tasks. For example, at least one embodiment pertains to training and simulating robots using artificial intelligence according to various novel techniques described herein.

BACKGROUND

Training and simulating robots to accurately perform tasks can use significant memory, time, or computing resources. Training a robotic control system to track the pose of an object held and manipulated by a robot hand is challenging for vision-based object pose tracking systems, because the object is under significant occlusion while the robot hand is holding it. Such occlusion reduces the amount of data that can be used in determining what movements a robot should make, creating risks that whatever task is being performed will be done incorrectly and/or inefficiently, perhaps damaging the object or other objects in the environment in the process. Such tracking is especially complex due to the fact that objects sometimes slip or otherwise move during the process, creating changes in the object's orientation that can go undetected and, therefore, unaccounted for. The amount of memory, time, or computing resources used to accurately train and simulate robots can be improved.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 illustrates an example of teleoperation across various tasks, according to at least one embodiment;

FIG. 2 illustrates an example of a robot with tactile sensors, according to at least one embodiment;

FIG. 3 illustrates an example of a system that tracks objects in real-time, according to at least one embodiment;

FIG. 4 illustrates an example of estimating hand pose, according to at least one embodiment;

FIG. 5 illustrates an example of a human hand pose, and a robotic gripper performing a corresponding pose, according to at least one embodiment;

FIG. 6 illustrates an example of human hand poses, and corresponding robot gripper poses, according to at least one embodiment;

FIG. 7 illustrates an example of an in-hand object pose tracking framework, according to at least one embodiment;

FIG. 8 illustrates an example of a comparison of optimizers, according to at least one embodiment;

FIG. 9 illustrates an example of an algorithm utilized by a system, according to at least one embodiment;

FIG. 10 illustrates a first example of results of ablation studies, according to at least one embodiment;

FIG. 11 illustrates a second example of results of ablation studies, according to at least one embodiment;

FIG. 12 illustrates an example of real-world experiment results, according to at least one embodiment;

FIG. 13 illustrates an example of a process that, as a result of being performed by a computer system, determines the pose of an object being manipulated by a robotic hand equipped with tactile force sensors;

FIG. 14A illustrates inference and/or training logic, according to at least one embodiment;

FIG. 14B illustrates inference and/or training logic, according to at least one embodiment;

FIG. 15 illustrates training and deployment of a neural network, according to at least one embodiment;

FIG. 16 illustrates an example data center system, according to at least one embodiment;

FIG. 17A illustrates an example of an autonomous vehicle, according to at least one embodiment;

FIG. 17B illustrates an example of camera locations and fields of view for the autonomous vehicle of FIG. 17A, according to at least one embodiment;

FIG. 17C is a block diagram illustrating an example system architecture for the autonomous vehicle of FIG. 17A, according to at least one embodiment;

FIG. 17D is a diagram illustrating a system for communication between cloud-based server(s) and the autonomous vehicle of FIG. 17A, according to at least one embodiment;

FIG. 18 is a block diagram illustrating a computer system, according to at least one embodiment;

FIG. 19 is a block diagram illustrating computer system, according to at least one embodiment;

FIG. 20 illustrates a computer system, according to at least one embodiment;

FIG. 21 illustrates a computer system, according at least one embodiment;

FIG. 22A illustrates a computer system, according to at least one embodiment;

FIG. 22B illustrates a computer system, according to at least one embodiment;

FIG. 22C illustrates a computer system, according to at least one embodiment;

FIG. 22D illustrates a computer system, according to at least one embodiment;

FIGS. 22E and 22F illustrate a shared programming model, according to at least one embodiment;

FIG. 23 illustrates exemplary integrated circuits and associated graphics processors, according to at least one embodiment;

FIGS. 24A and 24B illustrate exemplary integrated circuits and associated graphics processors, according to at least one embodiment;

FIGS. 25A and 25B illustrate additional exemplary graphics processor logic according to at least one embodiment;

FIG. 26 illustrates a computer system, according to at least one embodiment;

FIG. 27A illustrates a parallel processor, according to at least one embodiment;

FIG. 27B illustrates a partition unit, according to at least one embodiment;

FIG. 27C illustrates a processing cluster, according to at least one embodiment;

FIG. 27D illustrates a graphics multiprocessor, according to at least one embodiment;

FIG. 28 illustrates a multi-graphics processing unit (GPU) system, according to at least one embodiment;

FIG. 29 illustrates a graphics processor, according to at least one embodiment;

FIG. 30 is a block diagram illustrating a processor micro-architecture for a processor, according to at least one embodiment;

FIG. 31 illustrates a deep learning application processor, according to at least one embodiment;

FIG. 32 is a block diagram illustrating an example neuromorphic processor, according to at least one embodiment;

FIG. 33 illustrates at least portions of a graphics processor, according to one or more embodiments;

FIG. 34 illustrates at least portions of a graphics processor, according to one or more embodiments;

FIG. 35 illustrates at least portions of a graphics processor, according to one or more embodiments;

FIG. 36 is a block diagram of a graphics processing engine 3610 of a graphics processor in accordance with at least one embodiment.

FIG. 37 is a block diagram of at least portions of a graphics processor core, according to at least one embodiment;

FIGS. 38A and 38B illustrate thread execution logic 3800 including an array of processing elements of a graphics processor core according to at least one embodiment;

FIG. 39 illustrates a parallel processing unit (“PPU”), according to at least one embodiment;

FIG. 40 illustrates a general processing cluster (“GPC”), according to at least one embodiment;

FIG. 41 illustrates a memory partition unit of a parallel processing unit (“PPU”), according to at least one embodiment; and

FIG. 42 illustrates a streaming multi-processor, according to at least one embodiment.

DETAILED DESCRIPTION

The present document describes a system and method for estimating the pose of an object while the object is being manipulated by a robotic hand, claw, or manipulator. When an object is being held by a robot, image-based pose estimation systems may, in various examples, suffer from inaccurate pose estimation caused by object occlusion. In at least one embodiment, a robotic hand is equipped with tactile sensors, and while an object is manipulated by the robotic hand, sensor signals generated by the tactile sensors are used to improve the estimate of the pose of the object. In some situations, dynamic effects such as slipping may occur during active manipulation that increase the difficulty of pose estimation. In at least one embodiment, a physics model of the object is used to improve the modeling of robot-object interactions.

In at least one embodiment, a graphics processing units (“GPU”)-Accelerated physics engine with derivative-free, sample-based optimizers tracks in-hand object poses with contact feedback during manipulation. In at least one embodiment, the physics simulation is used as the forward model for robot-object interactions, and the techniques described herein jointly optimize for the state and the parameters of the simulations, such that the simulations more accurately estimate the real world.

In at least one embodiment, techniques described herein explicitly model the dynamics of robot-object interactions for object pose tracking and optimize for simulation parameters during pose tracking. In various examples, these features allow the system to track the object pose under complex dynamic behaviors, such as translational and torsional slippage due to inertial and external forces and breaking and re-establishing contact. In addition, by using GPU-Accelerated physics engine, these techniques can often be applied in real-time (30 Hz) by using a GPU.

Various embodiments demonstrate promising applications for GPU-Accelerated physics simulation for robotics. For example, in some embodiments, the speed of the physics engine allows expensive, contact-rich simulations and sample-based optimization methods, which rely on data from many concurrent simulations, all in real-time on the same machine, which is often more difficult with CPU-based simulations. Various embodiments may be used as a tool for estimating in-hand object pose and for relaxing the constraint that many dexterous manipulation researchers face, which is that the object is usually placed in such a way that it is mostly visible from a camera, limiting the range of manipulation tasks that can be studied.

Teleoperation may imbue lifeless robotic systems with sophisticated reasoning skills, intuition, and creativity. However, teleoperation solutions for high degree-of-actuation (“DoA”), multi-fingered robots may be complex. In at least one embodiment, a system is developed that allows for complete control over a high DoA robotic system by merely observing the bare human hand. The system may enable operators to solve a variety of complex manipulation tasks that go beyond simple pick-and-place operations. In various embodiments, the system may be implemented by one or more systems as described/depicted in FIGS. 13-40.

Tracking the pose of an object while it is being held and manipulated by a robot hand may be difficult for vision-based methods due to significant occlusions. The techniques described herein utilize GPU-accelerated parallel robot simulations and derivative-free, sample-based optimizers to track in-hand object poses with contact feedback during manipulation. In some examples, a physics simulation is used as the forward model for robot-object interactions, and the algorithm jointly optimizes for the state and the parameters of the simulations, so they better match with those of the real world. At least one embodiment runs in real-time (30 Hz) on a single GPU, and it achieves an average point cloud distance error of 6 mm in simulation experiments and 13 mm in the real-world ones.

In at least one embodiment, performing dexterous manipulation policies benefits from a robust estimate of the pose of the object held in-hand. However, in many implementations, in-hand object pose tracking still presents a challenge due to significant occlusions. In such implementations, works that require in-hand object poses may be limited to experiments where the object is mostly visible or rely on multiple cameras, or the hand-object transform is fixed or known. In some examples, the issue of visual occlusions is mitigated by studying object pose estimation via contacts or tactile feedback, often by using particle filters and knowledge of the object geometry and contact locations. In at least one embodiment, these techniques may be applied to a static-grasp setting, where an object is stationary and in-grasp. In at least one embodiment, these techniques are extended to tracking object poses during in-hand manipulation, requiring modeling of complex object-hand contact dynamics.

To provide in-hand object tracking during robot manipulation, at least one embodiment combines a GPU-accelerated, high-fidelity physics simulator as the forward dynamics model with a sample-based optimization framework to track object poses with contact feedback as shown in FIG. 7. In at least one embodiment, a concurrent set of simulations is initialized with the initial states of a real robot and the initial pose of the real object, which may be obtained from a vision-based pose registration algorithm assuming the object is not in occlusion in the beginning. In at least one embodiment, the initial poses of the simulated objects are slightly perturbed and reflect the uncertainty of the vision-based pose registration algorithm. In at least one embodiment, the GPU-accelerated physics simulator runs many concurrent simulations in real-time on a single GPU. In at least one embodiment, as a given policy controls the real robot to approach, grasp, and manipulate the object in-hand, the system runs the same robot control commands on the simulated robots. In at least one embodiment, observations of the real robot and the simulated robots are collected, which include terms like the magnitude and direction of contacts on the robot hand's contact sensors. In at least one embodiment, a sample-based optimization algorithm periodically updates the states and parameters of the simulations according to a cost function that captures how well the observations of each simulation matches with those of the real world. In addition, in some embodiments, the algorithm updates simulation parameters, such as mass and friction, to further improve the simulations' dynamics models of the real world. In at least one embodiment, at any point in time, the object pose estimate is the pose of the robot-object system.

In at least one embodiment, to evaluate the proposed algorithm, a total of 24 in-hand manipulation trajectories with three different objects in simulation and in the real world were collected. In at least one embodiment, a Kuka IIWA7 arm with the 4-finger Wonik Robotics Allegro hand as the end-effector was used, with each finger outfitted with a SynTouch BioTac contact sensor. In at least one embodiment, object manipulation trajectories are human demonstrations collected via a hand-tracking teleoperation system. In at least one embodiment, because ground-truth object poses in simulation are available, detailed ablation studies in simulation experiments to study the properties of the proposed algorithm are performed. In at least one embodiment, for real-world experiments, a vision-based algorithm is used to obtain the object pose in the first and last frame of the collected trajectories, where the object is not in occlusion. In at least one embodiment, the pose in the first frame is used to initialize the simulations, and the pose in the last frame is used to evaluate the accuracy of the proposed contact-based algorithm.

Various examples identify in-hand object-pose with vision only, usually by first segmenting out the robot or human hand in an image before performing pose estimation. However, vision-only approaches may degrade in performance for larger occlusions. Some embodiments use tactile feedback to aid object pose estimation. Tactile perception can identify object properties such as materials and pose, as well as provide feedback during object manipulation.

In at least one embodiment, experiments with dynamics models and particle filter techniques reveal that adding noise to applied forces instead of the underlying dynamics yield more accurate tracking results. At least one embodiment combines tactile feedback with a vision-based object tracker to track object trajectories during planar pushing tasks, and another applies incremental smoothing and mapping (“iSAM”) to combine global visual pose estimations with local contact pose readings.

In at least one embodiment, a robot hand grasps an object and localizes the object pose without moving. Some examples use point contact locations and some examples use a full tactile map to extract local geometry information around the contacts.

At least one embodiment uses contact location feedback for pose estimation, and some implementations use a variation of Bayesian or particle filtering. Some embodiments perform filtering jointly over visual features, hand joint positions, force-torque readings, and binary contact modes. Some techniques can be applied to pose estimation when the object is not held by the robot hand as well by using force probes.

In at least one embodiment, tactile maps are used for pose estimation, some examples use large, low-resolution tactile arrays to sense contacts in a grid, while other examples use high-resolution tactile sensors mounted on robot finger tips. In at least one embodiment, the system searches for similar local patches on an object surface to localize the object with respect to the contact location, and other systems fuse GelSight data with a point cloud perceived by a depth sensor before performing pose estimation.

Some embodiments implement in-hand object pose tracking during object manipulation, which is more challenging than if the object is static. In at least one embodiment, an algorithm that combines contact locations with Dense Articulated Real-time Tracking (“DART”) is used. In at least one embodiment, the algorithm fuses contact locations with color visual features, joint positions, and force-torque readings. In at least one embodiment the algorithm is sensitive to initialization of the object poses, especially when the object appears small in the depth image. In at least one embodiment, techniques described herein do not assume access to robust visual features during manipulation, but instead utilize a physics simulator to model both the kinematics and the dynamics of the robot-object system.

In various examples, robotic teleoperation may have applications in search and rescue, space, medicine, and applied machine learning. The motivation for teleoperative capability may be to allow a robot system to solve complex tasks by harnessing the cognition, creativity, and reactivity of humans through a human-machine interface (“HM”). In an embodiment, this system provides a glove-free solution to drive a multi-fingered, highly actuated robot system to solve a wide variety of grasping and manipulation tasks. In some examples, depth cameras and various graphics processing units (“GPU”) may be used along with deep learning and optimization to produce a minimal-footprint, dexterous teleoperation system. In some examples, a variety of physical tasks can be performed with visual feedback alone. Therefore, this system may utilize the human ability to plan, move, and predict the consequences of physical actions from vision alone, which may be a sufficient condition for solving a variety of tasks.

The developed system, in various embodiments, enables such dexterous robot manipulation using multi-camera depth observations of the bare human hand. In some examples, the system may be a glove-free and entirely vision-based teleoperation system that dexterously articulates a highly-actuated robotic hand-arm system through direct imitation. The system may also demonstrate a range of tasks particularly involving fine manipulation and dexterity (e.g., extracting paper money from a wallet and concurrently picking two cubes with four fingers as depicted in FIG. 1).

FIG. 1 illustrates an example of teleoperation across various tasks, according to at least one embodiment. In one example, a robotic gripper 104 grasps a cylinder using a grasp pose based on a human hand 102. In another example, a robotic gripper 108 grasps a cube using a grasp pose based on a human hand 106. In another example, a robotic gripper 112 grasps a cup using a grasp pose based on a human hand 110. In another example, a robotic gripper 116 grasps a wallet using a grasp pose based on a human hand 104.

The teleoperation setup may comprise a robot system and an adjacent human pilot arena as shown in FIG. 2. FIG. 2 illustrates an example of a robot with tactile sensors, according to at least one embodiment. In at least one embodiment, a robot 202 has a robotic gripper 204 that is used to grasp objects. In at least one embodiment a set of cameras 206, 208, 210, and 212 are used to observe the workspace of the robot 202. In at least one embodiment, the gripper 204 includes a set of tactile sensors 216, 218, 220, and 222 that provide sensory information to a control computer system. In at least one embodiment, the tactile sensors may be covered with a friction material to enhance and/or improve the robot's ability to grip an object.

In some embodiments, as depicted in FIG. 2, the robot system may be a KUKA LBR iiwa7 R800 series Arm with a Wonik Robotics Allegro hand retrofitted with four SynTouch BioTac tactile sensors at the fingertips and 3M TB641 grip tape applied to the inner surfaces of the phalanges and palm, in which the rubbery surfaces of both the BioTac sensors and 3M tape may improve friction of the hand while the BioTacs themselves may produce 23 signals that can later be used to learn sensorimotor control from demonstrations. The human arena may be a black-clothed table surrounded by four calibrated and time-synchronized cameras, such as Intel RealSense RGB-D cameras, which may be spatially arranged to cover a workspace of 80 cm×55 cm×38 cm. In some examples, the cameras may be directly adjacent to the robot to improve line-of-sight and visual proximity since teleoperation is entirely based on human vision and spatial reasoning. It should be noted that FIG. 2 is intended to be an illustrative example and, in various embodiments, the system may include any robot system utilizing any robot components (e.g., various types of robot arms, hands, tactile sensors, grip, other sensors, cameras, and/or variations thereof) in any suitable environment.

To produce a natural-feeling teleoperation system, an imitation-type paradigm may be adopted. The bare human hand motion pose and finger configuration may be constantly observed and measured by a visual perception module. The human hand motion may then be relayed to the robot system in such a way that the copied motion is self-evident. This approach may enable a human pilot to curl and arrange their fingers, form grasps, reorient and translate their palms, with the robot system following in a similar manner. In at least one embodiment, the system relies heavily on Dense Articulated Real-Time Tracking (“DART”), which may form backbone of tracking the pose and joint angles of the human hand. The full system architecture and component connections are depicted in FIG. 3, in an embodiment.

FIG. 3 illustrates an example of a system that tracks objects in real-time, according to at least one embodiment. In at least one embodiment, the system operates using three threads, which are independent processes running on one or more processors of a computer system. In at least one embodiment, one or more images of a hand are obtained from RGB-Depth (“RGB-D”) cameras 302. The images are processed by a pointnet:stage 1 304, a pointnet:stage 2 306, and a jointnet 308 to produce a hand pose for the hand in the images. In at least one embodiment, an articulated hand model 310 and the hand pose are processed using DART 312 and kinematic retargeting 314 to produce a corresponding hand pose for a robotic gripper. In at least one embodiment, a control thread applies Reimannian motion policies 318 to the gripper hand pose, and the resulting information is used to control the robot 320.

In at least one embodiment, DART is used for continuous pose and joint angle tracking of a human hand. In at least one embodiment, DART uses an articulated model of the hand that is registered against an input point cloud. A human hand model may be obtained and turned into a single mesh model. Utilizing computer-aided design (“CAD”) software, the fingers of the mesh model may be separated into their respective proximal, medial, and distal links, and re-exported as separate meshes along with an associated extensible markup language (“XML”) file that describes their kinematic arrangement. In total, the human hand model may possess 20 revolute joints: four joints per finger with one abduction joint and three flexion joints.

In at least one embodiment, DART is a model-based tracker that relies on non-linear optimization and initialization (e.g., from the previous frame or an initial guess). In some examples, if this initialization is not within the basin of convergence, the tracker can fail to converge to the correct solution. In various embodiments, when tracking the human hand model with point cloud data, the hand model may often snap to spurious local minima leading to tracking failures every few minutes. Therefore, to reliably track the human hand over long periods of time as needed for teleoperation it may be desirable to have reliable hand pose priors, clean hand segmentation, and a multi-view camera studio to prevent the hand model from snapping onto unexpected local minima. In various embodiments, one method for generating hand pose priors is training a neural network on a large dataset of human hand poses given camera images.

In at least one embodiment, data collection is initiated with DART and no hand pose priors, seeding the training of an initial network to produce hand priors. Subsequently, DART and the latest trained neural network may generate increasing amounts of data. In at least one embodiment, the network is perpetually updated with the latest datasets to generate increasingly better priors for DART, which may ultimately extend the range over which DART can operate without any failures. In some examples, the hand pose neural network may be a PointNet-based architecture which operates directly on fused point cloud data obtained by back-projecting depth images from extrinsically calibrated depth cameras into a single global reference frame with annotations provided by DART. In various embodiments, since the fused point cloud contains both the points on table as well as human body and arm, it may be imperative to first localize the hand. Points may be removed from the table by fitting a plane and feeding the remaining points containing the arm and human body to PointNet which may localize the hand as well as provide the hand pose. PointNet may be based on estimating hand pose via a vote-based regression scheme to the 3D positions of specified keypoints on the hand, a technique which may be associated with spatial-softmax often used in 2D keypoint localization. In various embodiments, PointNet may be trained to predict 3D coordinates of 23 keypoints specified on the hand four joint keypoints for each of the five fingers and three keypoints on the back of the hand for hand pose estimation. The loss function may be the Euclidean distance between predicted and the ground truth keypoints. Additionally, an auxiliary segmentation loss may be included to obtain hand segmentation. For efficiency reasons, any input point cloud may be sub-sampled uniformly to a fixed 8192×3 size before being fed to PointNet. In at least one embodiment, while reasonable hand pose estimation and segmentation may be achieved, high quality predictions for the 20 joint keypoints on the fingers may not yet be achieved. In at least one embodiment, the uniform sub-sampling used at the input may indicate that points on the fingers are not densely sampled, and therefore a second stage refinement may be needed which resamples points on the hand from the original raw point cloud given the pose and segmentation of the first stage. In at least one embodiment, the second stage may be trained on the same loss functions, but may only use the points sampled on the hand instead to predict accurately the 23 keypoints. In at least one embodiment, to enable robustness to any inaccuracies in the hand pose from the first stage, random perturbations may be added to the hand pose for second stage. FIG. 4 depicts the second stage refinement within the system, in accordance with at least one embodiment. In at least one embodiment, both stages of PointNet may be trained on 100K point clouds collected over a batch of 30-45 minutes each for 7-8 hours in total by running DART to provide annotations for keypoints, joint angles and segmentation. In at least one embodiment, to provide joint angle priors for fingers, a third neural network may be trained that maps keypoint locations predicted by PointNet to corresponding joint angles. This neural network, which may be referred to as JointNet, may be a two-layer fully connected network that takes input of size 23×3 and predicts 20-dimensional vector of joint angles for fingers.

In at least one embodiment, the neural networks are trained on data collected across multiple human hands, ensuring accurate pose fits for this system and enabling sensible priors for DART. In some embodiments, the hand tracker may work better for hands geometrically close to the DART human hand model.

In at least one embodiment, teleoperation of a robotic hand that is kinematically disparate from the human hand may require a module that can map the observed human hand joints to the robot hand joints, which can be referred to in some embodiments as the Allegro hand joints. FIG. 5 illustrates an example of a human hand pose 502, and a robotic gripper 504 performing a corresponding pose, according to at least one embodiment. There may be many different approaches to kinematic retargeting. For instance, in at least one embodiment, a module may be used to match the positions from the palm to the fingertips and medial joints, and the directionality of proximal phalanges and thumb distal phalange. In at least one embodiment, the optimized mapping may be used to label human depth images such that a deep network can ingest a depth image and output joint angles. In at least one embodiment, motion retargeting is also utilized. For instance, a deep recurrent neural network may be trained unsupervised to retarget motion between skeletons. In at least one embodiment, the system utilizes fingertip task-space metrics because distal regions may be of the highest priority in grasping and manipulation tasks as measured by their contact prevalence, degree of innervation, and heightened controllability for fine, in-hand manipulation skill. In at least one embodiment, the joint axes and locations between two hands may be different, and therefore, no metrics directly comparing joint angles between the two hands may be used. In at least one embodiment, to capture and optimize for the positioning of fingertips, both distance and direction among fingertips are considered. Specifically, in at least one embodiment, the cost function for kinematic retargeting may be chosen as:

C(q_h,q_a)=½Σ_i=0^Ns(d_i)∥r_i(q_a)−f(d_i)r_i{circumflex over ( )}(q_h)∥²+γ∥q_a∥²

where q_h, q_a may be the angles of the human hand model and Allegro hand, respectively, r_i ∈ R³ may be the vector pointing from the origin of one coordinate system to another, expressed in the origin coordinate system (see FIG. 5). Furthermore, in at least one embodiment,

$d_i=r_i\left(q_h\right)\mathrm{and}{r}_i^{\bigwedge }\left(q_h\right)=\frac{r_i\left(q_h\right)}{r_i\left(q_h\right)}.$

The switching weight function s(d_i) may be defined as:

$s\left(d_i\right)=\{\begin{array}{cc}1,& d_i>ϵ\\ 200,& d_i\le ϵ\bigwedge r_i\left(q_h\right)\in S_1\\ 400,& d_i\le ϵ\bigwedge r_i\left(q_h\right)\in S_2\end{array}$

where S₁ may be vectors that originate from a primary finger (index, middle, ring) and point to the thumb, and S₂ may be vectors between two primary fingers when both primary fingers have associated vectors ∈ S₁ (e.g., both primary fingers are being projected with the thumb). In at least one embodiment, the distancing function, f(d_i)∈ R is defined as:

$f\left(d_i\right)=\{\begin{array}{cc}\beta d_i,& d_i>ϵ\\ {\eta }_1,& d_i\le ϵ\bigwedge r_i\left(q_h\right)\in S_1\\ {\eta }_2,& d_i\le ϵ\bigwedge r_i\left(q_h\right)\in S_2\end{array}$

where β=1.6 may be a scaling factor, η₁=1×10⁻⁴ m may be a distance between a primary finger and the thumb, and η₂=3×10⁻² m may be a minimum separation distance between two primary fingers when both primary fingers are being projected with the thumb. In at least one embodiment, these projections ensure that contacts between primary fingers and the thumb are close without inducing primary finger collisions in a precision grasp. In at least one embodiment, this may be particularly useful in the presence of visual finger tracking inaccuracies. In some examples, the vectors r_i may not only capture distance and direction from one task space to another, but their expression in local coordinates may further contain information on how the coordinate systems, and thereby fingertips, are oriented with one another. In at least one embodiment, the coordinate systems of the human hand model may therefore have equivalent coordinate systems on the Allegro model with similarity in orientation and placement. The vectors shown in FIG. 5 may form a minimal set that produces the desired retargeting behavior. In some embodiments, γ=2.5×10⁻³ may be a weight on regularizing the Allegro angles to zero (equivalent to fully opened the hand). In at least one embodiment, this term helps with reducing redundancy in solution and ensure that the hand never enters strange minima that may be difficult to recover from (e.g., the fingers embedding themselves into the palm). In at least one embodiment, various mappings from human hand 602-617 to an Allegro robotic hand 618-633 as produced by the kinematic retargeting are shown in FIG. 6.

In at least one embodiment, the above cost function is minimized in real-time using the Sequential Least-Squares Quadratic Programming (“SLSQP”) algorithm. In at least one embodiment, the routine is initiated with Allegro joint angles set to zero, and every solution thereafter may be initiated with the preceding solution. In at least one embodiment, the forward kinematic calculations between the various coordinate systems of both the human hand model and Allegro hand are found. In at least one embodiment, a first-order low-pass filter is applied to the raw retargeted joint angles in order to remove high-frequency noise present in tracking the human hand and to smooth discrete events, like the projection algorithm inducing step-response changes in retargeted angles.

Riemannian Motion Policies (“RMPs”), in an embodiment, are real-time motion generation methods that calculate acceleration fields from potential function gradients and corresponding Riemannian metrics. RMPs may combine the generation of multi-priority Cartesian trajectories and collision avoidance behaviors together in one cohesive framework. In at least one embodiment. They are used to control the Cartesian pose of the Allegro palm given the observed human hand pose while avoiding arm-palm collisions with the table or operator using collision planes. Given these objectives, in at least one embodiment, the RMPs generated target arm joint trajectories are sent to the arm's torque-level impedance controller at 200 Hz. In at least one embodiment, the kinematically retargeted Allegro angles are sent to the torque-level joint controller at 30 Hz. In at least one embodiment, a teleoperation instance is initialized by registering the studio cameras with the robot base coordinate system via an initial, static robot pose and the initial observation of the human hand. In at least one embodiment, the human hand model axes and robot end-effector axes is approximately aligned such that direction of movements are preserved between human hand motion and robot motion.

Overall, the system can be reliably used to solve a variety of tasks spanning a range of difficulty. In some examples, the ability to solve these tasks reveals that the system may have the dexterity to exhibit precision and power grasps, multi-fingered prehensile and non-prehensile manipulation, in-hand finger gaiting, and compound in-hand manipulation (e.g., grasping with two fingers while simultaneously manipulating with the remaining fingers).

In at least one embodiment, the system may enable a highly-actuated hand-arm system to find a motor solution to a variety of manipulation tasks by translating observed human hand and finger motion to robot arm and finger motion. In at least one embodiment, several tasks, like extracting paper money from a wallet and opening a cardboard box within a plastic container, may be so complex that hand-engineering a robot solution or applying learning methods directly may be likely intractable. Solving these tasks and others through the embodied robotic may allow for these solutions to be generated on-demand for many demonstrations. Furthermore, creating these solutions on the system itself may allow for the reading, access, and storage of the various tactile signals in the robot's fingertips, various commanded and measured joint position and velocity signals through the hand and arm, various torque commands throughout the system, and any camera feeds associated with the system. In at least one embodiment, this rich source of data together with demonstrations of tasks may be used to solve complex, multi-stage, long horizon tasks.

In an embodiment, a system is developed to track in-hand objects during robot manipulation. In various embodiments, the system may be implemented by one or more systems as described/depicted in FIGS. 13-41. As depicted in FIG. 7, the system may comprise a GPU-accelerated, high-fidelity physics simulator as the forward dynamics model with a sample-based optimization framework to track object poses with contacts feedback. In at least one embodiment, a concurrent set of simulations is initialized with the initial states of a real robot and the initial pose of a real object, which may be obtained from a vision-based pose registration algorithm assuming the object is not in occlusion in the beginning. In at least one embodiment, the initial poses of the simulated objects are slightly perturbed and reflect the uncertainty of the vision-based pose registration algorithm. In at least one embodiment, the GPU-accelerated physics simulator runs many concurrent simulations in real-time on a single GPU. In at least one embodiment, a given policy is utilized that controls the real robot to approach, grasp, and manipulate the object in-hand, and the same robot control commands are run on the simulated robots. In at least one embodiment, observations of the real robot and the simulated robots are collected, which include terms like the magnitude and direction of contacts on the robot hand's contact sensors. In at least one embodiment, a sample-based optimization algorithm is utilized that periodically updates the states and parameters of the simulations according to a cost function that captures how well the observations of each simulation match with those of the real world. In addition, in at least one embodiment, the algorithm also updates simulation parameters, such as mass and friction, to further improve the simulations' dynamics models of the real world. At any point in time, the object pose estimate may be the pose of the robot-object system.

In various embodiments, to evaluate the proposed algorithm, a total of 24 in-hand manipulation trajectories with three different objects in simulation and in the real world may be collected, although any number of trajectories may be collected. At least one embodiment utilizes a robot arm such as the Kuka IIWA7 arm with the 4-finger Wonik Robotics Allegro hand as the end-effector, with each finger outfitted with a SynTouch BioTac contact sensor. In at least one embodiment, object manipulation trajectories are human demonstrations collected via a hand-tracking teleoperation system. In various embodiments, due to the ground-truth object poses in simulation, detailed ablation studies are performed in simulation experiments to evaluate the properties of the proposed algorithm. In at least one embodiment, a vision-based algorithm is utilized to obtain the object pose in the first and last frame of the collected trajectories, where the object is not in occlusion. In at least one embodiment, the pose in the first frame is used to initialize the simulations, and the pose in the last frame is used to evaluate the accuracy of the proposed contact-based algorithm.

FIG. 7 illustrates an embodiment of an in-hand object pose tracking framework. In at least one embodiment, robotic controls 702 are sent to a GPU-accelerated physics simulator that runs many robot simulations in parallel 708, each with different physics parameters and perturbed object poses. In at least one embodiment, costs based on observations, such as contact feedback from the real world and from the simulations, are passed to a sample-based derivative-free optimizer 704 that periodically updates the states and parameters of all simulations to better match that of the real world. In at least one embodiment, at any point in time, the pose of the simulation with the lowest cost is chosen as the current object pose estimate 706.

In an embodiment, a system tracks the pose of an object held in-hand by a robot manipulator during object manipulation. In some embodiments, for time, which may be represented by t, an object pose may be defined as p_t ∈ SE(3), and a physics dynamics model may be defined as s_t+1=f(s_t,u_t, θ), where s_t may be the state of the world (position and velocities of rigid bodies and of joint angles in articulated bodies), u_t ∈ ^M may be the robot controls (desired joint positions may be utilized as the action space), and θ ∈ ^N may be the fixed parameters of the simulation (e.g., mass and friction).

In various embodiments, for a simulation model f that exactly matches reality given perfect initializations of p₀, s₀, and θ, pose estimation may require only playing back the sequence of actions u_t applied to the robot in the simulation. However, as forward models may be imperfect and pose initializations may be noisy, pose estimation can be improved through observation feedback.

In some embodiments, D may be defined as a number of joints the robot has and L may be defined as a number of its contact sensors. An observation vector o_t may be defined as the concatenation of the joint position configuration of the robot q_t ∈^D, the position and rotation of the robot's contact sensors P_t^(l)∈³, R_t^(l)∈ SO(3) (which may be located on the fingertips), the force vectors of the sensed contacts c_t^(l)∈³, the unit vector in the direction of the translational slippage on the contact surface d_t^(l)∈², and the binary direction of the rotational slippage on the contact surface r_t^(l)∈ {0, 1}, where l may index into the lth contact sensor. In at least one embodiment, to determine general in-hand pose estimation, given the current and past observations o_1:t, the robot controls u_1:t, and the initial pose p₀, the most probable current object pose p_t is determined.

In various embodiments, a GPU-accelerated physics simulator may be utilized as a forward dynamics model to concurrently simulate many robot-object environments to track in-hand object pose, and derivative-free, sample-based optimizers may be utilized to jointly tune the state and parameters of the simulations to improve tracking performance. FIG. 9 depicts an example embodiment of an algorithm that may be utilized.

First, an estimate of an initial object pose may be obtained via a vision-based object pose estimator. The pose estimator may be assumed to give a reliable initial pose estimate p₀ when the robot is not in contact with the object and when the object is not occluded, such as before grasping. Then, given the initial object pose estimate and robot configuration, K concurrent simulations may be initialized, and at every time step the real robot actions u_t may be copied to all K simulations. In various examples, the object pose can change when the hand establishes contact, and this may be modeled by the simulator. In various embodiments, the object pose and the observation of the ith simulation may be defined as p_t⁽ⁱ⁾ and o_t⁽ⁱ⁾, and the ground truth observations may be defined as o_t^(gt). In various embodiments, given a cost function C, the current best pose estimate at time t may be the pose of the i*th simulation, p_t^(i*), where the i*th simulation may be the one that incurs the lowest average cost across some past time window T:

$C_i=\frac{1}{T}\sum _{\Delta t=0}^{T-1}C\left(o_{\left(t-\Delta t\right)}^{\left(i\right)},o_{\left(t-\Delta t\right)}^{\left(\mathrm{gt}\right)}\right)$ $i^{*}=\arg \min C_i$

The costs may be used to periodically update the simulations and their parameters, which may enable improved alignment with the real robot-object system.

In some embodiments, a cost function may be utilized that sufficiently correlates with object pose differences during in-hand object manipulation such that a lower cost corresponds to better pose estimations. The cost function may be represented by the following symbolic mathematical equation:

$C\left(o_{\left(t-\Delta t\right)}^{\left(i\right)},o_{\left(t-\Delta t\right)}^{\left(\mathrm{gt}\right)}\right)=w_1q_t^{\left(i\right)}-q_t^{\left(\mathrm{gt}\right)}2+\sum _{l=1}^L\left(w_2P_t^{\left(i,l\right)}-P_t^{\left(\mathrm{gt},l\right)}2+w_3\Delta \left(R_t^{\left(i,l\right)},R_t^{\left(\mathrm{gt},l\right)}\right)+w_4\left(1-{\alpha }_{\left(i,l\right)}\right)+w_5\left({\alpha }_{\left(i,l\right)}\right)\Delta M\left(c_t^{\left(i,l\right)},c_t^{\left(\mathrm{gt},l\right)}\right)+w_6\left({\alpha }_{\left(i,l\right)}\right)\mathrm{\Delta \phi }\left(c_t^{\left(i,l\right)},c_t^{\left(\mathrm{gt},l\right)}\right)+w_7\left(1-{\beta }_{\left(i,l\right)}\right)+w_8\left({\beta }_{\left(i,l\right)}\right)\mathrm{\Delta \phi }\left(d_t^{\left(i,l\right)},d_t^{\left(\mathrm{gt},l\right)}\right)+w_9\left(1-{\gamma }_{\left(i,l\right)}\right)+w_{10}\left({\gamma }_{\left(i,l\right)}\right)\left(r_t^{\left(i,l\right)}-r_t^{\left(\mathrm{gt},l\right)}\right)\right)$

For the first term in the cost function, comparing q_t's between the simulated and real-world robots may be useful even if they share the same u_t, because q_t may be different depending on the collision constraints imposed by the current pose of the object in contact with the robot hand, which may make it physically impossible for a joint to reach a commanded target angle.

In at least one embodiment, a contact sensor is considered in contact if its force magnitude is greater than a certain threshold. In at least one embodiment, α_(i,l) is 1 when the binary contact state of the lth contact sensor of the ith simulation agrees with that of the real contact sensor and 0 otherwise. In at least one embodiment, β_(i,l) is 1 when the lth contact sensor of the ith simulation agrees with the real contact sensor in whether or not the sensor is undergoing translational slippage, 0 otherwise; γ_(i,l) is the same but for rotational slippage.

In at least one embodiment, for any two vectors, ΔM(•,•) gives the difference of their magnitudes, and Δφ(•,•) gives the angle between them. In at least one embodiment, for any two rotations R_a and R_b, Δ(R_a, R_b) gives the angle of the axis-angle representation of R_a⁻¹ R_b. In at least one embodiment, the weights of the cost terms, w_is, is determined such that the corresponding mean magnitude of each term is roughly normalized to 1.

In various embodiments, there may be two sources of uncertainty regarding object pose estimation via simulation: 1) the initial pose estimation p₀ from the vision-based pose estimator may be noisy, and 2) there may be a mismatch between the simulated and real-world dynamics, partly caused by imperfect modeling and partly caused by the unknown real-world physics parameters θ.

In at least one embodiment, the first issue of initial pose uncertainty is addressed by perturbing the initial pose estimations across the different simulations by sampling from a distribution centered around the vision-based estimated pose p₀⁽ⁱ⁾˜N(p₀,Σ_p) and increasing the number of simulations K. In various embodiments, when K is arbitrarily large, then it may be with high probability that the true initial pose will be sufficiently represented in the set of simulations, and a well-designed cost function may select the correct simulation with the correct pose. In at least one embodiment, the translation and rotation is sampled separately to perform sampling over initial object poses. In at least one embodiment, translation is sampled from an isotropic normal distribution, while rotation may be sampled by drawing zero-mean, isotropic tangent vectors in so(3) and then applying it to the mean rotation.

In at least one embodiment, a second issue of mismatch between simulated and real-world physics (the “sim-to-real” gap) is addressed by utilizing derivative-free, sample-based optimization algorithms to tune θ during pose tracking. In various embodiments, after every T time steps, the average costs of all simulations during this window along with the simulation state and parameters may be passed to a given optimizer. In at least one embodiment, the optimizer determines the next set of simulations with their own updated parameters. In at least one embodiment, the simulations in the next set are sampled from simulations from the current set, with some added perturbations to the simulation parameters and object pose. In at least one embodiment, such exploration may maintain the diversity of the simulations, preventing them from getting stuck in sub-optimal simulation parameters or states due to noisy observations.

Although it may be desirable to have θ^(i*) converge to the true θ_(gt), this may not be necessary to achieve good pose estimation. In addition, due to differences in simulated and real-world dynamics, the optimal θ for reducing C to be their corresponding real-world values may have modest variation from calculated theoretical predictions.

The parameters of the K simulations may be optimized through three derivative-free, sample-based optimizers:

In an embodiment, a weighted resampling (“WRS”) optimizer is utilized. In various embodiments, WRS may form a probability mass function (“PMF”) over the existing simulation states s^(1:K) and sample K times with replacement from that distribution to form the next set of simulations. To form the PMF, WRS may apply softmax over the simulation costs:

$P\left(i\right)=\frac{\exp -\lambda \left(C_i-{\min }_j{C}_j\right)}{\sum _{i=1}^K\exp -\lambda \left(C_i-{\min }_j{C}_j\right)}$

Here, λ may be a temperature hyperparameter that determines the sharpness of the distribution. In at least one embodiment, after resampling, exploration on simulations are performed by perturbing the simulation parameters θ and the object pose.

In at least one embodiment, simulation parameters are perturbed by sampling from an isotropic normal distribution around the previous parameters: θ_τ+1⁽ⁱ⁾˜N(θ_τ⁽ⁱ⁾, Σ_θ), where Σ_θ may be predefined. The subscript τ may denote the optimizer update step (after T update steps the simulation has ran for a total of τT time steps).

In an embodiment, a relative entropy policy search (“REPS”) optimizer is utilized. At least one embodiment utilizes a sample-based variant of REPS that computes weights for each simulation and samples from a distribution formed by the softmax of those weights. Whereas WRS may use a fixed λ parameter to shape the distribution, REPS may solve for an adaptive temperature parameter η that best improves the performance of the overall distribution subject to ϵ, which may be a constraint on the KL-divergence between the old and updated sample distributions.

To use REPS, at least one embodiment reformulates the costs as rewards by setting R_i=max_jC_j+min_jC_j−C_i. The parameter η may be computed at every step by optimizing the dual function g(η), and then η may be utilized to form the PMF:

${\eta }^{*}=\mathrm{argmin}\eta ϵ+\mathrm{\eta log}\frac{1}{K}\sum _{i=1}^K\exp \frac{R_i}{\eta }$ $P\left(i\right)=\frac{\exp \frac{R_i}{{\eta }^{*}}}{\sum _{j=1}^K\exp \frac{R_i}{{\eta }^{*}}}$

After resampling, a simulation may be perturbed in the same manner as in WRS.

In various embodiments, a population-based optimization (“PBO”) algorithm may be utilized. In at least one embodiment, the PBO algorithm ranks all simulations by their average costs and finds the top K_best simulations with the lowest costs. In at least one embodiment, the algorithm exploits by replacing the remaining K-K_best simulations with copies of the K_best ones, sampled with replacement, and explores by perturbing the K_best simulations in the same way as WRS. In at least one embodiment, PBO effectively uses a shaped cost that depends only on the relative ordering of the simulation costs and not their magnitudes, potentially making the optimizer more robust to noisy costs.

In at least one embodiment, the above described optimizers utilize a distribution-shaping hyperparameter used to balance exploration with exploitation. In at least one embodiment, various embodiments may use combinations of additional hyperparameters such as the following:

• • T, which may represent the time steps an algorithm may wait for every update.
  • K, which may represent the number of concurrent simulations.
  • θ₀, which may represent the initial normal distribution over simulation parameters.
  • Σ_p, which may represent the diagonal covariance matrix for the normal distribution over initial pose perturbation.
  • Σ_θ and Σ_v, which may represent the diagonal covariances of normal distributions of perturbations used for exploration.

A larger K may be generally better than a smaller K, with the caveat that the resulting simulation may be slower and may not be practical in application. Σ_p may be large enough such that the actual initial pose is well represented in the initial pose distribution. However, K may be increased with a larger Σ_p and the convariance of θ₀ to ensure that the density of the samples may be high enough to capture a wider distribution.

In at least one embodiment, there are two additional trade-offs with these hyperparameters. In at least one embodiment, one trade-off is the exploration-exploitation trade-off in the context of optimizing for θ, and the other is the trade-off between optimizing for θ and for p_t^(i*). In at least one embodiment, making Σ_θ or Σ_v wider increases the speed at which the set of simulation parameters “move,” and the optimizer may explore more than it exploits. In at least one embodiment, increasing T improves the optimization for θ as the optimizer may have more samples to evaluate each simulation. However, updating the simulation parameters too slowly may lead to drift in pose estimation if the least-cost simulation is sufficiently different from the real world, potentially leading to divergent behavior in some examples. In some examples, divergent behavior may occur when force perturbation or some simulation parameters lead to an irrecoverable configuration, where the object falls out of the hand or brings the object into a pose such that small force perturbations cannot bring it back to the correct pose. In some examples, this may be acceptable if a few samples become divergent. Their costs may be high, so in some embodiments, they may be discarded and replaced by ones that are not divergent during optimizer updates.

In various embodiments, a physics simulator may be utilized as the forward model instead of a constant model. In addition to tracking the object pose, the proposed algorithm may also identify the context of the forward model by tuning the simulation parameters θ, which may not be affected by the forward or observation models. The optimizers may be based on discrete samples, because in-hand object poses from some distribution may not be easily sampled due to complicated mesh-penetration constraints imposed by contacts.

In various embodiments, the system may be evaluated with both simulation and real-world experiments using an Allegro Hand mounted on a Kuka IIWA7 robot arm, although in various embodiments any hand, robot arm, robot component, and/or variations thereof, may be utilized. In-hand object manipulation trajectories may first be collected with a hand-tracking teleoperation system, and pose estimation errors may be evaluated by running the proposed algorithms offline against the collected trajectories. FIG. 9 depicts an example of an algorithm which may be utilized by the system. These trajectories may start and end with the object not in the hand and not in occlusion. Due to the presence of ground truth object poses in simulation experiments, detailed ablation studies may be performed in simulation to study the effects of different hyperparameters on algorithm performance. In various embodiments, PoseRBPF, which may be recent RGB-D, particle-filter based pose estimation algorithm, may be utilized to obtain initial and final object poses. The initial and final object poses may be treated as ground truth and the final pose may be compared with the one predicted by the system.

In various embodiments, the 4-finger 16-DoF Allegro hand may be mounted on the 7-Dof Kuka IIWA7 robot arm. To obtain contact feedback in the real world, SynTouch BioTac sensors, or a variation thereof, may be attached to each of the fingertips. One or more processes performed in connection with various sensors and sensors' raw electrode readings may be utilized to predict contact force, slip direction, and grasp stability. A trained model may be utilized to estimate force vectors c_t. In various examples, the cost functions may not contain slippage terms. Simulations may be conducted on a computer with utilizing one or more graphics processing units, one or more central processing units, and one or more units of memory.

In various embodiments, 3 objects from the Yale-Columbia-Berkeley (“YCB”) objects dataset (spam can, foam brick, and toy banana) with models, textures, and point clouds may be utilized. The objects may be chosen based on constraints on the size of the robot hand and on keeping the objects light enough such that the robot hand (e.g., Allegro hand) can perform robust precision grasps with its fingertips.

In some examples, for each object, in both simulation and real-world experiments, 2 demonstrations of 2 types of manipulation trajectories may be utilized: 1) pick and place with finger-grasp and in-hand object rotation, and 2) the same but with finger tips breaking and re-establishing contact during the grasp (finger gaiting). This may give a total of 24 trajectories for analysis for both simulation and real-world experiments. In both trajectory types, the object may undergo translational and rotational slippage from both inertial forces and push-contacts with the table. Each trajectory may last about a minute. In various embodiments, the pose estimation algorithm may be run at approximately 30 Hz, which may result in a total of about 2k frames per trajectory.

In at least one embodiment, the input to the system is a point cloud of the hand of the human demonstrator. In at least one embodiment, a neural network, such as a neural network based on the PointNet++ neural network, maps the point cloud to an estimate of the hand's pose relative to the camera as well as the joint angles of the hand. In at least one embodiment, these estimates along with an articulated hand model and the original point cloud is then given to DART, which performs tracking by refining upon the neural network estimates. Finally, to perform kinematic retargeting, an optimization problem is solved that finds the Allegro hand joint angles that result in fingertip poses close to those of the human hand, in an embodiment.

In addition to the above mentioned optimizers (WRS, REPS, PBO), the following two baselines may also be evaluated: Open Loop (“OLP”) and Identity (“EYE”). OLP may track the object pose with 1 simulation. EYE may be initialized with a set of noisy initial poses and may always pick the pose of the lowest-cost simulation, but it may not perform any resampling or optimizer updates. In various embodiments, an Average Distance Deviation (“ADD”) may be utilized as the evaluation metric. ADD may compute the average distance between corresponding points in the object point cloud situated at the ground truth pose and at the predicted pose.

In various embodiments, arm and hand in the simulation may be controlled via a joint-angle PD controller, and the controller's gains may be tuned so that the joint angle step responses are similar to those of the real robot. To speed up simulation, the collision meshes of the robot and objects may be simplified. This may be done through various tools, such as by applying an algorithm such as the TetWild algorithm, which may give a mesh with triangles that are more equilateral than other algorithms such as MeshLab's Quadric Edge Collapse Decimation tool 1. In total, each simulation may generate at most 200 contacts during manipulation, and K=40 simulations at 30 Hz may be run.

The simulation experiments may be performed with varying amounts of initial pose noise. Three levels may be tested: “Low” may have a translation standard deviation of 1 mm and a rotation standard deviation of 0.01 radians. “Med” may be 5 mm and 0.1 radians, and “High” may be 10 mm 1 radian.

FIG. 8 depicts an example of a comparison of the optimizers on tracking in-hand object poses across all the simulation trajectories. ADD 802 may increase as the initial pose error increases, and the mean ADD for the optimizer-based methods may be lower. While EYE may achieve comparable mean ADD with the optimizer methods, the latter ones may have much smaller error variance and max error. The optimizers 804 may focus the distribution of simulations towards better performing ones over time. In the medium noise case, REPS and PBO may achieve the best ADD with a mean of 5.8 mm and 5.9 mm respectively.

In an embodiment, FIG. 10 and FIG. 11 depict examples of results of ablation studies in simulation performed over the hyperparameters governing exploration distance (how much simulations are perturbed), the number of parallel simulations, and whether or not contact and slip detection feedback is used in the cost function.

The algorithms may be evaluated on real-world trajectories similar to those collected in simulation. PoseRBPF may be utilized to register the object pose in the first and last frames of a trajectory. The initial pose estimate may be used to initialize the simulations, while the last one may be used to evaluate the accuracy of the contacts-based pose tracking algorithm. Unlike simulation experiments, the real-world experiments may initialize the object pose by sampling from the distribution over object poses from PoseRBPF, so the initial pose samples correspond to the uncertainties of the vision-based pose estimation algorithm.

FIG. 12 depicts an example of real-world experiment results. The ADDs may be higher than those from simulation experiments. This may be due to both that the real-world dynamics may be more dissimilar with simulations than are simulations with different parameters, and that real-world observations may be noisier than those in simulations. In various embodiments, few or no optimizers may be able to track the toy banana for the real-world data. The object's long moment arm and low friction coefficient may make its slippage behavior difficult to model precisely. This may be a failure mode of the algorithm, where if all of the simulations become divergent (e.g., the banana rotates in the wrong direction, or falls out of hand), then the algorithm may not recover in subsequent optimizer updates. The best ADD achieved with Foam may be 14.1 mm by PBO, and with Spam may be 12.2 mm by REPS.

In at least one embodiment, a sample-based optimization algorithm for tracking in-hand object poses during manipulation via contact feedback and GPU-accelerated robotic simulation is developed. In at least one embodiment, parallel simulations concurrently maintain many principles about the real world and model object pose changes that may be caused by complex contact dynamics. In at least one embodiment, the optimization algorithm tunes simulation parameters during object pose tracking to further improve tracking performance. In various embodiments, real-world contact sensing may be improved by utilizing vision in-the-loop.

FIG. 13 illustrates an example of a process that, as a result of being performed by a computer system, determines the pose of an object being manipulated by a robotic hand equipped with tactile force sensors. In at least one embodiment, a computer system having one or more processors executes executable instructions stored on a computer-readable memory that, as a result of being executed, cause the computer system to perform the operations illustrated in FIG. 13 and described below.

In at least one embodiment, at block 1302, the computer system obtains tactile information from one or more tactile sensors of a robotic hand being used to manipulate an object. In at least one embodiment, the tactile sensors are BioTac sensors equipped with a friction coating. In at least one embodiment, the tactile sensor data is a 2-dimensional array of force data that represents forces distributed over the surface of the tactile sensor.

In at least one embodiment, at block 1304, the computer system generates a set of simulations that simulate the manipulation performed by the robot, and determines a set of simulated tactile sensor information 1306 for each simulation. In at least one embodiment, each simulation uses a different pose for the object. In at least one embodiment, an estimated pose for the object is obtained from one or more images of the object.

In at least one embodiment, at block 1308, the computer system determines a cost value for each simulation based at least in part on differences between the tactile sensor information measured in the real-world and the tactile information determined in the simulation. Cost may be determined as described above, in an embodiment. Based at least in part on the cost determined for each simulation, a simulation that most closely matches the observations in the real world is identified 1310. In at least one embodiment, the pose of the object in the identified simulation is determined to be the pose of the object in the real world 1312.

In at least one embodiment, the pose can be used when performing a task where the object in-hand is manipulated. For example, the pose of a nut in-hand can be determined so that the robot can position the nut onto a threaded bolt. In another example, the pose of an in-hand object can be used when setting the object on a horizontal surface to ensure that the object is upright and stable when released by the robotic hand.

Inference and Training Logic

FIG. 14A illustrates inference and/or training logic 1415 used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided below in conjunction with FIGS. 14A and/or 14B.

In at least one embodiment, inference and/or training logic 1415 may include, without limitation, code and/or data storage 1401 to store forward and/or output weight and/or input/output data, and/or other parameters to configure neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, training logic 1415 may include, or be coupled to code and/or data storage 1401 to store graph code or other software to control timing and/or order, in which weight and/or other parameter information is to be loaded to configure, logic, including integer and/or floating point units (collectively, arithmetic logic units (ALUs)). In at least one embodiment, code, such as graph code, loads weight or other parameter information into processor ALUs based on an architecture of a neural network to which the code corresponds. In at least one embodiment code and/or data storage 1401 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during forward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. In at least one embodiment, any portion of code and/or data storage 1401 may be included with other on-chip or off-chip data storage, including a processor's L1, L2, or L3 cache or system memory.

In at least one embodiment, any portion of code and/or data storage 1401 may be internal or external to one or more processors or other hardware logic devices or circuits. In at least one embodiment, code and/or code and/or data storage 1401 may be cache memory, dynamic randomly addressable memory (“DRAM”), static randomly addressable memory (“SRAM”), non-volatile memory (e.g., Flash memory), or other storage. In at least one embodiment, choice of whether code and/or code and/or data storage 1401 is internal or external to a processor, for example, or comprised of DRAM, SRAM, Flash or some other storage type may depend on available storage on-chip versus off-chip, latency requirements of training and/or inferencing functions being performed, batch size of data used in inferencing and/or training of a neural network, or some combination of these factors.

In at least one embodiment, inference and/or training logic 1415 may include, without limitation, a code and/or data storage 1405 to store backward and/or output weight and/or input/output data corresponding to neurons or layers of a neural network trained and/or used for inferencing in aspects of one or more embodiments. In at least one embodiment, code and/or data storage 1405 stores weight parameters and/or input/output data of each layer of a neural network trained or used in conjunction with one or more embodiments during backward propagation of input/output data and/or weight parameters during training and/or inferencing using aspects of one or more embodiments. In at least one embodiment, training logic 1415 may include, or be coupled to code and/or data storage 1405 to store graph code or other software to control timing and/or order, in which weight and/or other parameter information is to be loaded to configure, logic, including integer and/or floating point units (collectively, arithmetic logic units (ALUs)). In at least one embodiment, code, such as graph code, loads weight or other parameter information into processor ALUs based on an architecture of a neural network to which the code corresponds. In at least one embodiment, any portion of code and/or data storage 1405 may be included with other on-chip or off-chip data storage, including a processor's L1, L2, or L3 cache or system memory. In at least one embodiment, any portion of code and/or data storage 1405 may be internal or external to on one or more processors or other hardware logic devices or circuits. In at least one embodiment, code and/or data storage 1405 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage. In at least one embodiment, choice of whether code and/or data storage 1405 is internal or external to a processor, for example, or comprised of DRAM, SRAM, Flash or some other storage type may depend on available storage on-chip versus off-chip, latency requirements of training and/or inferencing functions being performed, batch size of data used in inferencing and/or training of a neural network, or some combination of these factors.

In at least one embodiment, code and/or data storage 1401 and code and/or data storage 1405 may be separate storage structures. In at least one embodiment, code and/or data storage 1401 and code and/or data storage 1405 may be same storage structure. In at least one embodiment, code and/or data storage 1401 and code and/or data storage 1405 may be partially same storage structure and partially separate storage structures. In at least one embodiment, any portion of code and/or data storage 1401 and code and/or data storage 1405 may be included with other on-chip or off-chip data storage, including a processor's L1, L2, or L3 cache or system memory.

In at least one embodiment, inference and/or training logic 1415 may include, without limitation, one or more arithmetic logic unit(s) (“ALU(s)”) 1410, including integer and/or floating point units, to perform logical and/or mathematical operations based, at least in part on, or indicated by, training and/or inference code (e.g., graph code), a result of which may produce activations (e.g., output values from layers or neurons within a neural network) stored in an activation storage 1420 that are functions of input/output and/or weight parameter data stored in code and/or data storage 1401 and/or code and/or data storage 1405. In at least one embodiment, activations stored in activation storage 1420 are generated according to linear algebraic and or matrix-based mathematics performed by ALU(s) 1410 in response to performing instructions or other code, wherein weight values stored in code and/or data storage 1405 and/or data 1401 are used as operands along with other values, such as bias values, gradient information, momentum values, or other parameters or hyperparameters, any or all of which may be stored in code and/or data storage 1405 or code and/or data storage 1401 or another storage on or off-chip.

In at least one embodiment, ALU(s) 1410 are included within one or more processors or other hardware logic devices or circuits, whereas in another embodiment, ALU(s) 1410 may be external to a processor or other hardware logic device or circuit that uses them (e.g., a co-processor). In at least one embodiment, ALUs 1410 may be included within a processor's execution units or otherwise within a bank of ALUs accessible by a processor's execution units either within same processor or distributed between different processors of different types (e.g., central processing units, graphics processing units, fixed function units, etc.). In at least one embodiment, data storage 1401, code and/or data storage 1405, and activation storage 1420 may be on same processor or other hardware logic device or circuit, whereas in another embodiment, they may be in different processors or other hardware logic devices or circuits, or some combination of same and different processors or other hardware logic devices or circuits. In at least one embodiment, any portion of activation storage 1420 may be included with other on-chip or off-chip data storage, including a processor's L1, L2, or L3 cache or system memory. Furthermore, inferencing and/or training code may be stored with other code accessible to a processor or other hardware logic or circuit and fetched and/or processed using a processor's fetch, decode, scheduling, execution, retirement and/or other logical circuits.

In at least one embodiment, activation storage 1420 may be cache memory, DRAM, SRAM, non-volatile memory (e.g., Flash memory), or other storage. In at least one embodiment, activation storage 1420 may be completely or partially within or external to one or more processors or other logical circuits. In at least one embodiment, choice of whether activation storage 1420 is internal or external to a processor, for example, or comprised of DRAM, SRAM, Flash or some other storage type may depend on available storage on-chip versus off-chip, latency requirements of training and/or inferencing functions being performed, batch size of data used in inferencing and/or training of a neural network, or some combination of these factors. In at least one embodiment, inference and/or training logic 1415 illustrated in FIG. 14A may be used in conjunction with an application-specific integrated circuit (“ASIC”), such as Tensorflow® Processing Unit from Google, an inference processing unit (IPU) from Graphcore™, or a Nervana® (e.g., “Lake Crest”) processor from Intel Corp. In at least one embodiment, inference and/or training logic 1415 illustrated in FIG. 14A may be used in conjunction with central processing unit (“CPU”) hardware, graphics processing unit (“GPU”) hardware or other hardware, such as field programmable gate arrays (“FPGAs”).

FIG. 14B illustrates inference and/or training logic 1415, according to at least one embodiment various. In at least one embodiment, inference and/or training logic 1415 may include, without limitation, hardware logic in which computational resources are dedicated or otherwise exclusively used in conjunction with weight values or other information corresponding to one or more layers of neurons within a neural network. In at least one embodiment, inference and/or training logic 1415 illustrated in FIG. 14B may be used in conjunction with an application-specific integrated circuit (ASIC), such as Tensorflow® Processing Unit from Google, an inference processing unit (IPU) from Graphcore™, or a Nervana® (e.g., “Lake Crest”) processor from Intel Corp. In at least one embodiment, inference and/or training logic 1415 illustrated in FIG. 14B may be used in conjunction with central processing unit (CPU) hardware, graphics processing unit (GPU) hardware or other hardware, such as field programmable gate arrays (FPGAs). In at least one embodiment, inference and/or training logic 1415 includes, without limitation, code and/or data storage 1401 and code and/or data storage 1405, which may be used to store code (e.g., graph code), weight values and/or other information, including bias values, gradient information, momentum values, and/or other parameter or hyperparameter information. In at least one embodiment illustrated in FIG. 14B, each of code and/or data storage 1401 and code and/or data storage 1405 is associated with a dedicated computational resource, such as computational hardware 1402 and computational hardware 1406, respectively. In at least one embodiment, each of computational hardware 1402 and computational hardware 1406 comprises one or more ALUs that perform mathematical functions, such as linear algebraic functions, only on information stored in code and/or data storage 1401 and code and/or data storage 1405, respectively, result of which is stored in activation storage 1420.

In at least one embodiment, each of code and/or data storage 1401 and 1405 and corresponding computational hardware 1402 and 1406, respectively, correspond to different layers of a neural network, such that resulting activation from one “storage/computational pair 1401/1402” of code and/or data storage 1401 and computational hardware 1402 is provided as an input to next “storage/computational pair 1405/1406” of code and/or data storage 1405 and computational hardware 1406, in order to mirror conceptual organization of a neural network. In at least one embodiment, each of storage/computational pairs 1401/1402 and 1405/1406 may correspond to more than one neural network layer. In at least one embodiment, additional storage/computation pairs (not shown) subsequent to or in parallel with storage computation pairs 1401/1402 and 1405/1406 may be included in inference and/or training logic 1415.

Neural Network Training and Deployment

FIG. 15 illustrates training and deployment of a deep neural network, according to at least one embodiment. In at least one embodiment, untrained neural network 91506 is trained using a training dataset 1502. In at least one embodiment, training framework 1504 is a PyTorch framework, whereas in other embodiments, training framework 1504 is a Tensorflow, Boost, Caffe, Microsoft Cognitive Toolkit/CNTK, MXNet, Chainer, Keras, Deeplearning4j, or other training framework. In at least one embodiment training framework 1504 trains an untrained neural network 1506 and enables it to be trained using processing resources described herein to generate a trained neural network 1508. In at least one embodiment, weights may be chosen randomly or by pre-training using a deep belief network. In at least one embodiment, training may be performed in either a supervised, partially supervised, or unsupervised manner.

In at least one embodiment, untrained neural network 1506 is trained using supervised learning, wherein training dataset 1502 includes an input paired with a desired output for an input, or where training dataset 1502 includes input having a known output and an output of neural network 1506 is manually graded. In at least one embodiment, untrained neural network 1506 is trained in a supervised manner processes inputs from training dataset 1502 and compares resulting outputs against a set of expected or desired outputs. In at least one embodiment, errors are then propagated back through untrained neural network 1506. In at least one embodiment, training framework 1504 adjusts weights that control untrained neural network 1506. In at least one embodiment, training framework 1504 includes tools to monitor how well untrained neural network 1506 is converging towards a model, such as trained neural network 1508, suitable to generating correct answers, such as in result 1514, based on known input data, such as new data 1512. In at least one embodiment, training framework 1504 trains untrained neural network 1506 repeatedly while adjust weights to refine an output of untrained neural network 1506 using a loss function and adjustment algorithm, such as stochastic gradient descent. In at least one embodiment, training framework 1504 trains untrained neural network 1506 until untrained neural network 1506 achieves a desired accuracy. In at least one embodiment, trained neural network 1508 can then be deployed to implement any number of machine learning operations.

In at least one embodiment, untrained neural network 1506 is trained using unsupervised learning, wherein untrained neural network 1506 attempts to train itself using unlabeled data. In at least one embodiment, unsupervised learning training dataset 1502 will include input data without any associated output data or “ground truth” data. In at least one embodiment, untrained neural network 1506 can learn groupings within training dataset 1502 and can determine how individual inputs are related to untrained dataset 1502. In at least one embodiment, unsupervised training can be used to generate a self-organizing map, which is a type of trained neural network 1508 capable of performing operations useful in reducing dimensionality of new data 1512. In at least one embodiment, unsupervised training can also be used to perform anomaly detection, which allows identification of data points in a new dataset 1512 that deviate from normal patterns of new dataset 1512.

In at least one embodiment, semi-supervised learning may be used, which is a technique in which in training dataset 1502 includes a mix of labeled and unlabeled data. In at least one embodiment, training framework 1504 may be used to perform incremental learning, such as through transferred learning techniques. In at least one embodiment, incremental learning enables trained neural network 1508 to adapt to new data 1512 without forgetting knowledge instilled within network during initial training.

Data Center

FIG. 16 illustrates an example data center 1600, in which at least one embodiment may be used. In at least one embodiment, data center 1600 includes a data center infrastructure layer 1610, a framework layer 1620, a software layer 1630 and an application layer 1640.

In at least one embodiment, as shown in FIG. 16, data center infrastructure layer 1610 may include a resource orchestrator 1612, grouped computing resources 1614, and node computing resources (“node C.R.s”) 1616(1)-1616(N), where “N” represents any whole, positive integer. In at least one embodiment, node C.R.s 1616(1)-1616(N) may include, but are not limited to, any number of central processing units (“CPUs”) or other processors (including accelerators, field programmable gate arrays (FPGAs), graphics processors, etc.), memory devices (e.g., dynamic read-only memory), storage devices (e.g., solid state or disk drives), network input/output (“NW I/O”) devices, network switches, virtual machines (“VMs”), power modules, and cooling modules, etc. In at least one embodiment, one or more node C.R.s from among node C.R.s 1616(1)-1616(N) may be a server having one or more of above-mentioned computing resources.

In at least one embodiment, grouped computing resources 1614 may include separate groupings of node C.R.s housed within one or more racks (not shown), or many racks housed in data centers at various geographical locations (also not shown). Separate groupings of node C.R.s within grouped computing resources 1614 may include grouped compute, network, memory or storage resources that may be configured or allocated to support one or more workloads. In at least one embodiment, several node C.R.s including CPUs or processors may grouped within one or more racks to provide compute resources to support one or more workloads. In at least one embodiment, one or more racks may also include any number of power modules, cooling modules, and network switches, in any combination.

In at least one embodiment, resource orchestrator 1612 may configure or otherwise control one or more node C.R.s 1616(1)-1616(N) and/or grouped computing resources 1614. In at least one embodiment, resource orchestrator 1612 may include a software design infrastructure (“SDI”) management entity for data center 1600. In at least one embodiment, resource orchestrator may include hardware, software or some combination thereof.

In at least one embodiment, as shown in FIG. 16, framework layer 1620 includes a job scheduler 1632, a configuration manager 1634, a resource manager 1636 and a distributed file system 1638. In at least one embodiment, framework layer 1620 may include a framework to support software 1632 of software layer 1630 and/or one or more application(s) 1642 of application layer 1640. In at least one embodiment, software 1632 or application(s) 1642 may respectively include web-based service software or applications, such as those provided by Amazon Web Services, Google Cloud and Microsoft Azure. In at least one embodiment, framework layer 1620 may be, but is not limited to, a type of free and open-source software web application framework such as Apache Spark™ (hereinafter “Spark”) that may utilize distributed file system 1638 for large-scale data processing (e.g., “big data”). In at least one embodiment, job scheduler 1632 may include a Spark driver to facilitate scheduling of workloads supported by various layers of data center 1600. In at least one embodiment, configuration manager 1634 may be capable of configuring different layers such as software layer 1630 and framework layer 1620 including Spark and distributed file system 1638 for supporting large-scale data processing. In at least one embodiment, resource manager 1636 may be capable of managing clustered or grouped computing resources mapped to or allocated for support of distributed file system 1638 and job scheduler 1632. In at least one embodiment, clustered or grouped computing resources may include grouped computing resource 1614 at data center infrastructure layer 1610. In at least one embodiment, resource manager 1636 may coordinate with resource orchestrator 1612 to manage these mapped or allocated computing resources.

In at least one embodiment, software 1632 included in software layer 1630 may include software used by at least portions of node C.R.s 1616(1)-1616(N), grouped computing resources 1614, and/or distributed file system 1638 of framework layer 1620. One or more types of software may include, but are not limited to, Internet web page search software, e-mail virus scan software, database software, and streaming video content software.

In at least one embodiment, application(s) 1642 included in application layer 1640 may include one or more types of applications used by at least portions of node C.R.s 1616(1)-1616(N), grouped computing resources 1614, and/or distributed file system 1638 of framework layer 1620. One or more types of applications may include, but are not limited to, any number of a genomics application, a cognitive compute, and a machine learning application, including training or inferencing software, machine learning framework software (e.g., PyTorch, TensorFlow, Caffe, etc.) or other machine learning applications used in conjunction with one or more embodiments.

In at least one embodiment, any of configuration manager 1634, resource manager 1636, and resource orchestrator 1612 may implement any number and type of self-modifying actions based on any amount and type of data acquired in any technically feasible fashion. In at least one embodiment, self-modifying actions may relieve a data center operator of data center 1600 from making possibly bad configuration decisions and possibly avoiding underutilized and/or poor performing portions of a data center.

In at least one embodiment, data center 1600 may include tools, services, software or other resources to train one or more machine learning models or predict or infer information using one or more machine learning models according to one or more embodiments described herein. For example, in at least one embodiment, a machine learning model may be trained by calculating weight parameters according to a neural network architecture using software and computing resources described above with respect to data center 1600. In at least one embodiment, trained machine learning models corresponding to one or more neural networks may be used to infer or predict information using resources described above with respect to data center 1600 by using weight parameters calculated through one or more training techniques described herein.

In at least one embodiment, data center may use CPUs, application-specific integrated circuits (ASICs), GPUs, FPGAs, or other hardware to perform training and/or inferencing using above-described resources. Moreover, one or more software and/or hardware resources described above may be configured as a service to allow users to train or performing inferencing of information, such as image recognition, speech recognition, or other artificial intelligence services.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 16 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. For example, a generation network or an evaluation network may be constructed as described in FIGS. 14-16.

Autonomous Vehicle

FIG. 17A illustrates an example of an autonomous vehicle 1700, according to at least one embodiment. In at least one embodiment, autonomous vehicle 1700 (alternatively referred to herein as “vehicle 1700”) may be, without limitation, a passenger vehicle, such as a car, a truck, a bus, and/or another type of vehicle that accommodates one or more passengers. In at least one embodiment, vehicle 1700 may be a semi-tractor-trailer truck used for hauling cargo. In at least one embodiment, vehicle 1700 may be an airplane, robotic vehicle, or other kind of vehicle.

Autonomous vehicles may be described in terms of automation levels, defined by National Highway Traffic Safety Administration (“NHTSA”), a division of US Department of Transportation, and Society of Automotive Engineers (“SAE”) “Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles” (e.g., Standard No. J3016-201806, published on Jun. 15, 2018, Standard No. J3016-201609, published on Sep. 30, 2016, and previous and future versions of this standard). In one or more embodiments, vehicle 1700 may be capable of functionality in accordance with one or more of level 1-level 5 of autonomous driving levels. For example, in at least one embodiment, vehicle 1700 may be capable of conditional automation (Level 3), high automation (Level 4), and/or full automation (Level 5), depending on embodiment.

In at least one embodiment, vehicle 1700 may include, without limitation, components such as a chassis, a vehicle body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other components of a vehicle. In at least one embodiment, vehicle 1700 may include, without limitation, a propulsion system 1750, such as an internal combustion engine, hybrid electric power plant, an all-electric engine, and/or another propulsion system type. In at least one embodiment, propulsion system 1750 may be connected to a drive train of vehicle 1700, which may include, without limitation, a transmission, to enable propulsion of vehicle 1700. In at least one embodiment, propulsion system 1750 may be controlled in response to receiving signals from a throttle/accelerator(s) 1752.

In at least one embodiment, a steering system 1754, which may include, without limitation, a steering wheel, is used to steer a vehicle 1700 (e.g., along a desired path or route) when a propulsion system 1750 is operating (e.g., when vehicle is in motion). In at least one embodiment, a steering system 1754 may receive signals from steering actuator(s) 1756. Steering wheel may be optional for full automation (Level 5) functionality. In at least one embodiment, a brake sensor system 1746 may be used to operate vehicle brakes in response to receiving signals from brake actuator(s) 1748 and/or brake sensors.

In at least one embodiment, controller(s) 1736, which may include, without limitation, one or more system on chips (“SoCs”) (not shown in FIG. 17A) and/or graphics processing unit(s) (“GPU(s)”), provide signals (e.g., representative of commands) to one or more components and/or systems of vehicle 1700. For instance, in at least one embodiment, controller(s) 1736 may send signals to operate vehicle brakes via brake actuators 1748, to operate steering system 1754 via steering actuator(s) 1756, to operate propulsion system 1750 via throttle/accelerator(s) 1752. Controller(s) 1736 may include one or more onboard (e.g., integrated) computing devices (e.g., supercomputers) that process sensor signals, and output operation commands (e.g., signals representing commands) to enable autonomous driving and/or to assist a human driver in driving vehicle 1700. In at least one embodiment, controller(s) 1736 may include a first controller 1736 for autonomous driving functions, a second controller 1736 for functional safety functions, a third controller 1736 for artificial intelligence functionality (e.g., computer vision), a fourth controller 1736 for infotainment functionality, a fifth controller 1736 for redundancy in emergency conditions, and/or other controllers. In at least one embodiment, a single controller 1736 may handle two or more of above functionalities, two or more controllers 1736 may handle a single functionality, and/or any combination thereof.

In at least one embodiment, controller(s) 1736 provide signals for controlling one or more components and/or systems of vehicle 1700 in response to sensor data received from one or more sensors (e.g., sensor inputs). In at least one embodiment, sensor data may be received from, for example and without limitation, global navigation satellite systems (“GNSS”) sensor(s) 1758 (e.g., Global Positioning System sensor(s)), RADAR sensor(s) 1760, ultrasonic sensor(s) 1762, LIDAR sensor(s) 1764, inertial measurement unit (“IMU”)sensor(s) 1766 (e.g., accelerometer(s), gyroscope(s), magnetic compass(es), magnetometer(s), etc.), microphone(s) 1796, stereo camera(s) 1768, wide-view camera(s) 1770 (e.g., fisheye cameras), infrared camera(s) 1772, surround camera(s) 1774 (e.g., 360 degree cameras), long-range cameras (not shown in FIG. 17A), mid-range camera(s) (not shown in FIG. 17A), speed sensor(s) 1744 (e.g., for measuring speed of vehicle 1700), vibration sensor(s) 1742, steering sensor(s) 1740, brake sensor(s) (e.g., as part of brake sensor system 1746), and/or other sensor types.

In at least one embodiment, one or more of controller(s) 1736 may receive inputs (e.g., represented by input data) from an instrument cluster 1732 of vehicle 1700 and provide outputs (e.g., represented by output data, display data, etc.) via a human-machine interface (“HMI”) display 1734, an audible annunciator, a loudspeaker, and/or via other components of vehicle 1700. In at least one embodiment, outputs may include information such as vehicle velocity, speed, time, map data (e.g., a High Definition map (not shown in FIG. 17A), location data (e.g., vehicle's 1700 location, such as on a map), direction, location of other vehicles (e.g., an occupancy grid), information about objects and status of objects as perceived by controller(s) 1736, etc. For example, in at least one embodiment, HMI display 1734 may display information about presence of one or more objects (e.g., a street sign, caution sign, traffic light changing, etc.), and/or information about driving maneuvers vehicle has made, is making, or will make (e.g., changing lanes now, taking exit 34B in two miles, etc.).

In at least one embodiment, vehicle 1700 further includes a network interface 1724 which may use wireless antenna(s) 1726 and/or modem(s) to communicate over one or more networks. For example, in at least one embodiment, network interface 1724 may be capable of communication over Long-Term Evolution (“LTE”), Wideband Code Division Multiple Access (“WCDMA”), Universal Mobile Telecommunications System (“UMTS”), Global System for Mobile communication (“GSM”), IMT-CDMA Multi-Carrier (“CDMA2000”), etc. In at least one embodiment, wireless antenna(s) 1726 may also enable communication between objects in environment (e.g., vehicles, mobile devices, etc.), using local area network(s), such as Bluetooth, Bluetooth Low Energy (“LE”), Z-Wave, ZigBee, etc., and/or low power wide-area network(s) (“LPWANs”), such as LoRaWAN, SigFox, etc.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 17A for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the robot can be an autonomous vehicle, and the evaluation and generation networks can be implemented using a computer system on the autonomous vehicle.

FIG. 17B illustrates an example of camera locations and fields of view for autonomous vehicle 1700 of FIG. 17A, according to at least one embodiment. In at least one embodiment, cameras and respective fields of view are one example embodiment and are not intended to be limiting. For instance, in at least one embodiment, additional and/or alternative cameras may be included and/or cameras may be located at different locations on vehicle 1700.

In at least one embodiment, camera types for cameras may include, but are not limited to, digital cameras that may be adapted for use with components and/or systems of vehicle 1700. Camera(s) may operate at automotive safety integrity level (“ASIL”) B and/or at another ASIL. In at least one embodiment, camera types may be capable of any image capture rate, such as 60 frames per second (fps), 1220 fps, 240 fps, etc., depending on embodiment. In at least one embodiment, cameras may be capable of using rolling shutters, global shutters, another type of shutter, or a combination thereof. In at least one embodiment, color filter array may include a red clear clear clear (“RCCC”) color filter array, a red clear clear blue (“RCCB”) color filter array, a red blue green clear (“RBGC”) color filter array, a Foveon X3 color filter array, a Bayer sensors (“RGGB”) color filter array, a monochrome sensor color filter array, and/or another type of color filter array. In at least one embodiment, clear pixel cameras, such as cameras with an RCCC, an RCCB, and/or an RBGC color filter array, may be used in an effort to increase light sensitivity.

In at least one embodiment, one or more of camera(s) may be used to perform advanced driver assistance systems (“ADAS”) functions (e.g., as part of a redundant or fail-safe design). For example, in at least one embodiment, a Multi-Function Mono Camera may be installed to provide functions including lane departure warning, traffic sign assist and intelligent headlamp control. In at least one embodiment, one or more of camera(s) (e.g., all of cameras) may record and provide image data (e.g., video) simultaneously.

In at least one embodiment, one or more of cameras may be mounted in amounting assembly, such as a custom designed (three-dimensional (“3D”) printed) assembly, in order to cut out stray light and reflections from within car (e.g., reflections from dashboard reflected in windshield mirrors) which may interfere with camera's image data capture abilities. With reference to wing-mirror mounting assemblies, in at least one embodiment, wing-mirror assemblies may be custom 3D printed so that camera mounting plate matches shape of wing-mirror. In at least one embodiment, camera(s) may be integrated into wing-mirror. For side-view cameras, camera(s) may also be integrated within four pillars at each corner of cabIn at least one embodiment.

In at least one embodiment, cameras with afield of view that include portions of environment in front of vehicle 1700 (e.g., front-facing cameras) may be used for surround view, to help identify forward facing paths and obstacles, as well as aid in, with help of one or more of controllers 1736 and/or control SoCs, providing information critical to generating an occupancy grid and/or determining preferred vehicle paths. In at least one embodiment, front-facing cameras may be used to perform many of same ADAS functions as LIDAR, including, without limitation, emergency braking, pedestrian detection, and collision avoidance. In at least one embodiment, front-facing cameras may also be used for ADAS functions and systems including, without limitation, Lane Departure Warnings (“LDW”), Autonomous Cruise Control (“ACC”), and/or other functions such as traffic sign recognition.

In at least one embodiment, a variety of cameras may be used in a front-facing configuration, including, for example, a monocular camera platform that includes a CMOS (“complementary metal oxide semiconductor”) color imager. In at least one embodiment, wide-view camera 1770 may be used to perceive objects coming into view from periphery (e.g., pedestrians, crossing traffic or bicycles). Although only one wide-view camera 1770 is illustrated in FIG. 17B, in other embodiments, there may be any number (including zero) of wide-view camera(s) 1770 on vehicle 1700. In at least one embodiment, any number of long-range camera(s) 1798 (e.g., a long-view stereo camera pair) may be used for depth-based object detection, especially for objects for which a neural network has not yet been trained. In at least one embodiment, long-range camera(s) 1798 may also be used for object detection and classification, as well as basic object tracking.

In at least one embodiment, any number of stereo camera(s) 1768 may also be included in a front-facing configuration. In at least one embodiment, one or more of stereo camera(s) 1768 may include an integrated control unit comprising a scalable processing unit, which may provide a programmable logic (“FPGA”) and a multi-core micro-processor with an integrated Controller Area Network (“CAN”) or Ethernet interface on a single chip. In at least one embodiment, such a unit may be used to generate a 3D map of environment of vehicle 1700, including a distance estimate for all points in image. In at least one embodiment, one or more of stereo camera(s) 1768 may include, without limitation, compact stereo vision sensor(s) that may include, without limitation, two camera lenses (one each on left and right) and an image processing chip that may measure distance from vehicle 1700 to target object and use generated information (e.g., metadata) to activate autonomous emergency braking and lane departure warning functions. In at least one embodiment, other types of stereo camera(s) 1768 may be used in addition to, or alternatively from, those described herein.

In at least one embodiment, cameras with afield of view that include portions of environment to side of vehicle 1700 (e.g., side-view cameras) may be used for surround view, providing information used to create and update occupancy grid, as well as to generate side impact collision warnings. For example, in at least one embodiment, surround camera(s) 1774 (e.g., four surround cameras 1774 as illustrated in FIG. 17B) could be positioned on vehicle 1700. Surround camera(s) 1774 may include, without limitation, any number and combination of wide-view camera(s) 1770, fisheye camera(s), 360 degree camera(s), and/or like. For instance, in at least one embodiment, four fisheye cameras may be positioned on front, rear, and sides of vehicle 1700. In at least one embodiment, vehicle 1700 may use three surround camera(s) 1774 (e.g., left, right, and rear), and may leverage one or more other camera(s) (e.g., a forward-facing camera) as a fourth surround-view camera.

In at least one embodiment, cameras with afield of view that include portions of environment to rear of vehicle 1700 (e.g., rear-view cameras) may be used for park assistance, surround view, rear collision warnings, and creating and updating occupancy grid. In at least one embodiment, a wide variety of cameras may be used including, but not limited to, cameras that are also suitable as a front-facing camera(s) (e.g., long-range cameras 1798 and/or mid-range camera(s) 1776, stereo camera(s) 1768, infrared camera(s) 1772, etc.), as described herein.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 17B for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the robot can be an autonomous vehicle, and the evaluation and generation networks can be implemented using a computer system on the autonomous vehicle. In at least one embodiment, the depth camera used to capture the point cloud of an object to be grasped is a camera, sonar, radar, or lidar on an autonomous vehicle.

FIG. 17C is a block diagram illustrating an example system architecture for autonomous vehicle 1700 of FIG. 17A, according to at least one embodiment. In at least one embodiment, each of components, features, and systems of vehicle 1700 in FIG. 17C are illustrated as being connected via a bus 1702. In at least one embodiment, bus 1702 may include, without limitation, a CAN data interface (alternatively referred to herein as a “CAN bus”). In at least one embodiment, a CAN may be a network inside vehicle 1700 used to aid in control of various features and functionality of vehicle 1700, such as actuation of brakes, acceleration, braking, steering, windshield wipers, etc. In at least one embodiment, bus 1702 may be configured to have dozens or even hundreds of nodes, each with its own unique identifier (e.g., a CAN ID). In at least one embodiment, bus 1702 may be read to find steering wheel angle, ground speed, engine revolutions per minute (“RPMs”), button positions, and/or other vehicle status indicators. In at least one embodiment, bus 1702 may be a CAN bus that is ASIL B compliant.

In at least one embodiment, in addition to, or alternatively from CAN, FlexRay and/or Ethernet may be used. In at least one embodiment, there may be any number of busses 1702, which may include, without limitation, zero or more CAN busses, zero or more FlexRay busses, zero or more Ethernet busses, and/or zero or more other types of busses using a different protocol. In at least one embodiment, two or more busses 1702 may be used to perform different functions, and/or may be used for redundancy. For example, a first bus 1702 may be used for collision avoidance functionality and a second bus 1702 may be used for actuation control. In at least one embodiment, each bus 1702 may communicate with any of components of vehicle 1700, and two or more busses 1702 may communicate with same components. In at least one embodiment, each of any number of system(s) on chip(s) (“SoC(s)”) 1704, each of controller(s) 1736, and/or each computer within vehicle may have access to same input data (e.g., inputs from sensors of vehicle 1700), and may be connected to a common bus, such CAN bus.

In at least one embodiment, vehicle 1700 may include one or more controller(s) 1736, such as those described herein with respect to FIG. 17A. Controller(s) 1736 may be used for a variety of functions. In at least one embodiment, controller(s) 1736 may be coupled to any of various other components and systems of vehicle 1700, and may be used for control of vehicle 1700, artificial intelligence of vehicle 1700, infotainment for vehicle 1700, and/or like.

In at least one embodiment, vehicle 1700 may include any number of SoCs 1704. Each of SoCs 1704 may include, without limitation, central processing units (“CPU(s)”) 1706, graphics processing units (“GPU(s)”) 1708, processor(s) 1710, cache(s) 1712, accelerator(s) 1714, data store(s) 1716, and/or other components and features not illustrated. In at least one embodiment, SoC(s) 1704 may be used to control vehicle 1700 in a variety of platforms and systems. For example, in at least one embodiment, SoC(s) 1704 may be combined in a system (e.g., system of vehicle 1700) with a High Definition (“HD”) map 1722 which may obtain map refreshes and/or updates via network interface 1724 from one or more servers (not shown in FIG. 17C).

In at least one embodiment, CPU(s) 1706 may include a CPU cluster or CPU complex (alternatively referred to herein as a “CCPLEX”). In at least one embodiment, CPU(s) 1706 may include multiple cores and/or level two (“L2”) caches. For instance, in at least one embodiment, CPU(s) 1706 may include eight cores in a coherent multi-processor configuration. In at least one embodiment, CPU(s) 1706 may include four dual-core clusters where each cluster has a dedicated L2 cache (e.g., a 2 MB L2 cache). In at least one embodiment, CPU(s) 1706 (e.g., CCPLEX) may be configured to support simultaneous cluster operation enabling any combination of clusters of CPU(s) 1706 to be active at any given time.

In at least one embodiment, one or more of CPU(s) 1706 may implement power management capabilities that include, without limitation, one or more of following features: individual hardware blocks may be clock-gated automatically when idle to save dynamic power; each core clock may be gated when core is not actively executing instructions due to execution of Wait for Interrupt (“WFI”)/Wait for Event (“WFE”) instructions; each core may be independently power-gated; each core cluster may be independently clock-gated when all cores are clock-gated or power-gated; and/or each core cluster may be independently power-gated when all cores are power-gated. In at least one embodiment, CPU(s) 1706 may further implement an enhanced algorithm for managing power states, where allowed power states and expected wakeup times are specified, and hardware/microcode determines best power state to enter for core, cluster, and CCPLEX. In at least one embodiment, processing cores may support simplified power state entry sequences in software with work offloaded to microcode.

In at least one embodiment, GPU(s) 1708 may include an integrated GPU (alternatively referred to herein as an “iGPU”). In at least one embodiment, GPU(s) 1708 may be programmable and may be efficient for parallel workloads. In at least one embodiment, GPU(s) 1708, in at least one embodiment, may use an enhanced tensor instruction set. In on embodiment, GPU(s) 1708 may include one or more streaming microprocessors, where each streaming microprocessor may include a level one (“L1”) cache (e.g., an L1 cache with at least 96 KB storage capacity), and two or more of streaming microprocessors may share an L2 cache (e.g., an L2 cache with a 512 KB storage capacity). In at least one embodiment, GPU(s) 1708 may include at least eight streaming microprocessors. In at least one embodiment, GPU(s) 1708 may use compute application programming interface(s) (API(s)). In at least one embodiment, GPU(s) 1708 may use one or more parallel computing platforms and/or programming models (e.g., NVIDIA's CUDA).

In at least one embodiment, one or more of GPU(s) 1708 may be power-optimized for best performance in automotive and embedded use cases. For example, in on embodiment, GPU(s) 1708 could be fabricated on a Fin field-effect transistor (“FinFET”). In at least one embodiment, each streaming microprocessor may incorporate a number of mixed-precision processing cores partitioned into multiple blocks. For example, and without limitation, 64 PF32 cores and 32 PF64 cores could be partitioned into four processing blocks. In at least one embodiment, each processing block could be allocated 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two mixed-precision NVIDIA TENSOR COREs for deep learning matrix arithmetic, a level zero (“L0”) instruction cache, a warp scheduler, a dispatch unit, and/or a 64 KB register file. In at least one embodiment, streaming microprocessors may include independent parallel integer and floating-point data paths to provide for efficient execution of workloads with a mix of computation and addressing calculations. In at least one embodiment, streaming microprocessors may include independent thread scheduling capability to enable finer-grain synchronization and cooperation between parallel threads. In at least one embodiment, streaming microprocessors may include a combined L1 data cache and shared memory unit in order to improve performance while simplifying programming.

In at least one embodiment, one or more of GPU(s) 1708 may include a high bandwidth memory (“HBM”) and/or a 16 GB HBM2 memory subsystem to provide, in some examples, about 900 GB/second peak memory bandwidth. In at least one embodiment, in addition to, or alternatively from, HBM memory, a synchronous graphics random-access memory (“SGRAM”) may be used, such as a graphics double data rate type five synchronous random-access memory (“GDDR5”).

In at least one embodiment, GPU(s) 1708 may include unified memory technology. In at least one embodiment, address translation services (“ATS”) support may be used to allow GPU(s) 1708 to access CPU(s) 1706 page tables directly. In at least one embodiment, embodiment, when GPU(s) 1708 memory management unit (“MMU”) experiences a miss, an address translation request may be transmitted to CPU(s) 1706. In response, CPU(s) 1706 may look in its page tables for virtual-to-physical mapping for address and transmits translation back to GPU(s) 1708, in at least one embodiment. In at least one embodiment, unified memory technology may allow a single unified virtual address space for memory of both CPU(s) 1706 and GPU(s) 1708, thereby simplifying GPU(s) 1708 programming and porting of applications to GPU(s) 1708.

In at least one embodiment, GPU(s) 1708 may include any number of access counters that may keep track of frequency of access of GPU(s) 1708 to memory of other processors. In at least one embodiment, access counter(s) may help ensure that memory pages are moved to physical memory of processor that is accessing pages most frequently, thereby improving efficiency for memory ranges shared between processors.

In at least one embodiment, one or more of SoC(s) 1704 may include any number of cache(s) 1712, including those described herein. For example, in at least one embodiment, cache(s) 1712 could include a level three (“L3”) cache that is available to both CPU(s) 1706 and GPU(s) 1708 (e.g., that is connected both CPU(s) 1706 and GPU(s) 1708). In at least one embodiment, cache(s) 1712 may include a write-back cache that may keep track of states of lines, such as by using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). In at least one embodiment, L3 cache may include 4 MB or more, depending on embodiment, although smaller cache sizes may be used.

In at least one embodiment, one or more of SoC(s) 1704 may include one or more accelerator(s) 1714 (e.g., hardware accelerators, software accelerators, or a combination thereof). In at least one embodiment, SoC(s) 1704 may include a hardware acceleration cluster that may include optimized hardware accelerators and/or large on-chip memory. In at least one embodiment, large on-chip memory (e.g., 4 MB of SRAM), may enable hardware acceleration cluster to accelerate neural networks and other calculations. In at least one embodiment, hardware acceleration cluster may be used to complement GPU(s) 1708 and to off-load some of tasks of GPU(s) 1708 (e.g., to free up more cycles of GPU(s) 1708 for performing other tasks). In at least one embodiment, accelerator(s) 1714 could be used for targeted workloads (e.g., perception, convolutional neural networks (“CNNs”), recurrent neural networks (“RNNs”), etc.) that are stable enough to be amenable to acceleration. In at least one embodiment, a CNN may include a region-based or regional convolutional neural networks (“RCNNs”) and Fast RCNNs (e.g., as used for object detection) or other type of CNN.

In at least one embodiment, accelerator(s) 1714 (e.g., hardware acceleration cluster) may include a deep learning accelerator(s) (“DLA”). DLA(s) may include, without limitation, one or more Tensor processing units (“TPUs”) that may be configured to provide an additional ten trillion operations per second for deep learning applications and inferencing. In at least one embodiment, TPUs may be accelerators configured to, and optimized for, performing image processing functions (e.g., for CNNs, RCNNs, etc.). DLA(s) may further be optimized for a specific set of neural network types and floating point operations, as well as inferencing. In at least one embodiment, design of DLA(s) may provide more performance per millimeter than a typical general-purpose GPU, and typically vastly exceeds performance of a CPU. In at least one embodiment, TPU(s) may perform several functions, including a single-instance convolution function, supporting, for example, INT8, INT16, and FP16 data types for both features and weights, as well as post-processor functions. In at least one embodiment, DLA(s) may quickly and efficiently execute neural networks, especially CNNs, on processed or unprocessed data for any of a variety of functions, including, for example and without limitation: a CNN for object identification and detection using data from camera sensors; a CNN for distance estimation using data from camera sensors; a CNN for emergency vehicle detection and identification and detection using data from microphones 1796; a CNN for facial recognition and vehicle owner identification using data from camera sensors; and/or a CNN for security and/or safety related events.

In at least one embodiment, DLA(s) may perform any function of GPU(s) 1708, and by using an inference accelerator, for example, a designer may target either DLA(s) or GPU(s) 1708 for any function. For example, in at least one embodiment, designer may focus processing of CNNs and floating point operations on DLA(s) and leave other functions to GPU(s) 1708 and/or other accelerator(s) 1714.

In at least one embodiment, accelerator(s) 1714 (e.g., hardware acceleration cluster) may include a programmable vision accelerator(s) (“PVA”), which may alternatively be referred to herein as a computer vision accelerator. In at least one embodiment, PVA(s) may be designed and configured to accelerate computer vision algorithms for advanced driver assistance system (“ADAS”) 1738, autonomous driving, augmented reality (“AR”) applications, and/or virtual reality (“VR”) applications. PVA(s) may provide a balance between performance and flexibility. For example, in at least one embodiment, each PVA(s) may include, for example and without limitation, any number of reduced instruction set computer (“RISC”) cores, direct memory access (“DMA”), and/or any number of vector processors.

In at least one embodiment, RISC cores may interact with image sensors (e.g., image sensors of any of cameras described herein), image signal processor(s), and/or like. In at least one embodiment, each of RISC cores may include any amount of memory. In at least one embodiment, RISC cores may use any of a number of protocols, depending on embodiment. In at least one embodiment, RISC cores may execute a real-time operating system (“RTOS”). In at least one embodiment, RISC cores may be implemented using one or more integrated circuit devices, application specific integrated circuits (“ASICs”), and/or memory devices. For example, in at least one embodiment, RISC cores could include an instruction cache and/or a tightly coupled RAM.

In at least one embodiment, DMA may enable components of PVA(s) to access system memory independently of CPU(s) 1706. In at least one embodiment, DMA may support any number of features used to provide optimization to PVA including, but not limited to, supporting multi-dimensional addressing and/or circular addressing. In at least one embodiment, DMA may support up to six or more dimensions of addressing, which may include, without limitation, block width, block height, block depth, horizontal block stepping, vertical block stepping, and/or depth stepping.

In at least one embodiment, vector processors may be programmable processors that may be designed to efficiently and flexibly execute programming for computer vision algorithms and provide signal processing capabilities. In at least one embodiment, PVA may include a PVA core and two vector processing subsystem partitions. In at least one embodiment, PVA core may include a processor subsystem, DMA engine(s) (e.g., two DMA engines), and/or other peripherals. In at least one embodiment, vector processing subsystem may operate as primary processing engine of PVA, and may include a vector processing unit (“VPU”), an instruction cache, and/or vector memory (e.g., “VMEM”). In at least one embodiment, VPU core may include a digital signal processor such as, for example, a single instruction, multiple data (“SIMD”), very long instruction word (“VLIW”) digital signal processor. In at least one embodiment, a combination of SIMD and VLIW may enhance throughput and speed.

In at least one embodiment, each of vector processors may include an instruction cache and may be coupled to dedicated memory. As a result, in at least one embodiment, each of vector processors may be configured to execute independently of other vector processors. In at least one embodiment, vector processors that are included in a particular PVA may be configured to employ data parallelism. For instance, in at least one embodiment, plurality of vector processors included in a single PVA may execute same computer vision algorithm, but on different regions of an image. In at least one embodiment, vector processors included in a particular PVA may simultaneously execute different computer vision algorithms, on same image, or even execute different algorithms on sequential images or portions of an image. In at least one embodiment, among other things, any number of PVAs may be included in hardware acceleration cluster and any number of vector processors may be included in each of PVAs. In at least one embodiment, PVA(s) may include additional error correcting code (“ECC”) memory, to enhance overall system safety.

In at least one embodiment, accelerator(s) 1714 (e.g., hardware acceleration cluster) may include a computer vision network on-chip and static random-access memory (“SRAM”), for providing a high-bandwidth, low latency SRAM for accelerator(s) 1714. In at least one embodiment, on-chip memory may include at least 4 MB SRAM, consisting of, for example and without limitation, eight field-configurable memory blocks, that may be accessible by both PVA and DLA. In at least one embodiment, each pair of memory blocks may include an advanced peripheral bus (“APB”) interface, configuration circuitry, a controller, and a multiplexer. In at least one embodiment, any type of memory may be used. In at least one embodiment, PVA and DLA may access memory via a backbone that provides PVA and DLA with high-speed access to memory. In at least one embodiment, backbone may include a computer vision network on-chip that interconnects PVA and DLA to memory (e.g., using APB).

In at least one embodiment, computer vision network on-chip may include an interface that determines, before transmission of any control signal/address/data, that both PVA and DLA provide ready and valid signals. In at least one embodiment, an interface may provide for separate phases and separate channels for transmitting control signals/addresses/data, as well as burst-type communications for continuous data transfer. In at least one embodiment, an interface may comply with International Organization for Standardization (“ISO”) 26262 or International Electrotechnical Commission (“IEC”) 61508 standards, although other standards and protocols may be used.

In at least one embodiment, one or more of SoC(s) 1704 may include a real-time ray-tracing hardware accelerator. In at least one embodiment, real-time ray-tracing hardware accelerator may be used to quickly and efficiently determine positions and extents of objects (e.g., within a world model), to generate real-time visualization simulations, for RADAR signal interpretation, for sound propagation synthesis and/or analysis, for simulation of SONAR systems, for general wave propagation simulation, for comparison to LIDAR data for purposes of localization and/or other functions, and/or for other uses.

In at least one embodiment, accelerator(s) 1714 (e.g., hardware accelerator cluster) have a wide array of uses for autonomous driving. In at least one embodiment, PVA may be a programmable vision accelerator that may be used for key processing stages in ADAS and autonomous vehicles. In at least one embodiment, PVA's capabilities are a good match for algorithmic domains needing predictable processing, at low power and low latency. In other words, PVA performs well on semi-dense or dense regular computation, even on small data sets, which need predictable run-times with low latency and low power. In at least one embodiment, autonomous vehicles, such as vehicle 1700, PVAs are designed to run classic computer vision algorithms, as they are efficient at object detection and operating on integer math.

For example, according to at least one embodiment of technology, PVA is used to perform computer stereo vision. In at least one embodiment, semi-global matching-based algorithm may be used in some examples, although this is not intended to be limiting. In at least one embodiment, applications for Level 3-5 autonomous driving use motion estimation/stereo matching on-the-fly (e.g., structure from motion, pedestrian recognition, lane detection, etc.). In at least one embodiment, PVA may perform computer stereo vision function on inputs from two monocular cameras.

In at least one embodiment, PVA may be used to perform dense optical flow. For example, in at least one embodiment, PVA could process raw RADAR data (e.g., using a 4D Fast Fourier Transform) to provide processed RADAR data. In at least one embodiment, PVA is used for time of flight depth processing, by processing raw time of flight data to provide processed time of flight data, for example.

In at least one embodiment, DLA may be used to run any type of network to enhance control and driving safety, including for example and without limitation, a neural network that outputs a measure of confidence for each object detection. In at least one embodiment, confidence may be represented or interpreted as a probability, or as providing a relative “weight” of each detection compared to other detections. In at least one embodiment, confidence enables a system to make further decisions regarding which detections should be considered as true positive detections rather than false positive detections. For example, in at least one embodiment, a system may set a threshold value for confidence and consider only detections exceeding threshold value as true positive detections. In an embodiment in which an automatic emergency braking (“AEB”) system is used, false positive detections would cause vehicle to automatically perform emergency braking, which is obviously undesirable. In at least one embodiment, highly confident detections may be considered as triggers for AEB. In at least one embodiment, DLA may run a neural network for regressing confidence value. In at least one embodiment, neural network may take as its input at least some subset of parameters, such as bounding box dimensions, ground plane estimate obtained (e.g. from another subsystem), output from IMU sensor(s) 1766 that correlates with vehicle 1700 orientation, distance, 3D location estimates of object obtained from neural network and/or other sensors (e.g., LIDAR sensor(s) 1764 or RADAR sensor(s) 1760), among others.

In at least one embodiment, one or more of SoC(s) 1704 may include data store(s) 1716 (e.g., memory). In at least one embodiment, data store(s) 1716 may be on-chip memory of SoC(s) 1704, which may store neural networks to be executed on GPU(s) 1708 and/or DLA. In at least one embodiment, data store(s) 1716 may be large enough in capacity to store multiple instances of neural networks for redundancy and safety. In at least one embodiment, data store(s) 1712 may comprise L2 or L3 cache(s).

In at least one embodiment, one or more of SoC(s) 1704 may include any number of processor(s) 1710 (e.g., embedded processors). Processor(s) 1710 may include a boot and power management processor that may be a dedicated processor and subsystem to handle boot power and management functions and related security enforcement. In at least one embodiment, boot and power management processor may be a part of SoC(s) 1704 boot sequence and may provide runtime power management services. In at least one embodiment, boot power and management processor may provide clock and voltage programming, assistance in system low power state transitions, management of SoC(s) 1704 thermals and temperature sensors, and/or management of SoC(s) 1704 power states. In at least one embodiment, each temperature sensor may be implemented as a ring-oscillator whose output frequency is proportional to temperature, and SoC(s) 1704 may use ring-oscillators to detect temperatures of CPU(s) 1706, GPU(s) 1708, and/or accelerator(s) 1714. In at least one embodiment, if temperatures are determined to exceed a threshold, then boot and power management processor may enter a temperature fault routine and put SoC(s) 1704 into a lower power state and/or put vehicle 1700 into a chauffeur to safe stop mode (e.g., bring vehicle 1700 to a safe stop).

In at least one embodiment, processor(s) 1710 may further include a set of embedded processors that may serve as an audio processing engine. In at least one embodiment, audio processing engine may be an audio subsystem that enables full hardware support for multi-channel audio over multiple interfaces, and a broad and flexible range of audio I/O interfaces. In at least one embodiment, audio processing engine is a dedicated processor core with a digital signal processor with dedicated RAM.

In at least one embodiment, processor(s) 1710 may further include an always on processor engine that may provide necessary hardware features to support low power sensor management and wake use cases. In at least one embodiment, always on processor engine may include, without limitation, a processor core, a tightly coupled RAM, supporting peripherals (e.g., timers and interrupt controllers), various I/O controller peripherals, and routing logic.

In at least one embodiment, processor(s) 1710 may further include a safety cluster engine that includes, without limitation, a dedicated processor subsystem to handle safety management for automotive applications. In at least one embodiment, safety cluster engine may include, without limitation, two or more processor cores, a tightly coupled RAM, support peripherals (e.g., timers, an interrupt controller, etc.), and/or routing logic. In a safety mode, two or more cores may operate, in at least one embodiment, in a lockstep mode and function as a single core with comparison logic to detect any differences between their operations. In at least one embodiment, processor(s) 1710 may further include a real-time camera engine that may include, without limitation, a dedicated processor subsystem for handling real-time camera management. In at least one embodiment, processor(s) 1710 may further include a high-dynamic range signal processor that may include, without limitation, an image signal processor that is a hardware engine that is part of camera processing pipeline.

In at least one embodiment, processor(s) 1710 may include a video image compositor that may be a processing block (e.g., implemented on a microprocessor) that implements video post-processing functions needed by a video playback application to produce final image for player window. In at least one embodiment, video image compositor may perform lens distortion correction on wide-view camera(s) 1770, surround camera(s) 1774, and/or on in-cabin monitoring camera sensor(s). In at least one embodiment, in-cabin monitoring camera sensor(s) are preferably monitored by a neural network running on another instance of SoC 1704, configured to identify in cabin events and respond accordingly. In at least one embodiment, an in-cabin system may perform, without limitation, lip reading to activate cellular service and place a phone call, dictate emails, change vehicle's destination, activate or change vehicle's infotainment system and settings, or provide voice-activated web surfing. In at least one embodiment, certain functions are available to driver when vehicle is operating in an autonomous mode and are disabled otherwise.

In at least one embodiment, video image compositor may include enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, in at least one embodiment, where motion occurs in a video, noise reduction weights spatial information appropriately, decreasing weight of information provided by adjacent frames. In at least one embodiment, where an image or portion of an image does not include motion, temporal noise reduction performed by video image compositor may use information from previous image to reduce noise in current image.

In at least one embodiment, video image compositor may also be configured to perform stereo rectification on input stereo lens frames. In at least one embodiment, video image compositor may further be used for user interface composition when operating system desktop is in use, and GPU(s) 1708 are not required to continuously render new surfaces. In at least one embodiment, when GPU(s) 1708 are powered on and active doing 3D rendering, video image compositor may be used to offload GPU(s) 1708 to improve performance and responsiveness.

In at least one embodiment, one or more of SoC(s) 1704 may further include a mobile industry processor interface (“MIPI”) camera serial interface for receiving video and input from cameras, a high-speed interface, and/or a video input block that may be used for camera and related pixel input functions. In at least one embodiment, one or more of SoC(s) 1704 may further include an input/output controller(s) that may be controlled by software and may be used for receiving I/O signals that are uncommitted to a specific role.

In at least one embodiment, one or more of SoC(s) 1704 may further include a broad range of peripheral interfaces to enable communication with peripherals, audio encoders/decoders (“codecs”), power management, and/or other devices. SoC(s) 1704 may be used to process data from cameras (e.g., connected over Gigabit Multimedia Serial Link and Ethernet), sensors (e.g., LIDAR sensor(s) 1764, RADAR sensor(s) 1760, etc. that may be connected over Ethernet), data from bus 1702 (e.g., speed of vehicle 1700, steering wheel position, etc.), data from GNSS sensor(s) 1758 (e.g., connected over Ethernet or CAN bus), etc. In at least one embodiment, one or more of SoC(s) 1704 may further include dedicated high-performance mass storage controllers that may include their own DMA engines, and that may be used to free CPU(s) 1706 from routine data management tasks.

In at least one embodiment, SoC(s) 1704 may be an end-to-end platform with a flexible architecture that spans automation levels 3-5, thereby providing a comprehensive functional safety architecture that leverages and makes efficient use of computer vision and ADAS techniques for diversity and redundancy, provides a platform for a flexible, reliable driving software stack, along with deep learning tools. In at least one embodiment, SoC(s) 1704 may be faster, more reliable, and even more energy-efficient and space-efficient than conventional systems. For example, in at least one embodiment, accelerator(s) 1714, when combined with CPU(s) 1706, GPU(s) 1708, and data store(s) 1716, may provide for a fast, efficient platform for level 3-5 autonomous vehicles.

In at least one embodiment, computer vision algorithms may be executed on CPUs, which may be configured using high-level programming language, such as C programming language, to execute a wide variety of processing algorithms across a wide variety of visual data. However, in at least one embodiment, CPUs are oftentimes unable to meet performance requirements of many computer vision applications, such as those related to execution time and power consumption, for example. In at least one embodiment, many CPUs are unable to execute complex object detection algorithms in real-time, which is used in in-vehicle ADAS applications and in practical Level 3-5 autonomous vehicles.

Embodiments described herein allow for multiple neural networks to be performed simultaneously and/or sequentially, and for results to be combined together to enable Level 3-5 autonomous driving functionality. For example, in at least one embodiment, a CNN executing on DLA or discrete GPU (e.g., GPU(s) 1720) may include text and word recognition, allowing supercomputer to read and understand traffic signs, including signs for which neural network has not been specifically trained. In at least one embodiment, DLA may further include a neural network that is able to identify, interpret, and provide semantic understanding of sign, and to pass that semantic understanding to path planning modules running on CPU Complex.

In at least one embodiment, multiple neural networks may be run simultaneously, as for Level 3, 4, or 5 driving. For example, in at least one embodiment, a warning sign consisting of “Caution: flashing lights indicate icy conditions,” along with an electric light, may be independently or collectively interpreted by several neural networks. In at least one embodiment, sign itself may be identified as a traffic sign by a first deployed neural network (e.g., a neural network that has been trained), text “flashing lights indicate icy conditions” may be interpreted by a second deployed neural network, which informs vehicle's path planning software (preferably executing on CPU Complex) that when flashing lights are detected, icy conditions exist. In at least one embodiment, flashing light may be identified by operating a third deployed neural network over multiple frames, informing vehicle's path-planning software of presence (or absence) of flashing lights. In at least one embodiment, all three neural networks may run simultaneously, such as within DLA and/or on GPU(s) 1708.

In at least one embodiment, a CNN for facial recognition and vehicle owner identification may use data from camera sensors to identify presence of an authorized driver and/or owner of vehicle 1700. In at least one embodiment, an always on sensor processing engine may be used to unlock vehicle when owner approaches driver door and turn on lights, and, in security mode, to disable vehicle when owner leaves vehicle. In this way, SoC(s) 1704 provide for security against theft and/or carjacking.

In at least one embodiment, a CNN for emergency vehicle detection and identification may use data from microphones 1796 to detect and identify emergency vehicle sirens. In at least one embodiment, SoC(s) 1704 use CNN for classifying environmental and urban sounds, as well as classifying visual data. In at least one embodiment, CNN running on DLA is trained to identify relative closing speed of emergency vehicle (e.g., by using Doppler effect). In at least one embodiment, CNN may also be trained to identify emergency vehicles specific to local area in which vehicle is operating, as identified by GNSS sensor(s) 1758. In at least one embodiment, when operating in Europe, CNN will seek to detect European sirens, and when in United States CNN will seek to identify only North American sirens. In at least one embodiment, once an emergency vehicle is detected, a control program may be used to execute an emergency vehicle safety routine, slowing vehicle, pulling over to side of road, parking vehicle, and/or idling vehicle, with assistance of ultrasonic sensor(s) 1762, until emergency vehicle(s) passes.

In at least one embodiment, vehicle 1700 may include CPU(s) 1718 (e.g., discrete CPU(s), or dCPU(s)), that may be coupled to SoC(s) 1704 via a high-speed interconnect (e.g., PCIe). In at least one embodiment, CPU(s) 1718 may include an X86 processor, for example. CPU(s) 1718 may be used to perform any of a variety of functions, including arbitrating potentially inconsistent results between ADAS sensors and SoC(s) 1704, and/or monitoring status and health of controller(s) 1736 and/or an infotainment system on a chip (“infotainment SoC”) 1730, for example.

In at least one embodiment, vehicle 1700 may include GPU(s) 1720 (e.g., discrete GPU(s), or dGPU(s)), that may be coupled to SoC(s) 1704 via a high-speed interconnect (e.g., NVIDIA's NVLINK). In at least one embodiment, GPU(s) 1720 may provide additional artificial intelligence functionality, such as by executing redundant and/or different neural networks, and may be used to train and/or update neural networks based at least in part on input (e.g., sensor data) from sensors of vehicle 1700.

In at least one embodiment, vehicle 1700 may further include network interface 1724 which may include, without limitation, wireless antenna(s) 1726 (e.g., one or more wireless antennas 1726 for different communication protocols, such as a cellular antenna, a Bluetooth antenna, etc.). In at least one embodiment, network interface 1724 may be used to enable wireless connectivity over Internet with cloud (e.g., with server(s) and/or other network devices), with other vehicles, and/or with computing devices (e.g., client devices of passengers). In at least one embodiment, to communicate with other vehicles, a direct link may be established between vehicle 170 and other vehicle and/or an indirect link may be established (e.g., across networks and over Internet). In at least one embodiment, direct links may be provided using a vehicle-to-vehicle communication link. Vehicle-to-vehicle communication link may provide vehicle 1700 information about vehicles in proximity to vehicle 1700 (e.g., vehicles in front of, on side of, and/or behind vehicle 1700). In at least one embodiment, aforementioned functionality may be part of a cooperative adaptive cruise control functionality of vehicle 1700.

In at least one embodiment, network interface 1724 may include an SoC that provides modulation and demodulation functionality and enables controller(s) 1736 to communicate over wireless networks. In at least one embodiment, network interface 1724 may include a radio frequency front-end for up-conversion from baseband to radio frequency, and down conversion from radio frequency to baseband. In at least one embodiment, frequency conversions may be performed in any technically feasible fashion. For example, frequency conversions could be performed through well-known processes, and/or using super-heterodyne processes. In at least one embodiment, radio frequency front end functionality may be provided by a separate chip. In at least one embodiment, network interface may include wireless functionality for communicating over LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-Wave, ZigBee, LoRaWAN, and/or other wireless protocols.

In at least one embodiment, vehicle 1700 may further include data store(s) 1728 which may include, without limitation, off-chip (e.g., off SoC(s) 1704) storage. In at least one embodiment, data store(s) 1728 may include, without limitation, one or more storage elements including RAM, SRAM, dynamic random-access memory (“DRAM”), video random-access memory (“VRAM”), Flash, hard disks, and/or other components and/or devices that may store at least one bit of data.

In at least one embodiment, vehicle 1700 may further include GNSS sensor(s) 1758 (e.g., GPS and/or assisted GPS sensors), to assist in mapping, perception, occupancy grid generation, and/or path planning functions. In at least one embodiment, any number of GNSS sensor(s) 1758 may be used, including, for example and without limitation, a GPS using a USB connector with an Ethernet to Serial (e.g., RS-232) bridge.

In at least one embodiment, vehicle 1700 may further include RADAR sensor(s) 1760. RADAR sensor(s) 1760 may be used by vehicle 1700 for long-range vehicle detection, even in darkness and/or severe weather conditions. In at least one embodiment, RADAR functional safety levels may be ASIL B. RADAR sensor(s) 1760 may use CAN and/or bus 1702 (e.g., to transmit data generated by RADAR sensor(s) 1760) for control and to access object tracking data, with access to Ethernet to access raw data in some examples. In at least one embodiment, wide variety of RADAR sensor types may be used. For example, and without limitation, RADAR sensor(s) 1760 may be suitable for front, rear, and side RADAR use. In at least one embodiment, one or more of RADAR sensors(s) 1760 are Pulse Doppler RADAR sensor(s).

In at least one embodiment, RADAR sensor(s) 1760 may include different configurations, such as long-range with narrow field of view, short-range with wide field of view, short-range side coverage, etc. In at least one embodiment, long-range RADAR may be used for adaptive cruise control functionality. In at least one embodiment, long-range RADAR systems may provide a broad field of view realized by two or more independent scans, such as within a 250 m range. In at least one embodiment, RADAR sensor(s) 1760 may help in distinguishing between static and moving objects, and may be used by ADAS system 1738 for emergency brake assist and forward collision warning. Sensors 1760(s) included in a long-range RADAR system may include, without limitation, monostatic multimodal RADAR with multiple (e.g., six or more) fixed RADAR antennae and a high-speed CAN and FlexRay interface. In at least one embodiment, with six antennae, central four antennae may create a focused beam pattern, designed to record vehicle's 1700 surroundings at higher speeds with minimal interference from traffic in adjacent lanes. In at least one embodiment, other two antennae may expand field of view, making it possible to quickly detect vehicles entering or leaving vehicle's 1700 lane.

In at least one embodiment, mid-range RADAR systems may include, as an example, a range of up to 160 m (front) or 80 m (rear), and a field of view of up to 42 degrees (front) or 150 degrees (rear). In at least one embodiment, short-range RADAR systems may include, without limitation, any number of RADAR sensor(s) 1760 designed to be installed at both ends of rear bumper. When installed at both ends of rear bumper, in at least one embodiment, a RADAR sensor system may create two beams that constantly monitor blind spot in rear and next to vehicle. In at least one embodiment, short-range RADAR systems may be used in ADAS system 1738 for blind spot detection and/or lane change assist.

In at least one embodiment, vehicle 1700 may further include ultrasonic sensor(s) 1762. Ultrasonic sensor(s) 1762, which may be positioned at front, back, and/or sides of vehicle 1700, may be used for park assist and/or to create and update an occupancy grid. In at least one embodiment, a wide variety of ultrasonic sensor(s) 1762 may be used, and different ultrasonic sensor(s) 1762 may be used for different ranges of detection (e.g., 2.5 m, 4 m). In at least one embodiment, ultrasonic sensor(s) 1762 may operate at functional safety levels of ASIL B.

In at least one embodiment, vehicle 1700 may include LIDAR sensor(s) 1764. LIDAR sensor(s) 1764 may be used for object and pedestrian detection, emergency braking, collision avoidance, and/or other functions. In at least one embodiment, LIDAR sensor(s) 1764 may be functional safety level ASIL B. In at least one embodiment, vehicle 1700 may include multiple LIDAR sensors 1764 (e.g., two, four, six, etc.) that may use Ethernet (e.g., to provide data to a Gigabit Ethernet switch).

In at least one embodiment, LIDAR sensor(s) 1764 may be capable of providing a list of objects and their distances for a 360-degree field of view. In at least one embodiment, commercially available LIDAR sensor(s) 1764 may have an advertised range of approximately 100 m, with an accuracy of 2 cm-3 cm, and with support for a 100 Mbps Ethernet connection, for example. In at least one embodiment, one or more non-protruding LIDAR sensors 1764 may be used. In such an embodiment, LIDAR sensor(s) 1764 may be implemented as a small device that may be embedded into front, rear, sides, and/or corners of vehicle 1700. In at least one embodiment, LIDAR sensor(s) 1764, in such an embodiment, may provide up to a 120-degree horizontal and 35-degree vertical field-of-view, with a 200 m range even for low-reflectivity objects. In at least one embodiment, front-mounted LIDAR sensor(s) 1764 may be configured for a horizontal field of view between 45 degrees and 135 degrees.

In at least one embodiment, LIDAR technologies, such as 3D flash LIDAR, may also be used. 3D Flash LIDAR uses a flash of a laser as a transmission source, to illuminate surroundings of vehicle 1700 up to approximately 200 m. In at least one embodiment, a flash LIDAR unit includes, without limitation, a receptor, which records laser pulse transit time and reflected light on each pixel, which in turn corresponds to range from vehicle 1700 to objects. In at least one embodiment, flash LIDAR may allow for highly accurate and distortion-free images of surroundings to be generated with every laser flash. In at least one embodiment, four flash LIDAR sensors may be deployed, one at each side of vehicle 1700. In at least one embodiment, 3D flash LIDAR systems include, without limitation, a solid-state 3D staring array LIDAR camera with no moving parts other than a fan (e.g., a non-scanning LIDAR device). In at least one embodiment, flash LIDAR device may use a 5 nanosecond class I (eye-safe) laser pulse per frame and may capture reflected laser light in form of 3D range point clouds and co-registered intensity data.

In at least one embodiment, vehicle may further include IMU sensor(s) 1766. In at least one embodiment, IMU sensor(s) 1766 may be located at a center of rear axle of vehicle 1700, in at least one embodiment. In at least one embodiment, IMU sensor(s) 1766 may include, for example and without limitation, accelerometer(s), magnetometer(s), gyroscope(s), magnetic compass(es), and/or other sensor types. In at least one embodiment, such as in six-axis applications, IMU sensor(s) 1766 may include, without limitation, accelerometers and gyroscopes. In at least one embodiment, such as in nine-axis applications, IMU sensor(s) 1766 may include, without limitation, accelerometers, gyroscopes, and magnetometers.

In at least one embodiment, IMU sensor(s) 1766 may be implemented as a miniature, high performance GPS-Aided Inertial Navigation System (“GPS/INS”) that combines micro-electro-mechanical systems (“MEMS”) inertial sensors, a high-sensitivity GPS receiver, and advanced Kalman filtering algorithms to provide estimates of position, velocity, and attitude. In at least one embodiment, IMU sensor(s) 1766 may enable vehicle 1700 to estimate heading without requiring input from a magnetic sensor by directly observing and correlating changes in velocity from GPS to IMU sensor(s) 1766. In at least one embodiment, IMU sensor(s) 1766 and GNSS sensor(s) 1758 may be combined in a single integrated unit.

In at least one embodiment, vehicle 1700 may include microphone(s) 1796 placed in and/or around vehicle 1700. In at least one embodiment, microphone(s) 1796 may be used for emergency vehicle detection and identification, among other things.

In at least one embodiment, vehicle 1700 may further include any number of camera types, including stereo camera(s) 1768, wide-view camera(s) 1770, infrared camera(s) 1772, surround camera(s) 1774, long-range camera(s) 1798, mid-range camera(s) 1776, and/or other camera types. In at least one embodiment, cameras may be used to capture image data around an entire periphery of vehicle 1700. In at least one embodiment, types of cameras used depends vehicle 1700. In at least one embodiment, any combination of camera types may be used to provide necessary coverage around vehicle 1700. In at least one embodiment, number of cameras may differ depending on embodiment. For example, in at least one embodiment, vehicle 1700 could include six cameras, seven cameras, ten cameras, twelve cameras, or another number of cameras. Cameras may support, as an example and without limitation, Gigabit Multimedia Serial Link (“GMSL”) and/or Gigabit Ethernet. In at least one embodiment, each of camera(s) is described with more detail previously herein with respect to FIG. 17A and FIG. 17B.

In at least one embodiment, vehicle 1700 may further include vibration sensor(s) 1742. Vibration sensor(s) 1742 may measure vibrations of components of vehicle 1700, such as axle(s). For example, in at least one embodiment, changes in vibrations may indicate a change in road surfaces. In at least one embodiment, when two or more vibration sensors 1742 are used, differences between vibrations may be used to determine friction or slippage of road surface (e.g., when difference in vibration is between a power-driven axle and a freely rotating axle).

In at least one embodiment, vehicle 1700 may include ADAS system 1738. ADAS system 1738 may include, without limitation, an SoC, in some examples. In at least one embodiment, ADAS system 1738 may include, without limitation, any number and combination of an autonomous/adaptive/automatic cruise control (“ACC”) system, a cooperative adaptive cruise control (“CACC”) system, a forward crash warning (“FCW”) system, an automatic emergency braking (“AEB”) system, a lane departure warning (“LDW)” system, a lane keep assist (“LKA”) system, a blind spot warning (“BSW”) system, a rear cross-traffic warning (“RCTW”) system, a collision warning (“CW”) system, a lane centering (“LC”) system, and/or other systems, features, and/or functionality.

In at least one embodiment, ACC system may use RADAR sensor(s) 1760, LIDAR sensor(s) 1764, and/or any number of camera(s). In at least one embodiment, ACC system may include a longitudinal ACC system and/or a lateral ACC system. In at least one embodiment, longitudinal ACC system monitors and controls distance to vehicle immediately ahead of vehicle 1700 and automatically adjust speed of vehicle 1700 to maintain a safe distance from vehicles ahead. In at least one embodiment, lateral ACC system performs distance keeping, and advises vehicle 1700 to change lanes when necessary. In at least one embodiment, lateral ACC is related to other ADAS applications such as LC and CW.

In at least one embodiment, CACC system uses information from other vehicles that may be received via network interface 1724 and/or wireless antenna(s) 1726 from other vehicles via a wireless link, or indirectly, over a network connection (e.g., over Internet). In at least one embodiment, direct links may be provided by a vehicle-to-vehicle (“V2V”) communication link, while indirect links may be provided by an infrastructure-to-vehicle (“2V”) communication link. In general, V2V communication concept provides information about immediately preceding vehicles (e.g., vehicles immediately ahead of and in same lane as vehicle 1700), while I2V communication concept provides information about traffic further ahead. In at least one embodiment, CACC system may include either or both I2V and V2V information sources. In at least one embodiment, given information of vehicles ahead of vehicle 1700, CACC system may be more reliable and it has potential to improve traffic flow smoothness and reduce congestion on road.

In at least one embodiment, FCW system is designed to alert driver to a hazard, so that driver may take corrective action. In at least one embodiment, FCW system uses a front-facing camera and/or RADAR sensor(s) 1760, coupled to a dedicated processor, DSP, FPGA, and/or ASIC, that is electrically coupled to driver feedback, such as a display, speaker, and/or vibrating component. In at least one embodiment, FCW system may provide a warning, such as in form of a sound, visual warning, vibration and/or a quick brake pulse.

In at least one embodiment, AEB system detects an impending forward collision with another vehicle or other object, and may automatically apply brakes if driver does not take corrective action within a specified time or distance parameter. In at least one embodiment, AEB system may use front-facing camera(s) and/or RADAR sensor(s) 1760, coupled to a dedicated processor, DSP, FPGA, and/or ASIC. In at least one embodiment, when AEB system detects a hazard, AEB system typically first alerts driver to take corrective action to avoid collision and, if driver does not take corrective action, AEB system may automatically apply brakes in an effort to prevent, or at least mitigate, impact of predicted collision. In at least one embodiment, AEB system, may include techniques such as dynamic brake support and/or crash imminent braking.

In at least one embodiment, LDW system provides visual, audible, and/or tactile warnings, such as steering wheel or seat vibrations, to alert driver when vehicle 1700 crosses lane markings. In at least one embodiment, LDW system does not activate when driver indicates an intentional lane departure, by activating a turn signal. In at least one embodiment, LDW system may use front-side facing cameras, coupled to a dedicated processor, DSP, FPGA, and/or ASIC, that is electrically coupled to driver feedback, such as a display, speaker, and/or vibrating component. In at least one embodiment, LKA system is a variation of LDW system. LKA system provides steering input or braking to correct vehicle 1700 if vehicle 1700 starts to exit lane.

In at least one embodiment, BSW system detects and warns driver of vehicles in an automobile's blind spot. In at least one embodiment, BSW system may provide a visual, audible, and/or tactile alert to indicate that merging or changing lanes is unsafe. In at least one embodiment, BSW system may provide an additional warning when driver uses a turn signal. In at least one embodiment, BSW system may use rear-side facing camera(s) and/or RADAR sensor(s) 1760, coupled to a dedicated processor, DSP, FPGA, and/or ASIC, that is electrically coupled to driver feedback, such as a display, speaker, and/or vibrating component.

In at least one embodiment, RCTW system may provide visual, audible, and/or tactile notification when an object is detected outside rear-camera range when vehicle 1700 is backing up. In at least one embodiment, RCTW system includes AEB system to ensure that vehicle brakes are applied to avoid a crash. In at least one embodiment, RCTW system may use one or more rear-facing RADAR sensor(s) 1760, coupled to a dedicated processor, DSP, FPGA, and/or ASIC, that is electrically coupled to driver feedback, such as a display, speaker, and/or vibrating component.

In at least one embodiment, conventional ADAS systems may be prone to false positive results which may be annoying and distracting to a driver, but typically are not catastrophic, because conventional ADAS systems alert driver and allow driver to decide whether a safety condition truly exists and act accordingly. In at least one embodiment, vehicle 1700 itself decides, in case of conflicting results, whether to heed result from a primary computer or a secondary computer (e.g., first controller 1736 or second controller 1736). For example, in at least one embodiment, ADAS system 1738 may be a backup and/or secondary computer for providing perception information to a backup computer rationality module. In at least one embodiment, backup computer rationality monitor may run a redundant diverse software on hardware components to detect faults in perception and dynamic driving tasks. In at least one embodiment, outputs from ADAS system 1738 may be provided to a supervisory MCU. In at least one embodiment, if outputs from primary computer and secondary computer conflict, supervisory MCU determines how to reconcile conflict to ensure safe operation.

In at least one embodiment, primary computer may be configured to provide supervisory MCU with a confidence score, indicating primary computer's confidence in chosen result. In at least one embodiment, if confidence score exceeds a threshold, supervisory MCU may follow primary computer's direction, regardless of whether secondary computer provides a conflicting or inconsistent result. In at least one embodiment, where confidence score does not meet threshold, and where primary and secondary computer indicate different results (e.g., a conflict), supervisory MCU may arbitrate between computers to determine appropriate outcome.

In at least one embodiment, supervisory MCU may be configured to run a neural network(s) that is trained and configured to determine, based at least in part on outputs from primary computer and secondary computer, conditions under which secondary computer provides false alarms. In at least one embodiment, neural network(s) in supervisory MCU may learn when secondary computer's output may be trusted, and when it cannot. For example, in at least one embodiment, when secondary computer is a RADAR-based FCW system, a neural network(s) in supervisory MCU may learn when FCW system is identifying metallic objects that are not, in fact, hazards, such as a drainage grate or manhole cover that triggers an alarm. In at least one embodiment, when secondary computer is a camera-based LDW system, a neural network in supervisory MCU may learn to override LDW when bicyclists or pedestrians are present and a lane departure is, in fact, safest maneuver. In at least one embodiment, supervisory MCU may include at least one of a DLA or GPU suitable for running neural network(s) with associated memory. In at least one embodiment, supervisory MCU may comprise and/or be included as a component of SoC(s) 1704.

In at least one embodiment, ADAS system 1738 may include a secondary computer that performs ADAS functionality using traditional rules of computer vision. In at least one embodiment, secondary computer may use classic computer vision rules (if-then), and presence of a neural network(s) in supervisory MCU may improve reliability, safety and performance. For example, in at least one embodiment, diverse implementation and intentional non-identity makes overall system more fault-tolerant, especially to faults caused by software (or software-hardware interface) functionality. For example, in at least one embodiment, if there is a software bug or error in software running on primary computer, and non-identical software code running on secondary computer provides same overall result, then supervisory MCU may have greater confidence that overall result is correct, and bug in software or hardware on primary computer is not causing material error.

In at least one embodiment, output of ADAS system 1738 may be fed into primary computer's perception block and/or primary computer's dynamic driving task block. For example, in at least one embodiment, if ADAS system 1738 indicates a forward crash warning due to an object immediately ahead, perception block may use this information when identifying objects. In at least one embodiment, secondary computer may have its own neural network which is trained and thus reduces risk of false positives, as described herein.

In at least one embodiment, vehicle 1700 may further include infotainment SoC 1730 (e.g., an in-vehicle infotainment system (IVI)). Although illustrated and described as an SoC, infotainment system 1730, in at least one embodiment, may not be an SoC, and may include, without limitation, two or more discrete components. In at least one embodiment, infotainment SoC 1730 may include, without limitation, a combination of hardware and software that may be used to provide audio (e.g., music, a personal digital assistant, navigational instructions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), phone (e.g., hands-free calling), network connectivity (e.g., LTE, WiFi, etc.), and/or information services (e.g., navigation systems, rear-parking assistance, a radio data system, vehicle related information such as fuel level, total distance covered, brake fuel level, oil level, door open/close, air filter information, etc.) to vehicle 1700. For example, infotainment SoC 1730 could include radios, disk players, navigation systems, video players, USB and Bluetooth connectivity, carputers, in-car entertainment, WiFi, steering wheel audio controls, hands free voice control, a heads-up display (“HUD”), HMI display 1734, a telematics device, a control panel (e.g., for controlling and/or interacting with various components, features, and/or systems), and/or other components. In at least one embodiment, infotainment SoC 1730 may further be used to provide information (e.g., visual and/or audible) to user(s) of vehicle, such as information from ADAS system 1738, autonomous driving information such as planned vehicle maneuvers, trajectories, surrounding environment information (e.g., intersection information, vehicle information, road information, etc.), and/or other information.

In at least one embodiment, infotainment SoC 1730 may include any amount and type of GPU functionality. In at least one embodiment, infotainment SoC 1730 may communicate over bus 1702 (e.g., CAN bus, Ethernet, etc.) with other devices, systems, and/or components of vehicle 1700. In at least one embodiment, infotainment SoC 1730 may be coupled to a supervisory MCU such that GPU of infotainment system may perform some self-driving functions in event that primary controller(s) 1736 (e.g., primary and/or backup computers of vehicle 1700) fail. In at least one embodiment, infotainment SoC 1730 may put vehicle 1700 into a chauffeur to safe stop mode, as described herein.

In at least one embodiment, vehicle 1700 may further include instrument cluster 1732 (e.g., a digital dash, an electronic instrument cluster, a digital instrument panel, etc.). Instrument cluster 1732 may include, without limitation, a controller and/or supercomputer (e.g., a discrete controller or supercomputer). In at least one embodiment, instrument cluster 1732 may include, without limitation, any number and combination of a set of instrumentation such as a speedometer, fuel level, oil pressure, tachometer, odometer, turn indicators, gearshift position indicator, seat belt warning light(s), parking-brake warning light(s), engine-malfunction light(s), supplemental restraint system (e.g., airbag) information, lighting controls, safety system controls, navigation information, etc. In some examples, information may be displayed and/or shared among infotainment SoC 1730 and instrument cluster 1732. In at least one embodiment, instrument cluster 1732 may be included as part of infotainment SoC 1730, or vice versa.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 17C for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the robot can be an autonomous vehicle, and the evaluation and generation networks can be implemented using a computer system on the autonomous vehicle.

FIG. 17D is a diagram of a system 1776 for communication between cloud-based server(s) and autonomous vehicle 1700 of FIG. 17A, according to at least one embodiment. In at least one embodiment, system 1776 may include, without limitation, server(s) 1778, network(s) 1790, and any number and type of vehicles, including vehicle 1700. server(s) 1778 may include, without limitation, a plurality of GPUs 1784(A)-1784(H) (collectively referred to herein as GPUs 1784), PCIe switches 1782(A)-1782(H) (collectively referred to herein as PCIe switches 1782), and/or CPUs 1780(A)-1780(B) (collectively referred to herein as CPUs 1780). GPUs 1784, CPUs 1780, and PCIe switches 1782 may be interconnected with high-speed interconnects such as, for example and without limitation, NVLink interfaces 1788 developed by NVIDIA and/or PCIe connections 1786. In at least one embodiment, GPUs 1784 are connected via an NVLink and/or NVSwitch SoC and GPUs 1784 and PCIe switches 1782 are connected via PCIe interconnects. In at least one embodiment, although eight GPUs 1784, two CPUs 1780, and four PCIe switches 1782 are illustrated, this is not intended to be limiting. In at least one embodiment, each of server(s) 1778 may include, without limitation, any number of GPUs 1784, CPUs 1780, and/or PCIe switches 1782, in any combination. For example, in at least one embodiment, server(s) 1778 could each include eight, sixteen, thirty-two, and/or more GPUs 1784.

In at least one embodiment, server(s) 1778 may receive, over network(s) 1790 and from vehicles, image data representative of images showing unexpected or changed road conditions, such as recently commenced road-work. In at least one embodiment, server(s) 1778 may transmit, over network(s) 1790 and to vehicles, neural networks 1792, updated neural networks 1792, and/or map information 1794, including, without limitation, information regarding traffic and road conditions. In at least one embodiment, updates to map information 1794 may include, without limitation, updates for HD map 1722, such as information regarding construction sites, potholes, detours, flooding, and/or other obstructions. In at least one embodiment, neural networks 1792, updated neural networks 1792, and/or map information 1794 may have resulted from new training and/or experiences represented in data received from any number of vehicles in environment, and/or based at least in part on training performed at a data center (e.g., using server(s) 1778 and/or other servers).

In at least one embodiment, server(s) 1778 may be used to train machine learning models (e.g., neural networks) based at least in part on training data. Training data may be generated by vehicles, and/or may be generated in a simulation (e.g., using a game engine). In at least one embodiment, any amount of training data is tagged (e.g., where associated neural network benefits from supervised learning) and/or undergoes other pre-processing. In at least one embodiment, any amount of training data is not tagged and/or pre-processed (e.g., where associated neural network does not require supervised learning). In at least one embodiment, once machine learning models are trained, machine learning models may be used by vehicles (e.g., transmitted to vehicles over network(s) 1790, and/or machine learning models may be used by server(s) 1778 to remotely monitor vehicles).

In at least one embodiment, server(s) 1778 may receive data from vehicles and apply data to up-to-date real-time neural networks for real-time intelligent inferencing. In at least one embodiment, server(s) 1778 may include deep-learning supercomputers and/or dedicated AI computers powered by GPU(s) 1784, such as a DGX and DGX Station machines developed by NVIDIA. However, in at least one embodiment, server(s) 1778 may include deep learning infrastructure that use CPU-powered data centers.

In at least one embodiment, deep-learning infrastructure of server(s) 1778 may be capable of fast, real-time inferencing, and may use that capability to evaluate and verify health of processors, software, and/or associated hardware in vehicle 1700. For example, in at least one embodiment, deep-learning infrastructure may receive periodic updates from vehicle 1700, such as a sequence of images and/or objects that vehicle 1700 has located in that sequence of images (e.g., via computer vision and/or other machine learning object classification techniques). In at least one embodiment, deep-learning infrastructure may run its own neural network to identify objects and compare them with objects identified by vehicle 1700 and, if results do not match and deep-learning infrastructure concludes that AI in vehicle 1700 is malfunctioning, then server(s) 1778 may transmit a signal to vehicle 1700 instructing a fail-safe computer of vehicle 1700 to assume control, notify passengers, and complete a safe parking maneuver.

In at least one embodiment, server(s) 1778 may include GPU(s) 1784 and one or more programmable inference accelerators (e.g., NVIDIA's TensorRT 3). In at least one embodiment, combination of GPU-powered servers and inference acceleration may make real-time responsiveness possible. In at least one embodiment, such as where performance is less critical, servers powered by CPUs, FPGAs, and other processors may be used for inferencing. In at least one embodiment, hardware structure(s) 1415 are used to perform one or more embodiments. Details regarding hardware structure(x) 1415 are provided herein in conjunction with FIGS. 14A and/or 14B.

Computer Systems

FIG. 18 is a block diagram illustrating an exemplary computer system, which may be a system with interconnected devices and components, a system-on-a-chip (SOC) or some combination thereof 1800 formed with a processor that may include execution units to execute an instruction, according to at least one embodiment. In at least one embodiment, computer system 1800 may include, without limitation, a component, such as a processor 1802 to employ execution units including logic to perform algorithms for process data, in accordance with present disclosure, such as in embodiment described herein. In at least one embodiment, computer system 1800 may include processors, such as PENTIUM® Processor family, Xeon™, Itanium®, XScale™ and/or StrongARM™, Intel® Core™, or Intel® Nervana™ microprocessors available from Intel Corporation of Santa Clara, Calif., although other systems (including PCs having other microprocessors, engineering workstations, set-top boxes and like) may also be used. In at least one embodiment, computer system 1800 may execute a version of WINDOWS' operating system available from Microsoft Corporation of Redmond, Wash., although other operating systems (UNIX and Linux for example), embedded software, and/or graphical user interfaces, may also be used.

Embodiments may be used in other devices such as handheld devices and embedded applications. Some examples of handheld devices include cellular phones, Internet Protocol devices, digital cameras, personal digital assistants (“PDAs”), and handheld PCs. In at least one embodiment, embedded applications may include a microcontroller, a digital signal processor (“DSP”), system on a chip, network computers (“NetPCs”), set-top boxes, network hubs, wide area network (“WAN”) switches, or any other system that may perform one or more instructions in accordance with at least one embodiment.

In at least one embodiment, computer system 1800 may include, without limitation, processor 1802 that may include, without limitation, one or more execution units 1808 to perform machine learning model training and/or inferencing according to techniques described herein. In at least one embodiment, system 18 is a single processor desktop or server system, but in another embodiment system 18 may be a multiprocessor system. In at least one embodiment, processor 1802 may include, without limitation, a complex instruction set computer (“CISC”) microprocessor, a reduced instruction set computing (“RISC”) microprocessor, a very long instruction word (“VLIW”) microprocessor, a processor implementing a combination of instruction sets, or any other processor device, such as a digital signal processor, for example. In at least one embodiment, processor 1802 may be coupled to a processor bus 1810 that may transmit data signals between processor 1802 and other components in computer system 1800.

In at least one embodiment, processor 1802 may include, without limitation, a Level 1 (“L1”) internal cache memory (“cache”) 1804. In at least one embodiment, processor 1802 may have a single internal cache or multiple levels of internal cache. In at least one embodiment, cache memory may reside external to processor 1802. Other embodiments may also include a combination of both internal and external caches depending on particular implementation and needs. In at least one embodiment, register file 1806 may store different types of data in various registers including, without limitation, integer registers, floating point registers, status registers, and instruction pointer register.

In at least one embodiment, execution unit 1808, including, without limitation, logic to perform integer and floating point operations, also resides in processor 1802. Processor 1802 may also include a microcode (“ucode”) read only memory (“ROM”) that stores microcode for certain macro instructions. In at least one embodiment, execution unit 1808 may include logic to handle a packed instruction set 1809. In at least one embodiment, by including packed instruction set 1809 in instruction set of a general-purpose processor 1802, along with associated circuitry to execute instructions, operations used by many multimedia applications may be performed using packed data in a general-purpose processor 1802. In one or more embodiments, many multimedia applications may be accelerated and executed more efficiently by using full width of a processor's data bus for performing operations on packed data, which may eliminate need to transfer smaller units of data across processor's data bus to perform one or more operations one data element at a time.

In at least one embodiment, execution unit 1808 may also be used in microcontrollers, embedded processors, graphics devices, DSPs, and other types of logic circuits. In at least one embodiment, computer system 1800 may include, without limitation, a memory 1820. In at least one embodiment, memory 1820 may be implemented as a Dynamic Random Access Memory (“DRAM”) device, a Static Random Access Memory (“SRAM”) device, flash memory device, or other memory device. Memory 1820 may store instruction(s) 1819 and/or data 1821 represented by data signals that may be executed by processor 1802.

In at least one embodiment, system logic chip may be coupled to processor bus 1810 and memory 1820. In at least one embodiment, system logic chip may include, without limitation, a memory controller hub (“MCH”) 1816, and processor 1802 may communicate with MCH 1816 via processor bus 1810. In at least one embodiment, MCH 1816 may provide a high bandwidth memory path 1818 to memory 1820 for instruction and data storage and for storage of graphics commands, data and textures. In at least one embodiment, MCH 1816 may direct data signals between processor 1802, memory 1820, and other components in computer system 1800 and to bridge data signals between processor bus 1810, memory 1820, and a system I/O 1822. In at least one embodiment, system logic chip may provide a graphics port for coupling to a graphics controller. In at least one embodiment, MCH 1816 may be coupled to memory 1820 through a high bandwidth memory path 1818 and graphics/video card 1812 may be coupled to MCH 1816 through an Accelerated Graphics Port (“AGP”) interconnect 1814.

In at least one embodiment, computer system 1800 may use system I/O 1822 that is a proprietary hub interface bus to couple MCH 1816 to I/O controller hub (“ICH”) 1830. In at least one embodiment, ICH 1830 may provide direct connections to some I/O devices via a local I/O bus. In at least one embodiment, local I/O bus may include, without limitation, a high-speed I/O bus for connecting peripherals to memory 1820, chipset, and processor 1802. Examples may include, without limitation, an audio controller 1829, a firmware hub (“flash BIOS”) 1828, a wireless transceiver 1826, a data storage 1824, a legacy I/O controller 1823 containing user input and keyboard interfaces, a serial expansion port 1827, such as Universal Serial Bus (“USB”), and a network controller 1834. Data storage 1824 may comprise a hard disk drive, a floppy disk drive, a CD-ROM device, a flash memory device, or other mass storage device.

In at least one embodiment, FIG. 18 illustrates a system, which includes interconnected hardware devices or “chips”, whereas in other embodiments, FIG. 18 may illustrate an exemplary System on a Chip (“SoC”). In at least one embodiment, devices illustrated in FIG. 18 may be interconnected with proprietary interconnects, standardized interconnects (e.g., PCIe) or some combination thereof. In at least one embodiment, one or more components of system 1800 are interconnected using compute express link (CXL) interconnects.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 18 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

FIG. 19 is a block diagram illustrating an electronic device 1900 for utilizing a processor 1910, according to at least one embodiment. In at least one embodiment, electronic device 1900 may be, for example and without limitation, a notebook, a tower server, a rack server, a blade server, a laptop, a desktop, a tablet, a mobile device, a phone, an embedded computer, or any other suitable electronic device.

In at least one embodiment, system 1900 may include, without limitation, processor 1910 communicatively coupled to any suitable number or kind of components, peripherals, modules, or devices. In at least one embodiment, processor 1910 coupled using a bus or interface, such as a 1° C. bus, a System Management Bus (“SMBus”), a Low Pin Count (LPC) bus, a Serial Peripheral Interface (“SPI”), a High Definition Audio (“HDA”) bus, a Serial Advance Technology Attachment (“SATA”) bus, a Universal Serial Bus (“USB”) (versions 1, 2, 3), or a Universal Asynchronous Receiver/Transmitter (“UART”) bus. In at least one embodiment, FIG. 19 illustrates a system, which includes interconnected hardware devices or “chips”, whereas in other embodiments, FIG. 19 may illustrate an exemplary System on a Chip (“SoC”). In at least one embodiment, devices illustrated in FIG. 19 may be interconnected with proprietary interconnects, standardized interconnects (e.g., PCIe) or some combination thereof. In at least one embodiment, one or more components of FIG. 19 are interconnected using compute express link (CXL) interconnects.

In at least one embodiment, FIG. 19 may include a display 1924, a touch screen 1925, a touch pad 1930, a Near Field Communications unit (“NFC”) 1945, a sensor hub 1940, a thermal sensor 1946, an Express Chipset (“EC”) 1935, a Trusted Platform Module (“TPM”) 1938, BIOS/firmware/flash memory (“BIOS, FW Flash”) 1922, a DSP 1960, a drive (“SSD or HDD”) 1920 such as a Solid State Disk (“SSD”) or a Hard Disk Drive (“HDD”), a wireless local area network unit (“WLAN”) 1950, a Bluetooth unit 1952, a Wireless Wide Area Network unit (“WWAN”) 1956, a Global Positioning System (GPS) 1955, a camera (“USB 3.0 camera”) 1954 such as a USB 3.0 camera, or a Low Power Double Data Rate (“LPDDR”) memory unit (“LPDDR3”) 1915 implemented in, for example, LPDDR3 standard. These components may each be implemented in any suitable manner.

In at least one embodiment, other components may be communicatively coupled to processor 1910 through components discussed above. In at least one embodiment, an accelerometer 1941, Ambient Light Sensor (“ALS”) 1942, compass 1943, and a gyroscope 1944 may be communicatively coupled to sensor hub 1940. In at least one embodiment, thermal sensor 1939, a fan 1937, a keyboard 1946, and a touch pad 1930 may be communicatively coupled to EC 1935. In at least one embodiment, speaker 1963, a headphones 1964, and a microphone (“mic”) 1965 may be communicatively coupled to an audio unit (“audio codec and class d amp”) 1964, which may in turn be communicatively coupled to DSP 1960. In at least one embodiment, audio unit 1964 may include, for example and without limitation, an audio coder/decoder (“codec”) and a class D amplifier. In at least one embodiment, SIM card (“SIM”) 1957 may be communicatively coupled to WWAN unit 1956. In at least one embodiment, components such as WLAN unit 1950 and Bluetooth unit 1952, as well as WWAN unit 1956 may be implemented in a Next Generation Form Factor (“NGFF”).

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 19 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 20 illustrates a computer system 2000, according to at least one embodiment. In at least one embodiment, computer system 2000 is configured to implement various processes and methods described throughout this disclosure.

In at least one embodiment, computer system 2000 comprises, without limitation, at least one central processing unit (“CPU”) 2002 that is connected to a communication bus 2010 implemented using any suitable protocol, such as PCI (“Peripheral Component Interconnect”), peripheral component interconnect express (“PCI-Express”), AGP (“Accelerated Graphics Port”), HyperTransport, or any other bus or point-to-point communication protocol(s). In at least one embodiment, computer system 2000 includes, without limitation, a main memory 2004 and control logic (e.g., implemented as hardware, software, or a combination thereof) and data are stored in main memory 2004 which may take form of random access memory (“RAM”). In at least one embodiment, a network interface subsystem (“network interface”) 2022 provides an interface to other computing devices and networks for receiving data from and transmitting data to other systems from computer system 2000.

In at least one embodiment, computer system 2000, in at least one embodiment, includes, without limitation, input devices 2008, parallel processing system 2012, and display devices 2006 which can be implemented using a conventional cathode ray tube (“CRT”), liquid crystal display (“LCD”), light emitting diode (“LED”), plasma display, or other suitable display technologies. In at least one embodiment, user input is received from input devices 2008 such as keyboard, mouse, touchpad, microphone, and more. In at least one embodiment, each of foregoing modules can be situated on a single semiconductor platform to form a processing system.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 20 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a computer system as described above.

FIG. 21 illustrates a computer system 2100, according to at least one embodiment. In at least one embodiment, computer system 2100 includes, without limitation, a computer 2110 and a USB stick 2120. In at least one embodiment, computer 2110 may include, without limitation, any number and type of processor(s) (not shown) and a memory (not shown). In at least one embodiment, computer 2110 includes, without limitation, a server, a cloud instance, a laptop, and a desktop computer.

In at least one embodiment, USB stick 2120 includes, without limitation, a processing unit 2130, a USB interface 2140, and USB interface logic 2150. In at least one embodiment, processing unit 2130 may be any instruction execution system, apparatus, or device capable of executing instructions. In at least one embodiment, processing unit 2130 may include, without limitation, any number and type of processing cores (not shown). In at least one embodiment, processing core 2130 comprises an application specific integrated circuit (“ASIC”) that is optimized to perform any amount and type of operations associated with machine learning. For instance, in at least one embodiment, processing core 2130 is a tensor processing unit (“TPC”) that is optimized to perform machine learning inference operations. In at least one embodiment, processing core 2130 is a vision processing unit (“VPU”) that is optimized to perform machine vision and machine learning inference operations.

In at least one embodiment, USB interface 2140 may be any type of USB connector or USB socket. For instance, in at least one embodiment, USB interface 2140 is a USB 3.0 Type-C socket for data and power. In at least one embodiment, USB interface 2140 is a USB 3.0 Type-A connector. In at least one embodiment, USB interface logic 2150 may include any amount and type of logic that enables processing unit 2130 to interface with or devices (e.g., computer 2110) via USB connector 2140.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 21 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a computer system as described above.

FIG. 22A illustrates an exemplary architecture in which a plurality of GPUs 2210-2213 is communicatively coupled to a plurality of multi-core processors 2205-2206 over high-speed links 2240-2243 (e.g., buses, point-to-point interconnects, etc.). In one embodiment, high-speed links 2240-2243 support a communication throughput of 4 GB/s, 30 GB/s, 80 GB/s or higher. Various interconnect protocols may be used including, but not limited to, PCIe 4.0 or 5.0 and NVLink 2.0.

In addition, and in one embodiment, two or more of GPUs 2210-2213 are interconnected over high-speed links 2229-2230, which may be implemented using same or different protocols/links than those used for high-speed links 2240-2243. Similarly, two or more of multi-core processors 2205-2206 may be connected over high speed link 2228 which may be symmetric multi-processor (SMP) buses operating at 20 GB/s, 30 GB/s, 120 GB/s or higher. Alternatively, all communication between various system components shown in FIG. 22A may be accomplished using same protocols/links (e.g., over a common interconnection fabric).

In one embodiment, each multi-core processor 2205-2206 is communicatively coupled to a processor memory 2201-2202, via memory interconnects 2226-2227, respectively, and each GPU 2210-2213 is communicatively coupled to GPU memory 2220-2223 over GPU memory interconnects 2250-2253, respectively. Memory interconnects 2226-2227 and 2250-2253 may utilize same or different memory access technologies. By way of example, and not limitation, processor memories 2201-2202 and GPU memories 2220-2223 may be volatile memories such as dynamic random access memories (DRAMs) (including stacked DRAMs), Graphics DDR SDRAM (GDDR) (e.g., GDDR5, GDDR6), or High Bandwidth Memory (HBM) and/or may be non-volatile memories such as 3D XPoint or Nano-Ram. In one embodiment, some portion of processor memories 2201-2202 may be volatile memory and another portion may be non-volatile memory (e.g., using a two-level memory (2LM) hierarchy).

As described herein, although various processors 2205-2206 and GPUs 2210-2213 may be physically coupled to a particular memory 2201-2202, 2220-2223, respectively, a unified memory architecture may be implemented in which a same virtual system address space (also referred to as “effective address” space) is distributed among various physical memories. For example, processor memories 2201-2202 may each comprise 64 GB of system memory address space and GPU memories 2220-2223 may each comprise 32 GB of system memory address space (resulting in a total of 256 GB addressable memory in this example).

FIG. 22B illustrates additional details for an interconnection between a multi-core processor 2207 and a graphics acceleration module 2246 in accordance with one exemplary embodiment. Graphics acceleration module 2246 may include one or more GPU chips integrated on a line card which is coupled to processor 2207 via high-speed link 2240. Alternatively, graphics acceleration module 2246 may be integrated on a same package or chip as processor 2207.

In at least one embodiment, illustrated processor 2207 includes a plurality of cores 2260A-2260D, each with a translation lookaside buffer 2261A-2261D and one or more caches 2262A-2262D. In at least one embodiment, cores 2260A-2260D may include various other components for executing instructions and processing data which are not illustrated. Caches 2262A-2262D may comprise level 1 (L1) and level 2 (L2) caches. In addition, one or more shared caches 2256 may be included in caches 2262A-2262D and shared by sets of cores 2260A-2260D. For example, one embodiment of processor 2207 includes 24 cores, each with its own L1 cache, twelve shared L2 caches, and twelve shared L3 caches. In this embodiment, one or more L2 and L3 caches are shared by two adjacent cores. Processor 2207 and graphics acceleration module 2246 connect with system memory 2214, which may include processor memories 2201-2202 of FIG. 22A.

Coherency is maintained for data and instructions stored in various caches 2262A-2262D, 2256 and system memory 2214 via inter-core communication over a coherence bus 2264. For example, each cache may have cache coherency logic/circuitry associated therewith to communicate to over coherence bus 2264 in response to detected reads or writes to particular cache lines. In one implementation, a cache snooping protocol is implemented over coherence bus 2264 to snoop cache accesses.

In one embodiment, a proxy circuit 2225 communicatively couples graphics acceleration module 2246 to coherence bus 2264, allowing graphics acceleration module 2246 to participate in a cache coherence protocol as a peer of cores 2260A-2260D. In particular, an interface 2235 provides connectivity to proxy circuit 2225 over high-speed link 2240 (e.g., a PCIe bus, NVLink, etc.) and an interface 2237 connects graphics acceleration module 2246 to link 2240.

In one implementation, an accelerator integration circuit 2236 provides cache management, memory access, context management, and interrupt management services on behalf of a plurality of graphics processing engines 2231, 2232, N of graphics acceleration module 2246. Graphics processing engines 2231, 2232, N may each comprise a separate graphics processing unit (GPU). Alternatively, graphics processing engines 2231, 2232, N may comprise different types of graphics processing engines within a GPU such as graphics execution units, media processing engines (e.g., video encoders/decoders), samplers, and blit engines. In at least one embodiment, graphics acceleration module 2246 may be a GPU with a plurality of graphics processing engines 2231-2232, N or graphics processing engines 2231-2232, N may be individual GPUs integrated on a common package, line card, or chip.

In one embodiment, accelerator integration circuit 2236 includes a memory management unit (MMU) 2239 for performing various memory management functions such as virtual-to-physical memory translations (also referred to as effective-to-real memory translations) and memory access protocols for accessing system memory 2214. MMU 2239 may also include a translation lookaside buffer (TLB) (not shown) for caching virtual/effective to physical/real address translations. In one implementation, a cache 2238 stores commands and data for efficient access by graphics processing engines 2231-2232, N. In one embodiment, data stored in cache 2238 and graphics memories 2233-2234, M is kept coherent with core caches 2262A-2262D, 2256 and system memory 2214. As mentioned, this may be accomplished via proxy circuit 2225 on behalf of cache 2238 and memories 2233-2234, M (e.g., sending updates to cache 2238 related to modifications/accesses of cache lines on processor caches 2262A-2262D, 2256 and receiving updates from cache 2238).

A set of registers 2245 store context data for threads executed by graphics processing engines 2231-2232, N and a context management circuit 2248 manages thread contexts. For example, context management circuit 2248 may perform save and restore operations to save and restore contexts of various threads during contexts switches (e.g., where a first thread is saved and a second thread is stored so that a second thread can be execute by a graphics processing engine). For example, on a context switch, context management circuit 2248 may store current register values to a designated region in memory (e.g., identified by a context pointer). It may then restore register values when returning to a context. In one embodiment, an interrupt management circuit 2247 receives and processes interrupts received from system devices.

In one implementation, virtual/effective addresses from a graphics processing engine 2231 are translated to real/physical addresses in system memory 2214 by MMU 2239. One embodiment of accelerator integration circuit 2236 supports multiple (e.g., 4, 8, 16) graphics accelerator modules 2246 and/or other accelerator devices. Graphics accelerator module 2246 may be dedicated to a single application executed on processor 2207 or may be shared between multiple applications. In one embodiment, a virtualized graphics execution environment is presented in which resources of graphics processing engines 2231-2232, N are shared with multiple applications or virtual machines (VMs). In at least one embodiment, resources may be subdivided into “slices” which are allocated to different VMs and/or applications based on processing requirements and priorities associated with VMs and/or applications.

In at least one embodiment, accelerator integration circuit 2236 performs as a bridge to a system for graphics acceleration module 2246 and provides address translation and system memory cache services. In addition, accelerator integration circuit 2236 may provide virtualization facilities for a host processor to manage virtualization of graphics processing engines 2231-2232, interrupts, and memory management.

Because hardware resources of graphics processing engines 2231-2232, N are mapped explicitly to a real address space seen by host processor 2207, any host processor can address these resources directly using an effective address value. One function of accelerator integration circuit 2236, in one embodiment, is physical separation of graphics processing engines 2231-2232, N so that they appear to a system as independent units.

In at least one embodiment, one or more graphics memories 2233-2234, Mare coupled to each of graphics processing engines 2231-2232, N, respectively. Graphics memories 2233-2234, M store instructions and data being processed by each of graphics processing engines 2231-2232, N. Graphics memories 2233-2234, M may be volatile memories such as DRAMs (including stacked DRAMs), GDDR memory (e.g., GDDR5, GDDR6), or HBM, and/or may be non-volatile memories such as 3D XPoint or Nano-Ram.

In one embodiment, to reduce data traffic over link 2240, biasing techniques are used to ensure that data stored in graphics memories 2233-2234, M is data which will be used most frequently by graphics processing engines 2231-2232, N and preferably not used by cores 2260A-2260D (at least not frequently). Similarly, a biasing mechanism attempts to keep data needed by cores (and preferably not graphics processing engines 2231-2232, N) within caches 2262A-2262D, 2256 of cores and system memory 2214.

FIG. 22C illustrates another exemplary embodiment in which accelerator integration circuit 2236 is integrated within processor 2207. In this embodiment, graphics processing engines 2231-2232, N communicate directly over high-speed link 2240 to accelerator integration circuit 2236 via interface 2237 and interface 2235 (which, again, may be utilize any form of bus or interface protocol). Accelerator integration circuit 2236 may perform same operations as those described with respect to FIG. 22B, but potentially at a higher throughput given its close proximity to coherence bus 2264 and caches 2262A-2262D, 2256. One embodiment supports different programming models including a dedicated-process programming model (no graphics acceleration module virtualization) and shared programming models (with virtualization), which may include programming models which are controlled by accelerator integration circuit 2236 and programming models which are controlled by graphics acceleration module 2246.

In at least one embodiment, graphics processing engines 2231-2232, N are dedicated to a single application or process under a single operating system. In at least one embodiment, a single application can funnel other application requests to graphics processing engines 2231-2232, N, providing virtualization within a VM/partition.

In at least one embodiment, graphics processing engines 2231-2232, N, may be shared by multiple VM/application partitions. In at least one embodiment, shared models may use a system hypervisor to virtualize graphics processing engines 2231-2232, N to allow access by each operating system. For single-partition systems without a hypervisor, graphics processing engines 2231-2232, N are owned by an operating system. In at least one embodiment, an operating system can virtualize graphics processing engines 2231-2232, N to provide access to each process or application.

In at least one embodiment, graphics acceleration module 2246 or an individual graphics processing engine 2231-2232, N selects a process element using a process handle. In one embodiment, process elements are stored in system memory 2214 and are addressable using an effective address to real address translation techniques described herein. In at least one embodiment, a process handle may be an implementation-specific value provided to a host process when registering its context with graphics processing engine 2231-2232, N (that is, calling system software to add a process element to a process element linked list). In at least one embodiment, a lower 16-bits of a process handle may be an offset of the process element within a process element linked list.

FIG. 22D illustrates an exemplary accelerator integration slice 2290. As used herein, a “slice” comprises a specified portion of processing resources of accelerator integration circuit 2236. Application effective address space 2282 within system memory 2214 stores process elements 2283. In one embodiment, process elements 2283 are stored in response to GPU invocations 2281 from applications 2280 executed on processor 2207. A process element 2283 contains process state for corresponding application 2280. A work descriptor (WD) 2284 contained in process element 2283 can be a single job requested by an application or may contain a pointer to a queue of jobs. In at least one embodiment, WD 2284 is a pointer to a job request queue in an application's address space 2282.

Graphics acceleration module 2246 and/or individual graphics processing engines 2231-2232, N can be shared by all or a subset of processes in a system. In at least one embodiment, an infrastructure for setting up process state and sending a WD 2284 to a graphics acceleration module 2246 to start a job in a virtualized environment may be included.

In at least one embodiment, a dedicated-process programming model is implementation-specific. In this model, a single process owns graphics acceleration module 2246 or an individual graphics processing engine 2231. Because graphics acceleration module 2246 is owned by a single process, a hypervisor initializes accelerator integration circuit 2236 for an owning partition and an operating system initializes accelerator integration circuit 2236 for an owning process when graphics acceleration module 2246 is assigned.

In operation, a WD fetch unit 2291 in accelerator integration slice 2290 fetches next WD 2284 which includes an indication of work to be done by one or more graphics processing engines of graphics acceleration module 2246. Data from WD 2284 may be stored in registers 2245 and used by MMU 2239, interrupt management circuit 2247 and/or context management circuit 2248 as illustrated. For example, one embodiment of MMU 2239 includes segment/page walk circuitry for accessing segment/page tables 2286 within OS virtual address space 2285. Interrupt management circuit 2247 may process interrupt events 2292 received from graphics acceleration module 2246. When performing graphics operations, an effective address 2293 generated by a graphics processing engine 2231-2232, N is translated to a real address by MMU 2239.

In one embodiment, a same set of registers 2245 are duplicated for each graphics processing engine 2231-2232, N and/or graphics acceleration module 2246 and may be initialized by a hypervisor or operating system. Each of these duplicated registers may be included in an accelerator integration slice 2290. Exemplary registers that may be initialized by a hypervisor are shown in Table 1.

TABLE 1
Hypervisor Initialized Registers
1 Slice Control Register
2 Real Address (RA) Scheduled Processes Area Pointer
3 Authority Mask Override Register
4 Interrupt Vector Table Entry Offset
5 Interrupt Vector Table Entry Limit
6 State Register
7 Logical Partition ID
8 Real address (RA) Hypervisor Accelerator Utilization Record Pointer
9 Storage Description Register

Exemplary registers that may be initialized by an operating system are shown in Table 2.

TABLE 2
Operating System Initialized Registers
1 Process and Thread Identification
2 Effective Address (EA) Context Save/Restore Pointer
3 Virtual Address (VA) Accelerator Utilization Record Pointer
4 Virtual Address (VA) Storage Segment Table Pointer
5 Authority Mask
6 Work descriptor

In one embodiment, each WD 2284 is specific to a particular graphics acceleration module 2246 and/or graphics processing engines 2231-2232, N. It contains all information required by a graphics processing engine 2231-2232, N to do work or it can be a pointer to a memory location where an application has set up a command queue of work to be completed.

FIG. 22E illustrates additional details for one exemplary embodiment of a shared model. This embodiment includes a hypervisor real address space 2298 in which a process element list 2299 is stored. Hypervisor real address space 2298 is accessible via a hypervisor 2296 which virtualizes graphics acceleration module engines for operating system 2295.

In at least one embodiment, shared programming models allow for all or a subset of processes from all or a subset of partitions in a system to use a graphics acceleration module 2246. There are two programming models where graphics acceleration module 2246 is shared by multiple processes and partitions: time-sliced shared and graphics directed shared.

In this model, system hypervisor 2296 owns graphics acceleration module 2246 and makes its function available to all operating systems 2295. For a graphics acceleration module 2246 to support virtualization by system hypervisor 2296, graphics acceleration module 2246 may adhere to the following: 1) An application's job request must be autonomous (that is, state does not need to be maintained between jobs), or graphics acceleration module 2246 must provide a context save and restore mechanism. 2) An application's job request is guaranteed by graphics acceleration module 2246 to complete in a specified amount of time, including any translation faults, or graphics acceleration module 2246 provides an ability to preempt processing of a job. 3) Graphics acceleration module 2246 must be guaranteed fairness between processes when operating in a directed shared programming model.

In at least one embodiment, application 2280 is required to make an operating system 2295 system call with a graphics acceleration module 2246 type, a work descriptor (WD), an authority mask register (AMR) value, and a context save/restore area pointer (CSRP). In at least one embodiment, graphics acceleration module 2246 type describes a targeted acceleration function for a system call. In at least one embodiment, graphics acceleration module 2246 type may be a system-specific value. In at least one embodiment, WD is formatted specifically for graphics acceleration module 2246 and can be in a form of a graphics acceleration module 2246 command, an effective address pointer to a user-defined structure, an effective address pointer to a queue of commands, or any other data structure to describe work to be done by graphics acceleration module 2246. In one embodiment, an AMR value is an AMR state to use for a current process. In at least one embodiment, a value passed to an operating system is similar to an application setting an AMR. If accelerator integration circuit 2236 and graphics acceleration module 2246 implementations do not support a User Authority Mask Override Register (UAMOR), an operating system may apply a current UAMOR value to an AMR value before passing an AMR in a hypervisor call. Hypervisor 2296 may optionally apply a current Authority Mask Override Register (AMOR) value before placing an AMR into process element 2283. In at least one embodiment, CSRP is one of registers 2245 containing an effective address of an area in an application's address space 2282 for graphics acceleration module 2246 to save and restore context state. This pointer is optional if no state is required to be saved between jobs or when a job is preempted. In at least one embodiment, context save/restore area may be pinned system memory.

Upon receiving a system call, operating system 2295 may verify that application 2280 has registered and been given authority to use graphics acceleration module 2246. Operating system 2295 then calls hypervisor 2296 with information shown in Table 3.

TABLE 3
OS to Hypervisor Call Parameters
1 A work descriptor (WD)
2 An Authority Mask Register (AMR) value (potentially masked)
3 An effective address (EA) Context Save/Restore Area Pointer (CSRP)
4 A process ID (PID) and optional thread ID (TID)
5 A virtual address (VA) accelerator utilization record pointer (AURP)
6 Virtual address of storage segment table pointer (SSTP)
7 A logical interrupt service number (LISN)

Upon receiving a hypervisor call, hypervisor 2296 verifies that operating system 2295 has registered and been given authority to use graphics acceleration module 2246. Hypervisor 2296 then puts process element 2283 into a process element linked list fora corresponding graphics acceleration module 2246 type. A process element may include information shown in Table 4.

TABLE 4
Process Element Information
1  A work descriptor (WD)
2  An Authority Mask Register (AMR) value (potentially masked).
3  An effective address (EA) Context Save/Restore Area Pointer (CSRP)
4  A process ID (PID) and optional thread ID (TID)
5  A virtual address (VA) accelerator utilization record pointer (AURP)
6  Virtual address of storage segment table pointer (SSTP)
7  A logical interrupt service number (LISN)
8  Interrupt vector table, derived from hypervisor call parameters
9  A state register (SR) value
10 A logical partition ID (LPID)
11 A real address (RA) hypervisor accelerator utilization record pointer
12 Storage Descriptor Register (SDR)

In at least one embodiment, hypervisor initializes a plurality of accelerator integration slice 2290 registers 2245.

As illustrated in FIG. 22F, in at least one embodiment, a unified memory is used, addressable via a common virtual memory address space used to access physical processor memories 2201-2202 and GPU memories 2220-2223. In this implementation, operations executed on GPUs 2210-2213 utilize a same virtual/effective memory address space to access processor memories 2201-2202 and vice versa, thereby simplifying programmability. In one embodiment, a first portion of a virtual/effective address space is allocated to processor memory 2201, a second portion to second processor memory 2202, a third portion to GPU memory 2220, and so on. In at least one embodiment, an entire virtual/effective memory space (sometimes referred to as an effective address space) is thereby distributed across each of processor memories 2201-2202 and GPU memories 2220-2223, allowing any processor or GPU to access any physical memory with a virtual address mapped to that memory.

In one embodiment, bias/coherence management circuitry 2294A-2294E within one or more of MMUs 2239A-2239E ensures cache coherence between caches of one or more host processors (e.g., 2205) and GPUs 2210-2213 and implements biasing techniques indicating physical memories in which certain types of data should be stored. While multiple instances of bias/coherence management circuitry 2294A-2294E are illustrated in FIG. 22F, bias/coherence circuitry may be implemented within an MMU of one or more host processors 2205 and/or within accelerator integration circuit 2236.

One embodiment allows GPU-attached memory 2220-2223 to be mapped as part of system memory, and accessed using shared virtual memory (SVM) technology, but without suffering performance drawbacks associated with full system cache coherence. In at least one embodiment, an ability for GPU-attached memory 2220-2223 to be accessed as system memory without onerous cache coherence overhead provides a beneficial operating environment for GPU offload. This arrangement allows host processor 2205 software to setup operands and access computation results, without overhead of tradition I/O DMA data copies. Such traditional copies involve driver calls, interrupts and memory mapped I/O (MMIO) accesses that are all inefficient relative to simple memory accesses. In at least one embodiment, an ability to access GPU attached memory 2220-2223 without cache coherence overheads can be critical to execution time of an offloaded computation. In cases with substantial streaming write memory traffic, for example, cache coherence overhead can significantly reduce an effective write bandwidth seen by a GPU 2210-2213. In at least one embodiment, efficiency of operand setup, efficiency of results access, and efficiency of GPU computation may play a role in determining effectiveness of a GPU offload.

In at least one embodiment, selection of GPU bias and host processor bias is driven by a bias tracker data structure. A bias table may be used, for example, which may be a page-granular structure (i.e., controlled at a granularity of a memory page) that includes 1 or 2 bits per GPU-attached memory page. In at least one embodiment, a bias table may be implemented in a stolen memory range of one or more GPU-attached memories 2220-2223, with or without a bias cache in GPU 2210-2213 (e.g., to cache frequently/recently used entries of a bias table). Alternatively, an entire bias table may be maintained within a GPU.

In at least one embodiment, a bias table entry associated with each access to GPU-attached memory 2220-2223 is accessed prior to actual access to a GPU memory, causing the following operations. First, local requests from GPU 2210-2213 that find their page in GPU bias are forwarded directly to a corresponding GPU memory 2220-2223. Local requests from a GPU that find their page in host bias are forwarded to processor 2205 (e.g., over a high-speed link as discussed above). In one embodiment, requests from processor 2205 that find a requested page in host processor bias complete a request like a normal memory read. Alternatively, requests directed to a GPU-biased page may be forwarded to GPU 2210-2213. In at least one embodiment, a GPU may then transition a page to a host processor bias if it is not currently using a page. In at least one embodiment, bias state of a page can be changed either by a software-based mechanism, a hardware-assisted software-based mechanism, or, for a limited set of cases, a purely hardware-based mechanism.

One mechanism for changing bias state employs an API call (e.g. OpenCL), which, in turn, calls a GPU's device driver which, in turn, sends a message (or enqueues a command descriptor) to a GPU directing it to change a bias state and, for some transitions, perform a cache flushing operation in a host. In at least one embodiment, cache flushing operation is used for a transition from host processor 2205 bias to GPU bias, but is not for an opposite transition.

In one embodiment, cache coherency is maintained by temporarily rendering GPU-biased pages uncacheable by host processor 2205. To access these pages, processor 2205 may request access from GPU 2210 which may or may not grant access right away. Thus, to reduce communication between processor 2205 and GPU 2210 it is beneficial to ensure that GPU-biased pages are those which are required by a GPU but not host processor 2205 and vice versa.

Hardware structure(s) 1415 are used to perform one or more embodiments. Details regarding the hardware structure(x) 1415 are provided herein in conjunction with FIGS. 14A and/or 14B.

FIG. 23 illustrates exemplary integrated circuits and associated graphics processors that may be fabricated using one or more IP cores, according to various embodiments described herein. In addition to what is illustrated, other logic and circuits may be included in at least one embodiment, including additional graphics processors/cores, peripheral interface controllers, or general-purpose processor cores.

FIG. 23 is a block diagram illustrating an exemplary system on a chip integrated circuit 2300 that may be fabricated using one or more IP cores, according to at least one embodiment. In at least one embodiment, integrated circuit 2300 includes one or more application processor(s) 2305 (e.g., CPUs), at least one graphics processor 2310, and may additionally include an image processor 2315 and/or a video processor 2320, any of which may be a modular IP core. In at least one embodiment, integrated circuit 2300 includes peripheral or bus logic including a USB controller 2325, UART controller 2330, an SPI/SDIO controller 2335, and an I.sup.2S/I.sup.2C controller 2340. In at least one embodiment, integrated circuit 2300 can include a display device 2345 coupled to one or more of a high-definition multimedia interface (HDMI) controller 2350 and a mobile industry processor interface (MIPI) display interface 2355. In at least one embodiment, storage may be provided by a flash memory subsystem 2360 including flash memory and a flash memory controller. In at least one embodiment, memory interface may be provided via a memory controller 2365 for access to SDRAM or SRAM memory devices. In at least one embodiment, some integrated circuits additionally include an embedded security engine 2370.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in integrated circuit 2300 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIGS. 24A-24B illustrate exemplary integrated circuits and associated graphics processors that may be fabricated using one or more IP cores, according to various embodiments described herein. In addition to what is illustrated, other logic and circuits may be included in at least one embodiment, including additional graphics processors/cores, peripheral interface controllers, or general-purpose processor cores.

FIGS. 24A-24B are block diagrams illustrating exemplary graphics processors for use within an SoC, according to embodiments described herein. FIG. 24A illustrates an exemplary graphics processor 2410 of a system on a chip integrated circuit that may be fabricated using one or more IP cores, according to at least one embodiment. FIG. 24B illustrates an additional exemplary graphics processor 2440 of a system on a chip integrated circuit that may be fabricated using one or more IP cores, according to at least one embodiment. In at least one embodiment, graphics processor 2410 of FIG. 24A is a low power graphics processor core. In at least one embodiment, graphics processor 2440 of FIG. 24B is a higher performance graphics processor core. In at least one embodiment, each of graphics processors 2410, 2440 can be variants of graphics processor 2310 of FIG. 23.

In at least one embodiment, graphics processor 2410 includes a vertex processor 2405 and one or more fragment processor(s) 2415A-2415N (e.g., 2415A, 2415B, 2415C, 2415D, through 2415N-1, and 2415N). In at least one embodiment, graphics processor 2410 can execute different shader programs via separate logic, such that vertex processor 2405 is optimized to execute operations for vertex shader programs, while one or more fragment processor(s) 2415A-2415N execute fragment (e.g., pixel) shading operations for fragment or pixel shader programs. In at least one embodiment, vertex processor 2405 performs a vertex processing stage of a 3D graphics pipeline and generates primitives and vertex data. In at least one embodiment, fragment processor(s) 2415A-2415N use primitive and vertex data generated by vertex processor 2405 to produce a framebuffer that is displayed on a display device. In at least one embodiment, fragment processor(s) 2415A-2415N are optimized to execute fragment shader programs as provided for in an OpenGL API, which may be used to perform similar operations as a pixel shader program as provided for in a Direct 3D API.

In at least one embodiment, graphics processor 2410 additionally includes one or more memory management units (MMUs) 2420A-2420B, cache(s) 2425A-2425B, and circuit interconnect(s) 2430A-2430B. In at least one embodiment, one or more MNU(s) 2420A-2420B provide for virtual to physical address mapping for graphics processor 2410, including for vertex processor 2405 and/or fragment processor(s) 2415A-2415N, which may reference vertex or image/texture data stored in memory, in addition to vertex or image/texture data stored in one or more cache(s) 2425A-2425B. In at least one embodiment, one or more MNU(s) 2420A-2420B may be synchronized with other MMUs within system, including one or more MMUs associated with one or more application processor(s) 2305, image processors 2315, and/or video processors 2320 of FIG. 23, such that each processor 2305-2320 can participate in a shared or unified virtual memory system. In at least one embodiment, one or more circuit interconnect(s) 2430A-2430B enable graphics processor 2410 to interface with other IP cores within SoC, either via an internal bus of SoC or via a direct connection.

In at least one embodiment, graphics processor 2440 includes one or more MMU(s) 2420A-2420B, caches 2425A-2425B, and circuit interconnects 2430A-2430B of graphics processor 2410 of FIG. 24A. In at least one embodiment, graphics processor 2440 includes one or more shader core(s) 2455A-2455N (e.g., 2455A, 2455B, 2455C, 2455D, 2455E, 2455F, through 2455N-1, and 2455N), which provides for a unified shader core architecture in which a single core or type or core can execute all types of programmable shader code, including shader program code to implement vertex shaders, fragment shaders, and/or compute shaders. In at least one embodiment, a number of shader cores can vary. In at least one embodiment, graphics processor 2440 includes an inter-core task manager 2445, which acts as a thread dispatcher to dispatch execution threads to one or more shader cores 2455A-2455N and a tiling unit 2458 to accelerate tiling operations for tile-based rendering, in which rendering operations for a scene are subdivided in image space, for example to exploit local spatial coherence within a scene or to optimize use of internal caches.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in integrated circuit 24A and/or 24B for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIGS. 25A-25B illustrate additional exemplary graphics processor logic according to embodiments described herein. FIG. 25A illustrates a graphics core 2500 that may be included within graphics processor 2310 of FIG. 23, in at least one embodiment, and may be a unified shader core 2455A-2455N as in FIG. 24B in at least one embodiment. FIG. 25B illustrates a highly-parallel general-purpose graphics processing unit 2530 suitable for deployment on a multi-chip module in at least one embodiment.

In at least one embodiment, graphics core 2500 includes a shared instruction cache 2502, a texture unit 2518, and a cache/shared memory 2520 that are common to execution resources within graphics core 2500. In at least one embodiment, graphics core 2500 can include multiple slices 2501A-2501N or partition for each core, and a graphics processor can include multiple instances of graphics core 2500. Slices 2501A-2501N can include support logic including a local instruction cache 2504A-2504N, a thread scheduler 2506A-2506N, a thread dispatcher 2508A-2508N, and a set of registers 2510A-2510N. In at least one embodiment, slices 2501A-2501N can include a set of additional function units (AFUs 2512A-2512N), floating-point units (FPU 2514A-2514N), integer arithmetic logic units (ALUs 2516-2516N), address computational units (ACU 2513A-2513N), double-precision floating-point units (DPFPU 2515A-2515N), and matrix processing units (MPU 2517A-2517N).

In at least one embodiment, FPUs 2514A-2514N can perform single-precision (32-bit) and half-precision (16-bit) floating point operations, while DPFPUs 2515A-2515N perform double precision (64-bit) floating point operations. In at least one embodiment, ALUs 2516A-2516N can perform variable precision integer operations at 8-bit, 16-bit, and 32-bit precision, and can be configured for mixed precision operations. In at least one embodiment, MPUs 2517A-2517N can also be configured for mixed precision matrix operations, including half-precision floating point and 8-bit integer operations. In at least one embodiment, MPUs 2517-2517N can perform a variety of matrix operations to accelerate machine learning application frameworks, including enabling support for accelerated general matrix to matrix multiplication (GEMM). In at least one embodiment, AFUs 2512A-2512N can perform additional logic operations not supported by floating-point or integer units, including trigonometric operations (e.g., Sine, Cosine, etc.).

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in graphics core 2500 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 25B illustrates a general-purpose processing unit (GPGPU) 2530 that can be configured to enable highly-parallel compute operations to be performed by an array of graphics processing units, in at least one embodiment. In at least one embodiment, GPGPU 2530 can be linked directly to other instances of GPGPU 2530 to create a multi-GPU cluster to improve training speed for deep neural networks. In at least one embodiment, GPGPU 2530 includes a host interface 2532 to enable a connection with a host processor. In at least one embodiment, host interface 2532 is a PCI Express interface. In at least one embodiment, host interjace 2532 can be a vendor specific communications interface or communications fabric. In at least one embodiment, GPGPU 2530 receives commands from a host processor and uses a global scheduler 2534 to distribute execution threads associated with those commands to a set of compute clusters 2536A-2536H. In at least one embodiment, compute clusters 2536A-2536H share a cache memory 2538. In at least one embodiment, cache memory 2538 can serve as a higher-level cache for cache memories within compute clusters 2536A-2536H.

In at least one embodiment, GPGPU 2530 includes memory 2544A-2544B coupled with compute clusters 2536A-2536H via a set of memory controllers 2542A-2542B. In at least one embodiment, memory 2544A-2544B can include various types of memory devices including dynamic random access memory (DRAM) or graphics random access memory, such as synchronous graphics random access memory (SGRAM), including graphics double data rate (GDDR) memory.

In at least one embodiment, compute clusters 2536A-2536H each include a set of graphics cores, such as graphics core 2500 of FIG. 25A, which can include multiple types of integer and floating point logic units that can perform computational operations at a range of precisions including suited for machine learning computations. For example, in at least one embodiment, at least a subset of floating point units in each of compute clusters 2536A-2536H can be configured to perform 16-bit or 32-bit floating point operations, while a different subset of floating point units can be configured to perform 64-bit floating point operations.

In at least one embodiment, multiple instances of GPGPU 2530 can be configured to operate as a compute cluster. In at least one embodiment, communication used by compute clusters 2536A-2536H for synchronization and data exchange varies across embodiments. In at least one embodiment, multiple instances of GPGPU 2530 communicate over host interface 2532. In at least one embodiment, GPGPU 2530 includes an I/O hub 2539 that couples GPGPU 2530 with a GPU link 2540 that enables a direct connection to other instances of GPGPU 2530. In at least one embodiment, GPU link 2540 is coupled to a dedicated GPU-to-GPU bridge that enables communication and synchronization between multiple instances of GPGPU 2530. In at least one embodiment GPU link 2540 couples with a high speed interconnect to transmit and receive data to other GPGPUs or parallel processors. In at least one embodiment, multiple instances of GPGPU 2530 are located in separate data processing systems and communicate via a network device that is accessible via host interface 2532. In at least one embodiment GPU link 2540 can be configured to enable a connection to a host processor in addition to or as an alternative to host interface 2532.

In at least one embodiment, GPGPU 2530 can be configured to train neural networks. In at least one embodiment, GPGPU 2530 can be used within a inferencing platform. In at least one embodiment, in which GPGPU 2530 is used for inferencing, GPGPU may include fewer compute clusters 2536A-2536H relative to when GPGPU is used for training a neural network. In at least one embodiment, memory technology associated with memory 2544A-2544B may differ between inferencing and training configurations, with higher bandwidth memory technologies devoted to training configurations. In at least one embodiment, inferencing configuration of GPGPU 2530 can support inferencing specific instructions. For example, in at least one embodiment, an inferencing configuration can provide support for one or more 8-bit integer dot product instructions, which may be used during inferencing operations for deployed neural networks.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in GPGPU 2530 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 26 is a block diagram illustrating a computing system 2600 according to at least one embodiment. In at least one embodiment, computing system 2600 includes a processing subsystem 2601 having one or more processor(s) 2602 and a system memory 2604 communicating via an interconnection path that may include a memory hub 2605. In at least one embodiment, memory hub 2605 may be a separate component within a chipset component or may be integrated within one or more processor(s) 2602. In at least one embodiment, memory hub 2605 couples with an I/O subsystem 2611 via a communication link 2606. In at least one embodiment, I/O subsystem 2611 includes an I/O hub 2607 that can enable computing system 2600 to receive input from one or more input device(s) 2608. In at least one embodiment, I/O hub 2607 can enable a display controller, which may be included in one or more processor(s) 2602, to provide outputs to one or more display device(s) 2610A. In at least one embodiment, one or more display device(s) 2610A coupled with I/O hub 2607 can include a local, internal, or embedded display device.

In at least one embodiment, processing subsystem 2601 includes one or more parallel processor(s) 2612 coupled to memory hub 2605 via a bus or other communication link 2613. In at least one embodiment, communication link 2613 may be one of any number of standards based communication link technologies or protocols, such as, but not limited to PCI Express, or may be a vendor specific communications interface or communications fabric. In at least one embodiment, one or more parallel processor(s) 2612 form a computationally focused parallel or vector processing system that can include a large number of processing cores and/or processing clusters, such as a many integrated core (MIC) processor. In at least one embodiment, one or more parallel processor(s) 2612 form a graphics processing subsystem that can output pixels to one of one or more display device(s) 2610A coupled via I/O Hub 2607. In at least one embodiment, one or more parallel processor(s) 2612 can also include a display controller and display interface (not shown) to enable a direct connection to one or more display device(s) 2610B.

In at least one embodiment, a system storage unit 2614 can connect to I/O hub 2607 to provide a storage mechanism for computing system 2600. In at least one embodiment, an I/O switch 2616 can be used to provide an interface mechanism to enable connections between I/O hub 2607 and other components, such as a network adapter 2618 and/or wireless network adapter 2619 that may be integrated into platform, and various other devices that can be added via one or more add-in device(s) 2620. In at least one embodiment, network adapter 2618 can be an Ethernet adapter or another wired network adapter. In at least one embodiment, wireless network adapter 2619 can include one or more of a Wi-Fi, Bluetooth, near field communication (NFC), or other network device that includes one or more wireless radios.

In at least one embodiment, computing system 2600 can include other components not explicitly shown, including USB or other port connections, optical storage drives, video capture devices, and like, may also be connected to I/O hub 2607. In at least one embodiment, communication paths interconnecting various components in FIG. 26 may be implemented using any suitable protocols, such as PCI (Peripheral Component Interconnect) based protocols (e.g., PCI-Express), or other bus or point-to-point communication interfaces and/or protocol(s), such as NV-Link high-speed interconnect, or interconnect protocols.

In at least one embodiment, one or more parallel processor(s) 2612 incorporate circuitry optimized for graphics and video processing, including, for example, video output circuitry, and constitutes a graphics processing unit (GPU). In at least one embodiment, one or more parallel processor(s) 2612 incorporate circuitry optimized for general purpose processing. In at least embodiment, components of computing system 2600 may be integrated with one or more other system elements on a single integrated circuit. For example, in at least one embodiment, one or more parallel processor(s) 2612, memory hub 2605, processor(s) 2602, and I/O hub 2607 can be integrated into a system on chip (SoC) integrated circuit. In at least one embodiment, components of computing system 2600 can be integrated into a single package to form a system in package (SIP) configuration. In at least one embodiment, at least a portion of components of computing system 2600 can be integrated into a multi-chip module (MCM), which can be interconnected with other multi-chip modules into a modular computing system.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in system FIG. 2600 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a computer system as described above.

Processors

FIG. 27A illustrates a parallel processor 2700 according to at least on embodiment. In at least one embodiment, various components of parallel processor 2700 may be implemented using one or more integrated circuit devices, such as programmable processors, application specific integrated circuits (ASICs), or field programmable gate arrays (FPGA). In at least one embodiment, illustrated parallel processor 2700 is a variant of one or more parallel processor(s) 2612 shown in FIG. 26 according to an exemplary embodiment.

In at least one embodiment, parallel processor 2700 includes a parallel processing unit 2702. In at least one embodiment, parallel processing unit 2702 includes an I/O unit 2704 that enables communication with other devices, including other instances of parallel processing unit 2702. In at least one embodiment, I/O unit 2704 may be directly connected to other devices. In at least one embodiment, I/O unit 2704 connects with other devices via use of a hub or switch interface, such as memory hub 2605. In at least one embodiment, connections between memory hub 2605 and I/O unit 2704 form a communication link 2613. In at least one embodiment, I/O unit 2704 connects with a host interface 2706 and a memory crossbar 2716, where host interface 2706 receives commands directed to performing processing operations and memory crossbar 2716 receives commands directed to performing memory operations.

In at least one embodiment, when host interface 2706 receives a command buffer via I/O unit 2704, host interface 2706 can direct work operations to perform those commands to a front end 2708. In at least one embodiment, front end 2708 couples with a scheduler 2710, which is configured to distribute commands or other work items to a processing cluster array 2712. In at least one embodiment, scheduler 2710 ensures that processing cluster array 2712 is properly configured and in a valid state before tasks are distributed to processing cluster array 2712 of processing cluster array 2712. In at least one embodiment, scheduler 2710 is implemented via firmware logic executing on a microcontroller. In at least one embodiment, microcontroller implemented scheduler 2710 is configurable to perform complex scheduling and work distribution operations at coarse and fine granularity, enabling rapid preemption and context switching of threads executing on processing array 2712. In at least one embodiment, host software can prove workloads for scheduling on processing array 2712 via one of multiple graphics processing doorbells. In at least one embodiment, workloads can then be automatically distributed across processing array 2712 by scheduler 2710 logic within a microcontroller including scheduler 2710.

In at least one embodiment, processing cluster array 2712 can include up to “N” processing clusters (e.g., cluster 2714A, cluster 2714B, through cluster 2714N). In at least one embodiment, each cluster 2714A-2714N of processing cluster array 2712 can execute a large number of concurrent threads. In at least one embodiment, scheduler 2710 can allocate work to clusters 2714A-2714N of processing cluster array 2712 using various scheduling and/or work distribution algorithms, which may vary depending on workload arising for each type of program or computation. In at least one embodiment, scheduling can be handled dynamically by scheduler 2710, or can be assisted in part by compiler logic during compilation of program logic configured for execution by processing cluster array 2712. In at least one embodiment, different clusters 2714A-2714N of processing cluster array 2712 can be allocated for processing different types of programs or for performing different types of computations.

In at least one embodiment, processing cluster array 2712 can be configured to perform various types of parallel processing operations. In at least one embodiment, processing cluster array 2712 is configured to perform general-purpose parallel compute operations. For example, in at least one embodiment, processing cluster array 2712 can include logic to execute processing tasks including filtering of video and/or audio data, performing modeling operations, including physics operations, and performing data transformations.

In at least one embodiment, processing cluster array 2712 is configured to perform parallel graphics processing operations. In at least one embodiment, processing cluster array 2712 can include additional logic to support execution of such graphics processing operations, including, but not limited to texture sampling logic to perform texture operations, as well as tessellation logic and other vertex processing logic. In at least one embodiment, processing cluster array 2712 can be configured to execute graphics processing related shader programs such as, but not limited to vertex shaders, tessellation shaders, geometry shaders, and pixel shaders. In at least one embodiment, parallel processing unit 2702 can transfer data from system memory via I/O unit 2704 for processing. In at least one embodiment, during processing, transferred data can be stored to on-chip memory (e.g., parallel processor memory 2722) during processing, then written back to system memory.

In at least one embodiment, when parallel processing unit 2702 is used to perform graphics processing, scheduler 2710 can be configured to divide a processing workload into approximately equal sized tasks, to better enable distribution of graphics processing operations to multiple clusters 2714A-2714N of processing cluster array 2712. In at least one embodiment, portions of processing cluster array 2712 can be configured to perform different types of processing. For example, in at least one embodiment, a first portion may be configured to perform vertex shading and topology generation, a second portion may be configured to perform tessellation and geometry shading, and a third portion may be configured to perform pixel shading or other screen space operations, to produce a rendered image for display. In at least one embodiment, intermediate data produced by one or more of clusters 2714A-2714N may be stored in buffers to allow intermediate data to be transmitted between clusters 2714A-2714N for further processing.

In at least one embodiment, processing cluster array 2712 can receive processing tasks to be executed via scheduler 2710, which receives commands defining processing tasks from front end 2708. In at least one embodiment, processing tasks can include indices of data to be processed, e.g., surface (patch) data, primitive data, vertex data, and/or pixel data, as well as state parameters and commands defining how data is to be processed (e.g., what program is to be executed). In at least one embodiment, scheduler 2710 may be configured to fetch indices corresponding to tasks or may receive indices from front end 2708. In at least one embodiment, front end 2708 can be configured to ensure processing cluster array 2712 is configured to a valid state before a workload specified by incoming command buffers (e.g., batch-buffers, push buffers, etc.) is initiated.

In at least one embodiment, each of one or more instances of parallel processing unit 2702 can couple with parallel processor memory 2722. In at least one embodiment, parallel processor memory 2722 can be accessed via memory crossbar 2716, which can receive memory requests from processing cluster array 2712 as well as I/O unit 2704. In at least one embodiment, memory crossbar 2716 can access parallel processor memory 2722 via a memory interface 2718. In at least one embodiment, memory interface 2718 can include multiple partition units (e.g., partition unit 2720A, partition unit 2720B, through partition unit 2720N) that can each couple to a portion (e.g., memory unit) of parallel processor memory 2722. In at least one embodiment, a number of partition units 2720A-2720N is configured to be equal to a number of memory units, such that a first partition unit 2720A has a corresponding first memory unit 2724A, a second partition unit 2720B has a corresponding memory unit 2724B, and an Nth partition unit 2720N has a corresponding Nth memory unit 2724N. In at least one embodiment, a number of partition units 2720A-2720N may not be equal to a number of memory devices.

In at least one embodiment, memory units 2724A-2724N can include various types of memory devices, including dynamic random access memory (DRAM) or graphics random access memory, such as synchronous graphics random access memory (SGRAM), including graphics double data rate (GDDR) memory. In at least one embodiment, memory units 2724A-2724N may also include 3D stacked memory, including but not limited to high bandwidth memory (HBM). In at least one embodiment, render targets, such as frame buffers or texture maps may be stored across memory units 2724A-2724N, allowing partition units 2720A-2720N to write portions of each render target in parallel to efficiently use available bandwidth of parallel processor memory 2722. In at least one embodiment, a local instance of parallel processor memory 2722 may be excluded in favor of a unified memory design that utilizes system memory in conjunction with local cache memory.

In at least one embodiment, any one of clusters 2714A-2714N of processing cluster array 2712 can process data that will be written to any of memory units 2724A-2724N within parallel processor memory 2722. In at least one embodiment, memory crossbar 2716 can be configured to transfer an output of each cluster 2714A-2714N to any partition unit 2720A-2720N or to another cluster 2714A-2714N, which can perform additional processing operations on an output. In at least one embodiment, each cluster 2714A-2714N can communicate with memory interface 2718 through memory crossbar 2716 to read from or write to various external memory devices. In at least one embodiment, memory crossbar 2716 has a connection to memory interface 2718 to communicate with I/O unit 2704, as well as a connection to a local instance of parallel processor memory 2722, enabling processing units within different processing clusters 2714A-2714N to communicate with system memory or other memory that is not local to parallel processing unit 2702. In at least one embodiment, memory crossbar 2716 can use virtual channels to separate traffic streams between clusters 2714A-2714N and partition units 2720A-2720N.

In at least one embodiment, multiple instances of parallel processing unit 2702 can be provided on a single add-in card, or multiple add-in cards can be interconnected. In at least one embodiment, different instances of parallel processing unit 2702 can be configured to inter-operate even if different instances have different numbers of processing cores, different amounts of local parallel processor memory, and/or other configuration differences. For example, in at least one embodiment, some instances of parallel processing unit 2702 can include higher precision floating point units relative to other instances. In at least one embodiment, systems incorporating one or more instances of parallel processing unit 2702 or parallel processor 2700 can be implemented in a variety of configurations and form factors, including but not limited to desktop, laptop, or handheld personal computers, servers, workstations, game consoles, and/or embedded systems.

FIG. 27B is a block diagram of a partition unit 2720 according to at least one embodiment. In at least one embodiment, partition unit 2720 is an instance of one of partition units 2720A-2720N of FIG. 27A. In at least one embodiment, partition unit 2720 includes an L2 cache 2721, a frame buffer interface 2725, and a ROP 2726 (raster operations unit). L2 cache 2721 is a read/write cache that is configured to perform load and store operations received from memory crossbar 2716 and ROP 2726. In at least one embodiment, read misses and urgent write-back requests are output by L2 cache 2721 to frame buffer interface 2725 for processing. In at least one embodiment, updates can also be sent to a frame buffer via frame buffer interface 2725 for processing. In at least one embodiment, frame buffer interface 2725 interfaces with one of memory units in parallel processor memory, such as memory units 2724A-2724N of FIG. 27 (e.g., within parallel processor memory 2722).

In at least one embodiment, ROP 2726 is a processing unit that performs raster operations such as stencil, z test, blending, and like. In at least one embodiment, ROP 2726 then outputs processed graphics data that is stored in graphics memory. In at least one embodiment, ROP 2726 includes compression logic to compress depth or color data that is written to memory and decompress depth or color data that is read from memory. In at least one embodiment, compression logic can be lossless compression logic that makes use of one or more of multiple compression algorithms. The type of compression that is performed by ROP 2726 can vary based on statistical characteristics of data to be compressed. For example, in at least one embodiment, delta color compression is performed on depth and color data on a per-tile basis.

In at least one embodiment, ROP 2726 is included within each processing cluster (e.g., cluster 2714A-2714N of FIG. 27) instead of within partition unit 2720. In at least one embodiment, read and write requests for pixel data are transmitted over memory crossbar 2716 instead of pixel fragment data. In at least one embodiment, processed graphics data may be displayed on a display device, such as one of one or more display device(s) 2610 of FIG. 26, routed for further processing by processor(s) 2602, or routed for further processing by one of processing entities within parallel processor 2700 of FIG. 27A.

FIG. 27C is a block diagram of a processing cluster 2714 within a parallel processing unit according to at least one embodiment. In at least one embodiment, a processing cluster is an instance of one of processing clusters 2714A-2714N of FIG. 27. In at least one embodiment, processing cluster 2714 can be configured to execute many threads in parallel, where term “thread” refers to an instance of a particular program executing on a particular set of input data. In at least one embodiment, single-instruction, multiple-data (SIMD) instruction issue techniques are used to support parallel execution of a large number of threads without providing multiple independent instruction units. In at least one embodiment, single-instruction, multiple-thread (SIMT) techniques are used to support parallel execution of a large number of generally synchronized threads, using a common instruction unit configured to issue instructions to a set of processing engines within each one of processing clusters.

In at least one embodiment, operation of processing cluster 2714 can be controlled via a pipeline manager 2732 that distributes processing tasks to SIMT parallel processors. In at least one embodiment, pipeline manager 2732 receives instructions from scheduler 2710 of FIG. 27 and manages execution of those instructions via a graphics multiprocessor 2734 and/or a texture unit 2736. In at least one embodiment, graphics multiprocessor 2734 is an exemplary instance of a SIMT parallel processor. However, in at least one embodiment, various types of SIMT parallel processors of differing architectures may be included within processing cluster 2714. In at least one embodiment, one or more instances of graphics multiprocessor 2734 can be included within a processing cluster 2714. In at least one embodiment, graphics multiprocessor 2734 can process data and a data crossbar 2740 can be used to distribute processed data to one of multiple possible destinations, including other shader units. In at least one embodiment, pipeline manager 2732 can facilitate distribution of processed data by specifying destinations for processed data to be distributed vis data crossbar 2740.

In at least one embodiment, each graphics multiprocessor 2734 within processing cluster 2714 can include an identical set of functional execution logic (e.g., arithmetic logic units, load-store units, etc.). In at least one embodiment, functional execution logic can be configured in a pipelined manner in which new instructions can be issued before previous instructions are complete. In at least one embodiment, functional execution logic supports a variety of operations including integer and floating point arithmetic, comparison operations, Boolean operations, bit-shifting, and computation of various algebraic functions. In at least one embodiment, same functional-unit hardware can be leveraged to perform different operations and any combination of functional units may be present.

In at least one embodiment, instructions transmitted to processing cluster 2714 constitute a thread. In at least one embodiment, a set of threads executing across a set of parallel processing engines is a thread group. In at least one embodiment, thread group executes a program on different input data. In at least one embodiment, each thread within a thread group can be assigned to a different processing engine within a graphics multiprocessor 2734. In at least one embodiment, a thread group may include fewer threads than a number of processing engines within graphics multiprocessor 2734. In at least one embodiment, when a thread group includes fewer threads than a number of processing engines, one or more of processing engines may be idle during cycles in which that thread group is being processed. In at least one embodiment, a thread group may also include more threads than a number of processing engines within graphics multiprocessor 2734. In at least one embodiment, when a thread group includes more threads than number of processing engines within graphics multiprocessor 2734, processing can be performed over consecutive clock cycles. In at least one embodiment, multiple thread groups can be executed concurrently on a graphics multiprocessor 2734.

In at least one embodiment, graphics multiprocessor 2734 includes an internal cache memory to perform load and store operations. In at least one embodiment, graphics multiprocessor 2734 can forego an internal cache and use a cache memory (e.g., L1 cache 2748) within processing cluster 2714. In at least one embodiment, each graphics multiprocessor 2734 also has access to L2 caches within partition units (e.g., partition units 2720A-2720N of FIG. 27) that are shared among all processing clusters 2714 and may be used to transfer data between threads. In at least one embodiment, graphics multiprocessor 2734 may also access off-chip global memory, which can include one or more of local parallel processor memory and/or system memory. In at least one embodiment, any memory external to parallel processing unit 2702 may be used as global memory. In at least one embodiment, processing cluster 2714 includes multiple instances of graphics multiprocessor 2734 can share common instructions and data, which may be stored in L1 cache 2748.

In at least one embodiment, each processing cluster 2714 may include an MMU 2745 (memory management unit) that is configured to map virtual addresses into physical addresses. In at least one embodiment, one or more instances of MMU 2745 may reside within memory interface 2718 of FIG. 27. In at least one embodiment, MMU 2745 includes a set of page table entries (PTEs) used to map a virtual address to a physical address of a tile (talk more about tiling) and optionally a cache line index. In at least one embodiment, MMU 2745 may include address translation lookaside buffers (TLB) or caches that may reside within graphics multiprocessor 2734 or L1 cache or processing cluster 2714. In at least one embodiment, physical address is processed to distribute surface data access locality to allow efficient request interleaving among partition units. In at least one embodiment, cache line index may be used to determine whether a request for a cache line is a hit or miss.

In at least one embodiment, a processing cluster 2714 may be configured such that each graphics multiprocessor 2734 is coupled to a texture unit 2736 for performing texture mapping operations, e.g., determining texture sample positions, reading texture data, and filtering texture data. In at least one embodiment, texture data is read from an internal texture L1 cache (not shown) or from an L1 cache within graphics multiprocessor 2734 and is fetched from an L2 cache, local parallel processor memory, or system memory, as needed. In at least one embodiment, each graphics multiprocessor 2734 outputs processed tasks to data crossbar 2740 to provide processed task to another processing cluster 2714 for further processing or to store processed task in an L2 cache, local parallel processor memory, or system memory via memory crossbar 2716. In at least one embodiment, preROP 2742 (pre-raster operations unit) is configured to receive data from graphics multiprocessor 2734, direct data to ROP units, which may be located with partition units as described herein (e.g., partition units 2720A-2720N of FIG. 27). In at least one embodiment, PreROP 2742 unit can perform optimizations for color blending, organize pixel color data, and perform address translations.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in graphics processing cluster 2714 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 27D shows a graphics multiprocessor 2734 according to at least one embodiment. In at least one embodiment, graphics multiprocessor 2734 couples with pipeline manager 2732 of processing cluster 2714. In at least one embodiment, graphics multiprocessor 2734 has an execution pipeline including but not limited to an instruction cache 2752, an instruction unit 2754, an address mapping unit 2756, a register file 2758, one or more general purpose graphics processing unit (GPGPU) cores 2762, and one or more load/store units 2766. GPGPU cores 2762 and load/store units 2766 are coupled with cache memory 2772 and shared memory 2770 via a memory and cache interconnect 2768.

In at least one embodiment, instruction cache 2752 receives a stream of instructions to execute from pipeline manager 2732. In at least one embodiment, instructions are cached in instruction cache 2752 and dispatched for execution by instruction unit 2754. In at least one embodiment, instruction unit 2754 can dispatch instructions as thread groups (e.g., warps), with each thread of thread group assigned to a different execution unit within GPGPU core 2762. In at least one embodiment, an instruction can access any of a local, shared, or global address space by specifying an address within a unified address space. In at least one embodiment, address mapping unit 2756 can be used to translate addresses in a unified address space into a distinct memory address that can be accessed by load/store units 2766.

In at least one embodiment, register file 2758 provides a set of registers for functional units of graphics multiprocessor 2734. In at least one embodiment, register file 2758 provides temporary storage for operands connected to data paths of functional units (e.g., GPGPU cores 2762, load/store units 2766) of graphics multiprocessor 2734. In at least one embodiment, register file 2758 is divided between each of functional units such that each functional unit is allocated a dedicated portion of register file 2758. In at least one embodiment, register file 2758 is divided between different warps being executed by graphics multiprocessor 2734.

In at least one embodiment, GPGPU cores 2762 can each include floating point units (FPUs) and/or integer arithmetic logic units (ALUs) that are used to execute instructions of graphics multiprocessor 2734. GPGPU cores 2762 can be similar in architecture or can differ in architecture. In at least one embodiment, a first portion of GPGPU cores 2762 include a single precision FPU and an integer ALU while a second portion of GPGPU cores include a double precision FPU. In at least one embodiment, FPUs can implement IEEE 754-2008 standard for floating point arithmetic or enable variable precision floating point arithmetic. In at least one embodiment, graphics multiprocessor 2734 can additionally include one or more fixed function or special function units to perform specific functions such as copy rectangle or pixel blending operations. In at least one embodiment one or more of GPGPU cores can also include fixed or special function logic.

In at least one embodiment, GPGPU cores 2762 include SIMD logic capable of performing a single instruction on multiple sets of data. In at least one embodiment GPGPU cores 2762 can physically execute SIMD4, SIMD8, and SIMD16 instructions and logically execute SIMD1, SIMD2, and SIMD32 instructions. In at least one embodiment, SIMD instructions for GPGPU cores can be generated at compile time by a shader compiler or automatically generated when executing programs written and compiled for single program multiple data (SPMD) or SIMT architectures. In at least one embodiment, multiple threads of a program configured for an SIMT execution model can executed via a single SIMD instruction. For example, in at least one embodiment, eight SIMT threads that perform same or similar operations can be executed in parallel via a single SIMD8 logic unit.

In at least one embodiment, memory and cache interconnect 2768 is an interconnect network that connects each functional unit of graphics multiprocessor 2734 to register file 2758 and to shared memory 2770. In at least one embodiment, memory and cache interconnect 2768 is a crossbar interconnect that allows load/store unit 2766 to implement load and store operations between shared memory 2770 and register file 2758. In at least one embodiment, register file 2758 can operate at a same frequency as GPGPU cores 2762, thus data transfer between GPGPU cores 2762 and register file 2758 is very low latency. In at least one embodiment, shared memory 2770 can be used to enable communication between threads that execute on functional units within graphics multiprocessor 2734. In at least one embodiment, cache memory 2772 can be used as a data cache for example, to cache texture data communicated between functional units and texture unit 2736. In at least one embodiment, shared memory 2770 can also be used as a program managed cached. In at least one embodiment, threads executing on GPGPU cores 2762 can programmatically store data within shared memory in addition to automatically cached data that is stored within cache memory 2772.

In at least one embodiment, a parallel processor or GPGPU as described herein is communicatively coupled to host/processor cores to accelerate graphics operations, machine-learning operations, pattern analysis operations, and various general purpose GPU (GPGPU) functions. In at least one embodiment, GPU may be communicatively coupled to host processor/cores over a bus or other interconnect (e.g., a high speed interconnect such as PCIe or NVLink). In at least one embodiment, GPU may be integrated on same package or chip as cores and communicatively coupled to cores over an internal processor bus/interconnect (i.e., internal to package or chip). In at least one embodiment, regardless of manner in which GPU is connected, processor cores may allocate work to GPU in form of sequences of commands/instructions contained in a work descriptor. In at least one embodiment, GPU then uses dedicated circuitry/logic for efficiently processing these commands/instructions.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in graphics multiprocessor 2734 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 28 illustrates a multi-GPU computing system 2800, according to at least one embodiment. In at least one embodiment, multi-GPU computing system 2800 can include a processor 2802 coupled to multiple general purpose graphics processing units (GPGPUs) 2806A-D via a host interface switch 2804. In at least one embodiment, host interface switch 2804 is a PCI express switch device that couples processor 2802 to a PCI express bus over which processor 2802 can communicate with GPGPUs 2806A-D. GPGPUs 2806A-D can interconnect via a set of high-speed point to point GPU to GPU links 2816. In at least one embodiment, GPU to GPU links 2816 connect to each of GPGPUs 2806A-D via a dedicated GPU link. In at least one embodiment, P2P GPU links 2816 enable direct communication between each of GPGPUs 2806A-D without requiring communication over host interface bus 2804 to which processor 2802 is connected. In at least one embodiment, with GPU-to-GPU traffic directed to P2P GPU links 2816, host interface bus 2804 remains available for system memory access or to communicate with other instances of multi-GPU computing system 2800, for example, via one or more network devices. While in at least one embodiment GPGPUs 2806A-D connect to processor 2802 via host interface switch 2804, in at least one embodiment processor 2802 includes direct support for P2P GPU links 2816 and can connect directly to GPGPUs 2806A-D.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in multi-GPU computing system 2800 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a GPU as described above.

FIG. 29 is a block diagram of a graphics processor 2900, according to at least one embodiment. In at least one embodiment, graphics processor 2900 includes a ring interconnect 2902, a pipeline front-end 2904, a media engine 2937, and graphics cores 2980A-2980N. In at least one embodiment, ring interconnect 2902 couples graphics processor 2900 to other processing units, including other graphics processors or one or more general-purpose processor cores. In at least one embodiment, graphics processor 2900 is one of many processors integrated within a multi-core processing system.

In at least one embodiment, graphics processor 2900 receives batches of commands via ring interconnect 2902. In at least one embodiment, incoming commands are interpreted by a command streamer 2903 in pipeline front-end 2904. In at least one embodiment, graphics processor 2900 includes scalable execution logic to perform 3D geometry processing and media processing via graphics core(s) 2980A-2980N. In at least one embodiment, for 3D geometry processing commands, command streamer 2903 supplies commands to geometry pipeline 2936. In at least one embodiment, for at least some media processing commands, command streamer 2903 supplies commands to a video front end 2934, which couples with a media engine 2937. In at least one embodiment, media engine 2937 includes a Video Quality Engine (VQE) 2930 for video and image post-processing and a multi-format encode/decode (MFX) 2933 engine to provide hardware-accelerated media data encode and decode. In at least one embodiment, geometry pipeline 2936 and media engine 2937 each generate execution threads for thread execution resources provided by at least one graphics core 2980A.

In at least one embodiment, graphics processor 2900 includes scalable thread execution resources featuring modular cores 2980A-2980N (sometimes referred to as core slices), each having multiple sub-cores 2950A-550N, 2960A-2960N (sometimes referred to as core sub-slices). In at least one embodiment, graphics processor 2900 can have any number of graphics cores 2980A through 2980N. In at least one embodiment, graphics processor 2900 includes a graphics core 2980A having at least a first sub-core 2950A and a second sub-core 2960A. In at least one embodiment, graphics processor 2900 is a low power processor with a single sub-core (e.g., 2950A). In at least one embodiment, graphics processor 2900 includes multiple graphics cores 2980A-2980N, each including a set of first sub-cores 2950A-2950N and a set of second sub-cores 2960A-2960N. In at least one embodiment, each sub-core in first sub-cores 2950A-2950N includes at least a first set of execution units 2952A-2952N and media/texture samplers 2954A-2954N. In at least one embodiment, each sub-core in second sub-cores 2960A-2960N includes at least a second set of execution units 2962A-2962N and samplers 2964A-2964N. In at least one embodiment, each sub-core 2950A-2950N, 2960A-2960N shares a set of shared resources 2970A-2970N. In at least one embodiment, shared resources include shared cache memory and pixel operation logic.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, inference and/or training logic 1415 may be used in graphics processor 2900 for inferencing or predicting operations based, at least in part, on weight parameters calculated using neural network training operations, neural network functions and/or architectures, or neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a GPU as described above.

FIG. 30 is a block diagram illustrating micro-architecture for a processor 3000 that may include logic circuits to perform instructions, according to at least one embodiment. In at least one embodiment, processor 3000 may perform instructions, including x86 instructions, ARM instructions, specialized instructions for application-specific integrated circuits (ASICs), etc. In at least one embodiment, processor 3010 may include registers to store packed data, such as 64-bit wide MMX™ registers in microprocessors enabled with MMX technology from Intel Corporation of Santa Clara, Calif. In at least one embodiment, MMX registers, available in both integer and floating point forms, may operate with packed data elements that accompany single instruction, multiple data (“SIMD”) and streaming SIMD extensions (“SSE”) instructions. In at least one embodiment, 128-bit wide XMM registers relating to SSE2, SSE3, SSE4, AVX, or beyond (referred to generically as “SSEx”) technology may hold such packed data operands. In at least one embodiment, processors 3010 may perform instructions to accelerate machine learning or deep learning algorithms, training, or inferencing.

In at least one embodiment, processor 3000 includes an in-order front end (“front end”) 3001 to fetch instructions to be executed and prepare instructions to be used later in processor pipeline. In at least one embodiment, front end 3001 may include several units. In at least one embodiment, an instruction prefetcher 3026 fetches instructions from memory and feeds instructions to an instruction decoder 3028 which in turn decodes or interprets instructions. For example, in at least one embodiment, instruction decoder 3028 decodes a received instruction into one or more operations called “micro-instructions” or “micro-operations” (also called “micro ops” or “uops”) that machine may execute. In at least one embodiment, instruction decoder 3028 parses instruction into an opcode and corresponding data and control fields that may be used by micro-architecture to perform operations in accordance with at least one embodiment. In at least one embodiment, a trace cache 3030 may assemble decoded uops into program ordered sequences or traces in a uop queue 3034 for execution. In at least one embodiment, when trace cache 3030 encounters a complex instruction, a microcode ROM 3032 provides uops needed to complete operation.

In at least one embodiment, some instructions may be converted into a single micro-op, whereas others need several micro-ops to complete full operation. In at least one embodiment, if more than four micro-ops are needed to complete an instruction, instruction decoder 3028 may access microcode ROM 3032 to perform instruction. In at least one embodiment, an instruction may be decoded into a small number of micro-ops for processing at instruction decoder 3028. In at least one embodiment, an instruction may be stored within microcode ROM 3032 should a number of micro-ops be needed to accomplish operation. In at least one embodiment, trace cache 3030 refers to an entry point programmable logic array (“PLA”) to determine a correct micro-instruction pointer for reading microcode sequences to complete one or more instructions from microcode ROM 3032 in accordance with at least one embodiment. In at least one embodiment, after microcode ROM 3032 finishes sequencing micro-ops for an instruction, front end 3001 of machine may resume fetching micro-ops from trace cache 3030.

In at least one embodiment, out-of-order execution engine (“out of order engine”) 3003 may prepare instructions for execution. In at least one embodiment, out-of-order execution logic has a number of buffers to smooth out and re-order flow of instructions to optimize performance as they go down pipeline and get scheduled for execution. out-of-order execution engine 3003 includes, without limitation, an allocator/register renamer 3040, a memory uop queue 3042, an integer/floating point uop queue 3044, a memory scheduler 3046, a fast scheduler 3002, a slow/general floating point scheduler (“slow/general FP scheduler”) 3004, and a simple floating point scheduler (“simple FP scheduler”) 3006. In at least one embodiment, fast schedule 3002, slow/general floating point scheduler 3004, and simple floating point scheduler 3006 are also collectively referred to herein as “uop schedulers 3002, 3004, 3006.” allocator/register renamer 3040 allocates machine buffers and resources that each uop needs in order to execute. In at least one embodiment, allocator/register renamer 3040 renames logic registers onto entries in a register file. In at least one embodiment, allocator/register renamer 3040 also allocates an entry for each uop in one of two uop queues, memory uop queue 3042 for memory operations and integer/floating point uop queue 3044 for non-memory operations, in front of memory scheduler 3046 and uop schedulers 3002, 3004, 3006. In at least one embodiment, uop schedulers 3002, 3004, 3006, determine when a uop is ready to execute based on readiness of their dependent input register operand sources and availability of execution resources uops need to complete their operation. In at least one embodiment, fast scheduler 3002 of at least one embodiment may schedule on each half of main clock cycle while slow/general floating point scheduler 3004 and simple floating point scheduler 3006 may schedule once per main processor clock cycle. In at least one embodiment, uop schedulers 3002, 3004, 3006 arbitrate for dispatch ports to schedule uops for execution.

In at least one embodiment, execution block b11 includes, without limitation, an integer register file/bypass network 3008, a floating point register file/bypass network (“FP register file/bypass network”) 3010, address generation units (“AGUs”) 3012 and 3014, fast Arithmetic Logic Units (ALUs) (“fast ALUs”) 3016 and 3018, a slow Arithmetic Logic Unit (“slow ALU”) 3020, a floating point ALU (“FP”) 3022, and a floating point move unit (“FP move”) 3024. In at least one embodiment, integer register file/bypass network 3008 and floating point register file/bypass network 3010 are also referred to herein as “register files 3008, 3010.” In at least one embodiment, AGUSs 3012 and 3014, fast ALUs 3016 and 3018, slow ALU 3020, floating point ALU 3022, and floating point move unit 3024 are also referred to herein as “execution units 3012, 3014, 3016, 3018, 3020, 3022, and 3024.” In at least one embodiment, execution block b11 may include, without limitation, any number (including zero) and type of register files, bypass networks, address generation units, and execution units, in any combination.

In at least one embodiment, register files 3008, 3010 may be arranged between uop schedulers 3002, 3004, 3006, and execution units 3012, 3014, 3016, 3018, 3020, 3022, and 3024. In at least one embodiment, integer register file/bypass network 3008 performs integer operations. In at least one embodiment, floating point register file/bypass network 3010 performs floating point operations. In at least one embodiment, each of register files 3008, 3010 may include, without limitation, a bypass network that may bypass or forward just completed results that have not yet been written into register file to new dependent uops. In at least one embodiment, register files 3008, 3010 may communicate data with each other. In at least one embodiment, integer register file/bypass network 3008 may include, without limitation, two separate register files, one register file for low-order thirty-two bits of data and a second register file for high order thirty-two bits of data. In at least one embodiment, floating point register file/bypass network 3010 may include, without limitation, 128-bit wide entries because floating point instructions typically have operands from 64 to 128 bits in width.

In at least one embodiment, execution units 3012, 3014, 3016, 3018, 3020, 3022, 3024 may execute instructions. In at least one embodiment, register files 3008, 3010 store integer and floating point data operand values that micro-instructions need to execute. In at least one embodiment, processor 3000 may include, without limitation, any number and combination of execution units 3012, 3014, 3016, 3018, 3020, 3022, 3024. In at least one embodiment, floating point ALU 3022 and floating point move unit 3024, may execute floating point, MMX, SIMD, AVX and SSE, or other operations, including specialized machine learning instructions. In at least one embodiment, floating point ALU 3022 may include, without limitation, a 64-bit by 64-bit floating point divider to execute divide, square root, and remainder micro ops. In at least one embodiment, instructions involving a floating point value may be handled with floating point hardware. In at least one embodiment, ALU operations may be passed to fast ALUs 3016, 3018. In at least one embodiment, fast ALUS 3016, 3018 may execute fast operations with an effective latency of half a clock cycle. In at least one embodiment, most complex integer operations go to slow ALU 3020 as slow ALU 3020 may include, without limitation, integer execution hardware for long-latency type of operations, such as a multiplier, shifts, flag logic, and branch processing. In at least one embodiment, memory load/store operations may be executed by AGUS 3012, 3014. In at least one embodiment, fast ALU 3016, fast ALU 3018, and slow ALU 3020 may perform integer operations on 64-bit data operands. In at least one embodiment, fast ALU 3016, fast ALU 3018, and slow ALU 3020 may be implemented to support a variety of data bit sizes including sixteen, thirty-two, 128, 256, etc. In at least one embodiment, floating point ALU 3022 and floating point move unit 3024 may be implemented to support a range of operands having bits of various widths. In at least one embodiment, floating point ALU 3022 and floating point move unit 3024 may operate on 128-bit wide packed data operands in conjunction with SIMD and multimedia instructions.

In at least one embodiment, uop schedulers 3002, 3004, 3006, dispatch dependent operations before parent load has finished executing. In at least one embodiment, as uops may be speculatively scheduled and executed in processor 3000, processor 3000 may also include logic to handle memory misses. In at least one embodiment, if a data load misses in data cache, there may be dependent operations in flight in pipeline that have left scheduler with temporarily incorrect data. In at least one embodiment, a replay mechanism tracks and re-executes instructions that use incorrect data. In at least one embodiment, dependent operations might need to be replayed and independent ones may be allowed to complete. In at least one embodiment, schedulers and replay mechanism of at least one embodiment of a processor may also be designed to catch instruction sequences for text string comparison operations.

In at least one embodiment, term “registers” may refer to on-board processor storage locations that may be used as part of instructions to identify operands. In at least one embodiment, registers may be those that may be usable from outside of processor (from a programmer's perspective). In at least one embodiment, registers might not be limited to a particular type of circuit. Rather, in at least one embodiment, a register may store data, provide data, and perform functions described herein. In at least one embodiment, registers described herein may be implemented by circuitry within a processor using any number of different techniques, such as dedicated physical registers, dynamically allocated physical registers using register renaming, combinations of dedicated and dynamically allocated physical registers, etc. In at least one embodiment, integer registers store 32-bit integer data. A register file of at least one embodiment also contains eight multimedia SIMD registers for packed data.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment portions or all of inference and/or training logic 1415 may be incorporated into EXE Block 3011 and other memory or registers shown or not shown. For example, in at least one embodiment, training and/or inferencing techniques described herein may use one or more of ALUs illustrated in EXE Block 3011. Moreover, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of EXE Block 3011 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a GPU as described above.

FIG. 31 illustrates a deep learning application processor 3100, according to at least one embodiment. In at least one embodiment, deep learning application processor 3100 uses instructions that, if executed by deep learning application processor 3100, cause deep learning application processor 3100 to perform some or all of processes and techniques described throughout this disclosure. In at least one embodiment, deep learning application processor 3100 is an application-specific integrated circuit (ASIC). In at least one embodiment, application processor 3100 performs matrix multiply operations either “hard-wired” into hardware as a result of performing one or more instructions or both. In at least one embodiment, deep learning application processor 3100 includes, without limitation, processing clusters 3110(1)-3110(12), Inter-Chip Links (“ICLs”) 3120(1)-3120(12), Inter-Chip Controllers (“ICCs”) 3130(1)-3130(2), high bandwidth memory second generation (“HBM2”) 3140(1)-3140(4), memory controllers (“Mem Ctrlrs”) 3142(1)-3142(4), high bandwidth memory physical layer (“HBM PHY”) 3144(1)-3144(4), a management-controller central processing unit (“management-controller CPU”) 3150, a Serial Peripheral Interface, Inter-Integrated Circuit, and General Purpose Input/Output block (“SPI, I2C, GPIO”) 3160, a peripheral component interconnect express controller and direct memory access block (“PCIe Controller and DMA”) 3170, and a sixteen-lane peripheral component interconnect express port (“PCI Express x 16”) 3180.

In at least one embodiment, processing clusters 3110 may perform deep learning operations, including inference or prediction operations based on weight parameters calculated one or more training techniques, including those described herein. In at least one embodiment, each processing cluster 3110 may include, without limitation, any number and type of processors. In at least one embodiment, deep learning application processor 3100 may include any number and type of processing clusters 3100. In at least one embodiment, Inter-Chip Links 3120 are bi-directional. In at least one embodiment, Inter-Chip Links 3120 and Inter-Chip Controllers 3130 enable multiple deep learning application processors 3100 to exchange information, including activation information resulting from performing one or more machine learning algorithms embodied in one or more neural networks. In at least one embodiment, deep learning application processor 3100 may include any number (including zero) and type of ICLs 3120 and ICCs 3130.

In at least one embodiment, HBM2s 3140 provide a total of 32 Gigabytes (GB) of memory. HBM2 3140(i) is associated with both memory controller 3142(i) and HBM PHY 3144(i). In at least one embodiment, any number of HBM2s 3140 may provide any type and total amount of high bandwidth memory and may be associated with any number (including zero) and type of memory controllers 3142 and HBM PHYs 3144. In at least one embodiment, SPI, I2C, GPIO 3160, PCIe Controller and DMA 3170, and/or PCIe 3180 may be replaced with any number and type of blocks that enable any number and type of communication standards in any technically feasible fashion.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, deep learning application processor is used to train a machine learning model, such as a neural network, to predict or infer information provided to deep learning application processor 3100. In at least one embodiment, deep learning application processor 3100 is used to infer or predict information based on a trained machine learning model (e.g., neural network) that has been trained by another processor or system or by deep learning application processor 3100. In at least one embodiment, processor 3100 may be used to perform one or more neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a GPU as described above.

FIG. 32 is a block diagram of a neuromorphic processor 3200, according to at least one embodiment. In at least one embodiment, neuromorphic processor 3200 may receive one or more inputs from sources external to neuromorphic processor 3200. In at least one embodiment, these inputs may be transmitted to one or more neurons 3202 within neuromorphic processor 3200. In at least one embodiment, neurons 3202 and components thereof may be implemented using circuitry or logic, including one or more arithmetic logic units (ALUs). In at least one embodiment, neuromorphic processor 3200 may include, without limitation, thousands or millions of instances of neurons 3202, but any suitable number of neurons 3202 may be used. In at least one embodiment, each instance of neuron 3202 may include a neuron input 3204 and a neuron output 3206. In at least one embodiment, neurons 3202 may generate outputs that may be transmitted to inputs of other instances of neurons 3202. For example, in at least one embodiment, neuron inputs 3204 and neuron outputs 3206 may be interconnected via synapses 3208.

In at least one embodiment, neurons 3202 and synapses 3208 may be interconnected such that neuromorphic processor 3200 operates to process or analyze information received by neuromorphic processor 3200. In at least one embodiment, neurons 3202 may transmit an output pulse (or “fire” or “spike”) when inputs received through neuron input 3204 exceed a threshold. In at least one embodiment, neurons 3202 may sum or integrate signals received at neuron inputs 3204. For example, in at least one embodiment, neurons 3202 may be implemented as leaky integrate-and-fire neurons, wherein if a sum (referred to as a “membrane potential”) exceeds a threshold value, neuron 3202 may generate an output (or “fire”) using a transfer function such as a sigmoid or threshold function. In at least one embodiment, a leaky integrate-and-fire neuron may sum signals received at neuron inputs 3204 into a membrane potential and may also apply a decay factor (or leak) to reduce a membrane potential. In at least one embodiment, a leaky integrate-and-fire neuron may fire if multiple input signals are received at neuron inputs 3204 rapidly enough to exceed a threshold value (i.e., before a membrane potential decays too low to fire). In at least one embodiment, neurons 3202 may be implemented using circuits or logic that receive inputs, integrate inputs into a membrane potential, and decay a membrane potential. In at least one embodiment, inputs may be averaged, or any other suitable transfer function may be used. Furthermore, in at least one embodiment, neurons 3202 may include, without limitation, comparator circuits or logic that generate an output spike at neuron output 3206 when result of applying a transfer function to neuron input 3204 exceeds a threshold. In at least one embodiment, once neuron 3202 fires, it may disregard previously received input information by, for example, resetting a membrane potential to 0 or another suitable default value. In at least one embodiment, once membrane potential is reset to 0, neuron 3202 may resume normal operation after a suitable period of time (or refractory period).

In at least one embodiment, neurons 3202 may be interconnected through synapses 3208. In at least one embodiment, synapses 3208 may operate to transmit signals from an output of a first neuron 3202 to an input of a second neuron 3202. In at least one embodiment, neurons 3202 may transmit information over more than one instance of synapse 3208. In at least one embodiment, one or more instances of neuron output 3206 may be connected, via an instance of synapse 3208, to an instance of neuron input 3204 in same neuron 3202. In at least one embodiment, an instance of neuron 3202 generating an output to be transmitted over an instance of synapse 3208 may be referred to as a “pre-synaptic neuron” with respect to that instance of synapse 3208. In at least one embodiment, an instance of neuron 3202 receiving an input transmitted over an instance of synapse 3208 may be referred to as a “post-synaptic neuron” with respect to that instance of synapse 3208. Because an instance of neuron 3202 may receive inputs from one or more instances of synapse 3208, and may also transmit outputs over one or more instances of synapse 3208, a single instance of neuron 3202 may therefore be both a “pre-synaptic neuron” and “post-synaptic neuron,” with respect to various instances of synapses 3208, in at least one embodiment.

In at least one embodiment, neurons 3202 may be organized into one or more layers. Each instance of neuron 3202 may have one neuron output 3206 that may fan out through one or more synapses 3208 to one or more neuron inputs 3204. In at least one embodiment, neuron outputs 3206 of neurons 3202 in a first layer 3210 may be connected to neuron inputs 3204 of neurons 3202 in a second layer 3212. In at least one embodiment, layer 3210 may be referred to as a “feed-forward layer.” In at least one embodiment, each instance of neuron 3202 in an instance of first layer 3210 may fan out to each instance of neuron 3202 in second layer 3212. In at least one embodiment, first layer 3210 may be referred to as a “fully connected feed-forward layer.” In at least one embodiment, each instance of neuron 3202 in an instance of second layer 3212 may fan out to fewer than all instances of neuron 3202 in a third layer 3214. In at least one embodiment, second layer 3212 may be referred to as a “sparsely connected feed-forward layer.” In at least one embodiment, neurons 3202 in second layer 3212 may fan out to neurons 3202 in multiple other layers, including to neurons 3202 in (same) second layer 3212. In at least one embodiment, second layer 3212 may be referred to as a “recurrent layer.” neuromorphic processor 3200 may include, without limitation, any suitable combination of recurrent layers and feed-forward layers, including, without limitation, both sparsely connected feed-forward layers and fully connected feed-forward layers.

In at least one embodiment, neuromorphic processor 3200 may include, without limitation, a reconfigurable interconnect architecture or dedicated hard wired interconnects to connect synapse 3208 to neurons 3202. In at least one embodiment, neuromorphic processor 3200 may include, without limitation, circuitry or logic that allows synapses to be allocated to different neurons 3202 as needed based on neural network topology and neuron fan-in/out. For example, in at least one embodiment, synapses 3208 may be connected to neurons 3202 using an interconnect fabric, such as network-on-chip, or with dedicated connections. In at least one embodiment, synapse interconnections and components thereof may be implemented using circuitry or logic.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a GPU as described above.

FIG. 33 is a block diagram of a processing system, according to at least one embodiment. In at least one embodiment, system 3300 includes one or more processors 3302 and one or more graphics processors 3308, and may be a single processor desktop system, a multiprocessor workstation system, or a server system having a large number of processors 3302 or processor cores 3307. In at least one embodiment, system 3300 is a processing platform incorporated within a system-on-a-chip (SoC) integrated circuit for use in mobile, handheld, or embedded devices.

In at least one embodiment, system 3300 can include, or be incorporated within a server-based gaming platform, a game console, including a game and media console, a mobile gaming console, a handheld game console, or an online game console. In at least one embodiment, system 3300 is a mobile phone, smart phone, tablet computing device or mobile Internet device. In at least one embodiment, processing system 3300 can also include, couple with, or be integrated within a wearable device, such as a smart watch wearable device, smart eyewear device, augmented reality device, or virtual reality device. In at least one embodiment, processing system 3300 is a television or set top box device having one or more processors 3302 and a graphical interface generated by one or more graphics processors 3308.

In at least one embodiment, one or more processors 3302 each include one or more processor cores 3307 to process instructions which, when executed, perform operations for system and user software. In at least one embodiment, each of one or more processor cores 3307 is configured to process a specific instruction set 3309. In at least one embodiment, instruction set 3309 may facilitate Complex Instruction Set Computing (CISC), Reduced Instruction Set Computing (RISC), or computing via a Very Long Instruction Word (VLIW). In at least one embodiment, processor cores 3307 may each process a different instruction set 3309, which may include instructions to facilitate emulation of other instruction sets. In at least one embodiment, processor core 3307 may also include other processing devices, such a Digital Signal Processor (DSP).

In at least one embodiment, processor 3302 includes cache memory 3304. In at least one embodiment, processor 3302 can have a single internal cache or multiple levels of internal cache. In at least one embodiment, cache memory is shared among various components of processor 3302. In at least one embodiment, processor 3302 also uses an external cache (e.g., a Level-3 (L3) cache or Last Level Cache (LLC)) (not shown), which may be shared among processor cores 3307 using known cache coherency techniques. In at least one embodiment, register file 3306 is additionally included in processor 3302 which may include different types of registers for storing different types of data (e.g., integer registers, floating point registers, status registers, and an instruction pointer register). In at least one embodiment, register file 3306 may include general-purpose registers or other registers.

In at least one embodiment, one or more processor(s) 3302 are coupled with one or more interface bus(es) 3310 to transmit communication signals such as address, data, or control signals between processor 3302 and other components in system 3300. In at least one embodiment interface bus 3310, in one embodiment, can be a processor bus, such as a version of a Direct Media Interface (DMI) bus. In at least one embodiment, interface 3310 is not limited to a DMI bus, and may include one or more Peripheral Component Interconnect buses (e.g., PCI, PCI Express), memory busses, or other types of interface busses. In at least one embodiment processor(s) 3302 include an integrated memory controller 3316 and a platform controller hub 3330. In at least one embodiment, memory controller 3316 facilitates communication between a memory device and other components of system 3300, while platform controller hub (PCH) 3330 provides connections to I/O devices via a local I/O bus.

In at least one embodiment, memory device 3320 can be a dynamic random access memory (DRAM) device, a static random access memory (SRAM) device, flash memory device, phase-change memory device, or some other memory device having suitable performance to serve as process memory. In at least one embodiment memory device 3320 can operate as system memory for system 3300, to store data 3322 and instructions 3321 for use when one or more processors 3302 executes an application or process. In at least one embodiment, memory controller 3316 also couples with an optional external graphics processor 3312, which may communicate with one or more graphics processors 3308 in processors 3302 to perform graphics and media operations. In at least one embodiment, a display device 3311 can connect to processor(s) 3302. In at least one embodiment display device 3311 can include one or more of an internal display device, as in a mobile electronic device or a laptop device or an external display device attached via a display interface (e.g., DisplayPort, etc.). In at least one embodiment, display device 3311 can include a head mounted display (MID) such as a stereoscopic display device for use in virtual reality (VR) applications or augmented reality (AR) applications.

In at least one embodiment, platform controller hub 3330 enables peripherals to connect to memory device 3320 and processor 3302 via a high-speed I/O bus. In at least one embodiment, I/O peripherals include, but are not limited to, an audio controller 3346, a network controller 3334, a firmware interface 3328, a wireless transceiver 3326, touch sensors 3325, a data storage device 3324 (e.g., hard disk drive, flash memory, etc.). In at least one embodiment, data storage device 3324 can connect via a storage interface (e.g., SATA) or via a peripheral bus, such as a Peripheral Component Interconnect bus (e.g., PCI, PCI Express). In at least one embodiment, touch sensors 3325 can include touch screen sensors, pressure sensors, or fingerprint sensors. In at least one embodiment, wireless transceiver 3326 can be a Wi-Fi transceiver, a Bluetooth transceiver, or a mobile network transceiver such as a 3G, 4G, or Long Term Evolution (LTE) transceiver. In at least one embodiment, firmware interface 3328 enables communication with system firmware, and can be, for example, a unified extensible firmware interface (UEFI). In at least one embodiment, network controller 3334 can enable a network connection to a wired network. In at least one embodiment, a high-performance network controller (not shown) couples with interface bus 3310. In at least one embodiment, audio controller 3346 is a multi-channel high definition audio controller. In at least one embodiment, system 3300 includes an optional legacy I/O controller 3340 for coupling legacy (e.g., Personal System 2 (PS/2)) devices to system. In at least one embodiment, platform controller hub 3330 can also connect to one or more Universal Serial Bus (USB) controllers 3342 connect input devices, such as keyboard and mouse 3343 combinations, a camera 3344, or other USB input devices.

In at least one embodiment, an instance of memory controller 3316 and platform controller hub 3330 may be integrated into a discreet external graphics processor, such as external graphics processor 3312. In at least one embodiment, platform controller hub 3330 and/or memory controller 3316 may be external to one or more processor(s) 3302. For example, in at least one embodiment, system 3300 can include an external memory controller 3316 and platform controller hub 3330, which may be configured as a memory controller hub and peripheral controller hub within a system chipset that is in communication with processor(s) 3302.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment portions or all of inference and/or training logic 1415 may be incorporated into graphics processor 3300. For example, in at least one embodiment, training and/or inferencing techniques described herein may use one or more of ALUs embodied in 3D pipeline 3312. Moreover, in at least one embodiment, inferencing and/or training operations described herein may be done using logic other than logic illustrated in FIG. 14A or 14B. In at least one embodiment, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of graphics processor 3300 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 34 is a block diagram of a processor 3400 having one or more processor cores 3402A-3402N, an integrated memory controller 3414, and an integrated graphics processor 3408, according to at least one embodiment. In at least one embodiment, processor 3400 can include additional cores up to and including additional core 3402N represented by dashed lined boxes. In at least one embodiment, each of processor cores 3402A-3402N includes one or more internal cache units 3404A-3404N. In at least one embodiment, each processor core also has access to one or more shared cached units 3406.

In at least one embodiment, internal cache units 3404A-3404N and shared cache units 3406 represent a cache memory hierarchy within processor 3400. In at least one embodiment, cache memory units 3404A-3404N may include at least one level of instruction and data cache within each processor core and one or more levels of shared mid-level cache, such as a Level 2 (L2), Level 3 (L3), Level 4 (L4), or other levels of cache, where a highest level of cache before external memory is classified as an LLC. In at least one embodiment, cache coherency logic maintains coherency between various cache units 3406 and 3404A-3404N.

In at least one embodiment, processor 3400 may also include a set of one or more bus controller units 3416 and a system agent core 3410. In at least one embodiment, one or more bus controller units 3416 manage a set of peripheral buses, such as one or more PCI or PCI express busses. In at least one embodiment, system agent core 3410 provides management functionality for various processor components. In at least one embodiment, system agent core 3410 includes one or more integrated memory controllers 3414 to manage access to various external memory devices (not shown).

In at least one embodiment, one or more of processor cores 3402A-3402N include support for simultaneous multi-threading. In at least one embodiment, system agent core 3410 includes components for coordinating and operating cores 3402A-3402N during multi-threaded processing. In at least one embodiment, system agent core 3410 may additionally include a power control unit (PCU), which includes logic and components to regulate one or more power states of processor cores 3402A-3402N and graphics processor 3408.

In at least one embodiment, processor 3400 additionally includes graphics processor 3408 to execute graphics processing operations. In at least one embodiment, graphics processor 3408 couples with shared cache units 3406, and system agent core 3410, including one or more integrated memory controllers 3414. In at least one embodiment, system agent core 3410 also includes a display controller 3411 to drive graphics processor output to one or more coupled displays. In at least one embodiment, display controller 3411 may also be a separate module coupled with graphics processor 3408 via at least one interconnect, or may be integrated within graphics processor 3408.

In at least one embodiment, a ring based interconnect unit 3412 is used to couple internal components of processor 3400. In at least one embodiment, an alternative interconnect unit may be used, such as a point-to-point interconnect, a switched interconnect, or other techniques. In at least one embodiment, graphics processor 3408 couples with ring interconnect 3412 via an I/O link 3413.

In at least one embodiment, I/O link 3413 represents at least one of multiple varieties of I/O interconnects, including an on package I/O interconnect which facilitates communication between various processor components and a high-performance embedded memory module 3418, such as an eDRAM module. In at least one embodiment, each of processor cores 3402A-3402N and graphics processor 3408 use embedded memory modules 3418 as a shared Last Level Cache.

In at least one embodiment, processor cores 3402A-3402N are homogenous cores executing a common instruction set architecture. In at least one embodiment, processor cores 3402A-3402N are heterogeneous in terms of instruction set architecture (ISA), where one or more of processor cores 3402A-3402N execute a common instruction set, while one or more other cores of processor cores 3402A-34-02N executes a subset of a common instruction set or a different instruction set. In at least one embodiment, processor cores 3402A-3402N are heterogeneous in terms of microarchitecture, where one or more cores having a relatively higher power consumption couple with one or more power cores having a lower power consumption. In at least one embodiment, processor 3400 can be implemented on one or more chips or as an SoC integrated circuit.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment portions or all of inference and/or training logic 1415 may be incorporated into graphics processor 3410. For example, in at least one embodiment, training and/or inferencing techniques described herein may use one or more of ALUs embodied in 3D pipeline 3312, graphics core(s) 3415A, shared function logic 3416, graphics core(s) 3415B, shared function logic 3420, or other logic in FIG. 34. Moreover, in at least one embodiment, inferencing and/or training operations described herein may be done using logic other than logic illustrated in FIG. 14A or 14B. In at least one embodiment, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of graphics processor 3410 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 35 is a block diagram of a graphics processor 3500, which may be a discrete graphics processing unit, or may be a graphics processor integrated with a plurality of processing cores. In at least one embodiment, graphics processor 3500 communicates via a memory mapped I/O interface to registers on graphics processor 3500 and with commands placed into memory. In at least one embodiment, graphics processor 3500 includes a memory interface 3514 to access memory. In at least one embodiment, memory interface 3514 is an interface to local memory, one or more internal caches, one or more shared external caches, and/or to system memory.

In at least one embodiment, graphics processor 3500 also includes a display controller 3502 to drive display output data to a display device 3520. In at least one embodiment, display controller 3502 includes hardware for one or more overlay planes for display device 3520 and composition of multiple layers of video or user interface elements. In at least one embodiment, display device 3520 can be an internal or external display device. In at least one embodiment, display device 3520 is a head mounted display device, such as a virtual reality (VR) display device or an augmented reality (AR) display device. In at least one embodiment, graphics processor 3500 includes a video codec engine 3506 to encode, decode, or transcode media to, from, or between one or more media encoding formats, including, but not limited to Moving Picture Experts Group (MPEG) formats such as MPEG-2, Advanced Video Coding (AVC) formats such as H.264/MPEG-4 AVC, as well as the Society of Motion Picture & Television Engineers (SMPTE) 421M/VC-1, and Joint Photographic Experts Group (JPEG) formats such as JPEG, and Motion JPEG (MJPEG) formats.

In at least one embodiment, graphics processor 3500 includes a block image transfer (BLIT) engine 3504 to perform two-dimensional (2D) rasterizer operations including, for example, bit-boundary block transfers. However, in at least one embodiment, 2D graphics operations are performed using one or more components of graphics processing engine (GPE) 3510. In at least one embodiment, GPE 3510 is a compute engine for performing graphics operations, including three-dimensional (3D) graphics operations and media operations.

In at least one embodiment, GPE 3510 includes a 3D pipeline 3512 for performing 3D operations, such as rendering three-dimensional images and scenes using processing functions that act upon 3D primitive shapes (e.g., rectangle, triangle, etc.). 3D pipeline 3512 includes programmable and fixed function elements that perform various tasks and/or spawn execution threads to a 3D/Media sub-system 3515. While 3D pipeline 3512 can be used to perform media operations, in at least one embodiment, GPE 3510 also includes a media pipeline 3516 that is used to perform media operations, such as video post-processing and image enhancement.

In at least one embodiment, media pipeline 3516 includes fixed function or programmable logic units to perform one or more specialized media operations, such as video decode acceleration, video de-interlacing, and video encode acceleration in place of, or on behalf of video codec engine 3506. In at least one embodiment, media pipeline 3516 additionally includes a thread spawning unit to spawn threads for execution on 3D/Media sub-system 3515. In at least one embodiment, spawned threads perform computations for media operations on one or more graphics execution units included in 3D/Media sub-system 3515.

In at least one embodiment, 3D/Media subsystem 3515 includes logic for executing threads spawned by 3D pipeline 3512 and media pipeline 3516. In at least one embodiment, 3D pipeline 3512 and media pipeline 3516 send thread execution requests to 3D/Media subsystem 3515, which includes thread dispatch logic for arbitrating and dispatching various requests to available thread execution resources. In at least one embodiment, execution resources include an array of graphics execution units to process 3D and media threads. In at least one embodiment, 3D/Media subsystem 3515 includes one or more internal caches for thread instructions and data. In at least one embodiment, subsystem 3515 also includes shared memory, including registers and addressable memory, to share data between threads and to store output data.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment portions or all of inference and/or training logic 1415 may be incorporated into graphics processor 3500. For example, in at least one embodiment, training and/or inferencing techniques described herein may use one or more of ALUs embodied in 3D pipeline 3512. Moreover, in at least one embodiment, inferencing and/or training operations described herein may be done using logic other than logic illustrated in FIG. 14A or 14B. In at least one embodiment, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of graphics processor 3500 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 36 is a block diagram of a graphics processing engine 3610 of a graphics processor in accordance with at least one embodiment. In at least one embodiment, graphics processing engine (GPE) 3610 is a version of GPE 3510 shown in FIG. 35. In at least one embodiment, media pipeline 3516 is optional and may not be explicitly included within GPE 3610. In at least one embodiment, a separate media and/or image processor is coupled to GPE 3610.

In at least one embodiment, GPE 3610 is coupled to or includes a command streamer 3603, which provides a command stream to 3D pipeline 3512 and/or media pipelines 3516. In at least one embodiment, command streamer 3603 is coupled to memory, which can be system memory, or one or more of internal cache memory and shared cache memory. In at least one embodiment, command streamer 3603 receives commands from memory and sends commands to 3D pipeline 3512 and/or media pipeline 3516. In at least one embodiment, commands are instructions, primitives, or micro-operations fetched from a ring buffer, which stores commands for 3D pipeline 3512 and media pipeline 3516. In at least one embodiment, a ring buffer can additionally include batch command buffers storing batches of multiple commands. In at least one embodiment, commands for 3D pipeline 3512 can also include references to data stored in memory, such as but not limited to vertex and geometry data for 3D pipeline 3512 and/or image data and memory objects for media pipeline 3516. In at least one embodiment, 3D pipeline 3512 and media pipeline 3516 process commands and data by performing operations or by dispatching one or more execution threads to a graphics core array 3614. In at least one embodiment graphics core array 3614 includes one or more blocks of graphics cores (e.g., graphics core(s) 3615A, graphics core(s) 3615B), each block including one or more graphics cores. In at least one embodiment, each graphics core includes a set of graphics execution resources that includes general-purpose and graphics specific execution logic to perform graphics and compute operations, as well as fixed function texture processing and/or machine learning and artificial intelligence acceleration logic, including inference and/or training logic 1415 in FIG. 14A and FIG. 14B.

In at least one embodiment, 3D pipeline 3512 includes fixed function and programmable logic to process one or more shader programs, such as vertex shaders, geometry shaders, pixel shaders, fragment shaders, compute shaders, or other shader programs, by processing instructions and dispatching execution threads to graphics core array 3614. In at least one embodiment, graphics core array 3614 provides a unified block of execution resources for use in processing shader programs. In at least one embodiment, multi-purpose execution logic (e.g., execution units) within graphics core(s) 3615A-3615B of graphic core array 3614 includes support for various 3D API shader languages and can execute multiple simultaneous execution threads associated with multiple shaders.

In at least one embodiment, graphics core array 3614 also includes execution logic to perform media functions, such as video and/or image processing. In at least one embodiment, execution units additionally include general-purpose logic that is programmable to perform parallel general-purpose computational operations, in addition to graphics processing operations.

In at least one embodiment, output data generated by threads executing on graphics core array 3614 can output data to memory in a unified return buffer (URB) 3618. URB 3618 can store data for multiple threads. In at least one embodiment, URB 3618 may be used to send data between different threads executing on graphics core array 3614. In at least one embodiment, URB 3618 may additionally be used for synchronization between threads on graphics core array 3614 and fixed function logic within shared function logic 3620.

In at least one embodiment, graphics core array 3614 is scalable, such that graphics core array 3614 includes a variable number of graphics cores, each having a variable number of execution units based on a target power and performance level of GPE 3610. In at least one embodiment, execution resources are dynamically scalable, such that execution resources may be enabled or disabled as needed.

In at least one embodiment, graphics core array 3614 is coupled to shared function logic 3620 that includes multiple resources that are shared between graphics cores in graphics core array 3614. In at least one embodiment, shared functions performed by shared function logic 3620 are embodied in hardware logic units that provide specialized supplemental functionality to graphics core array 3614. In at least one embodiment, shared function logic 3620 includes but is not limited to sampler 3621, math 3622, and inter-thread communication (ITC) 3623 logic. In at least one embodiment, one or more cache(s) 3625 are in included in or couple to shared function logic 3620.

In at least one embodiment, a shared function is used if demand for a specialized function is insufficient for inclusion within graphics core array 3614. In at least one embodiment, a single instantiation of a specialized function is used in shared function logic 3620 and shared among other execution resources within graphics core array 3614. In at least one embodiment, specific shared functions within shared function logic 3620 that are used extensively by graphics core array 3614 may be included within shared function logic 3616 within graphics core array 3614. In at least one embodiment, shared function logic 3616 within graphics core array 3614 can include some or all logic within shared function logic 3620. In at least one embodiment, all logic elements within shared function logic 3620 may be duplicated within shared function logic 3616 of graphics core array 3614. In at least one embodiment, shared function logic 3620 is excluded in favor of shared function logic 3616 within graphics core array 3614.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment portions or all of inference and/or training logic 1415 may be incorporated into graphics processor 3610. For example, in at least one embodiment, training and/or inferencing techniques described herein may use one or more of ALUs embodied in 3D pipeline 3512, graphics core(s) 3615A, shared function logic 3616, graphics core(s) 3615B, shared function logic 3620, or other logic in FIG. 36. Moreover, in at least one embodiment, inferencing and/or training operations described herein may be done using logic other than logic illustrated in FIG. 14A or 14B. In at least one embodiment, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of graphics processor 3610 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 37 is a block diagram of hardware logic of a graphics processor core 3700, according to at least one embodiment described herein. In at least one embodiment, graphics processor core 3700 is included within a graphics core array. In at least one embodiment, graphics processor core 3700, sometimes referred to as a core slice, can be one or multiple graphics cores within a modular graphics processor. In at least one embodiment, graphics processor core 3700 is exemplary of one graphics core slice, and a graphics processor as described herein may include multiple graphics core slices based on target power and performance envelopes. In at least one embodiment, each graphics core 3700 can include a fixed function block 3730 coupled with multiple sub-cores 3701A-3701F, also referred to as sub-slices, that include modular blocks of general-purpose and fixed function logic.

In at least one embodiment, fixed function block 3730 includes a geometry/fixed function pipeline 3736 that can be shared by all sub-cores in graphics processor 3700, for example, in lower performance and/or lower power graphics processor implementations. In at least one embodiment, geometry/fixed function pipeline 3736 includes a 3D fixed function pipeline, a video front-end unit, a thread spawner and thread dispatcher, and a unified return buffer manager, which manages unified return buffers.

In at least one embodiment fixed function block 3730 also includes a graphics SoC interface 3737, a graphics microcontroller 3738, and a media pipeline 3739. Graphics SoC interface 3737 provides an interface between graphics core 3700 and other processor cores within a system on a chip integrated circuit. In at least one embodiment, graphics microcontroller 3738 is a programmable sub-processor that is configurable to manage various functions of graphics processor 3700, including thread dispatch, scheduling, and pre-emption. In at least one embodiment, media pipeline 3739 includes logic to facilitate decoding, encoding, pre-processing, and/or post-processing of multimedia data, including image and video data. In at least one embodiment, media pipeline 3739 implement media operations via requests to compute or sampling logic within sub-cores 3701-3701F.

In at least one embodiment, SoC interface 3737 enables graphics core 3700 to communicate with general-purpose application processor cores (e.g., CPUs) and/or other components within an SoC, including memory hierarchy elements such as a shared last level cache memory, system RAM, and/or embedded on-chip or on-package DRAM. In at least one embodiment, SoC interface 3737 can also enable communication with fixed function devices within an SoC, such as camera imaging pipelines, and enables use of and/or implements global memory atomics that may be shared between graphics core 3700 and CPUs within an SoC. In at least one embodiment, SoC interface 3737 can also implement power management controls for graphics core 3700 and enable an interface between a clock domain of graphic core 3700 and other clock domains within an SoC. In at least one embodiment, SoC interface 3737 enables receipt of command buffers from a command streamer and global thread dispatcher that are configured to provide commands and instructions to each of one or more graphics cores within a graphics processor. In at least one embodiment, commands and instructions can be dispatched to media pipeline 3739, when media operations are to be performed, or a geometry and fixed function pipeline (e.g., geometry and fixed function pipeline 3736, geometry and fixed function pipeline 3714) when graphics processing operations are to be performed.

In at least one embodiment, graphics microcontroller 3738 can be configured to perform various scheduling and management tasks for graphics core 3700. In at least one embodiment, graphics microcontroller 3738 can perform graphics and/or compute workload scheduling on various graphics parallel engines within execution unit (EU) arrays 3702A-3702F, 3704A-3704F within sub-cores 3701A-3701F. In at least one embodiment, host software executing on a CPU core of an SoC including graphics core 3700 can submit workloads one of multiple graphic processor doorbells, which invokes a scheduling operation on an appropriate graphics engine. In at least one embodiment, scheduling operations include determining which workload to run next, submitting a workload to a command streamer, pre-empting existing workloads running on an engine, monitoring progress of a workload, and notifying host software when a workload is complete. In at least one embodiment, graphics microcontroller 3738 can also facilitate low-power or idle states for graphics core 3700, providing graphics core 3700 with an ability to save and restore registers within graphics core 3700 across low-power state transitions independently from an operating system and/or graphics driver software on a system.

In at least one embodiment, graphics core 3700 may have greater than or fewer than illustrated sub-cores 3701A-3701F, up to N modular sub-cores. For each set of N sub-cores, in at least one embodiment, graphics core 3700 can also include shared function logic 3710, shared and/or cache memory 3712, a geometry/fixed function pipeline 3714, as well as additional fixed function logic 3716 to accelerate various graphics and compute processing operations. In at least one embodiment, shared function logic 3710 can include logic units (e.g., sampler, math, and/or inter-thread communication logic) that can be shared by each N sub-cores within graphics core 3700. Shared and/or cache memory 3712 can be a last-level cache for N sub-cores 3701A-3701F within graphics core 3700 and can also serve as shared memory that is accessible by multiple sub-cores. In at least one embodiment, geometry/fixed function pipeline 3714 can be included instead of geometry/fixed function pipeline 3736 within fixed function block 3730 and can include same or similar logic units.

In at least one embodiment, graphics core 3700 includes additional fixed function logic 3716 that can include various fixed function acceleration logic for use by graphics core 3700. In at least one embodiment, additional fixed function logic 3716 includes an additional geometry pipeline for use in position only shading. In position-only shading, at least two geometry pipelines exist, whereas in a full geometry pipeline within geometry/fixed function pipeline 3716, 3736, and a cull pipeline, which is an additional geometry pipeline which may be included within additional fixed function logic 3716. In at least one embodiment, cull pipeline is a trimmed down version of a full geometry pipeline. In at least one embodiment, a full pipeline and a cull pipeline can execute different instances of an application, each instance having a separate context. In at least one embodiment, position only shading can hide long cull runs of discarded triangles, enabling shading to be completed earlier in some instances. For example, in at least one embodiment, cull pipeline logic within additional fixed function logic 3716 can execute position shaders in parallel with a main application and generally generates critical results faster than a full pipeline, as cull pipeline fetches and shades position attribute of vertices, without performing rasterization and rendering of pixels to a frame buffer. In at least one embodiment, cull pipeline can use generated critical results to compute visibility information for all triangles without regard to whether those triangles are culled. In at least one embodiment, full pipeline (which in this instance may be referred to as a replay pipeline) can consume visibility information to skip culled triangles to shade only visible triangles that are finally passed to a rasterization phase.

In at least one embodiment, additional fixed function logic 3716 can also include machine-learning acceleration logic, such as fixed function matrix multiplication logic, for implementations including optimizations for machine learning training or inferencing.

In at least one embodiment, within each graphics sub-core 3701A-3701F includes a set of execution resources that may be used to perform graphics, media, and compute operations in response to requests by graphics pipeline, media pipeline, or shader programs. In at least one embodiment, graphics sub-cores 3701A-3701F include multiple EU arrays 3702A-3702F, 3704A-3704F, thread dispatch and inter-thread communication (TD/IC) logic 3703A-3703F, a 3D (e.g., texture) sampler 3705A-3705F, a media sampler 3706A-3706F, a shader processor 3707A-3707F, and shared local memory (SLM) 3708A-3708F. EU arrays 3702A-3702F, 3704A-3704F each include multiple execution units, which are general-purpose graphics processing units capable of performing floating-point and integer/fixed-point logic operations in service of a graphics, media, or compute operation, including graphics, media, or compute shader programs. In at least one embodiment, TD/IC logic 3703A-3703F performs local thread dispatch and thread control operations for execution units within a sub-core and facilitate communication between threads executing on execution units of a sub-core. In at least one embodiment, 3D sampler 3705A-3705F can read texture or other 3D graphics related data into memory. In at least one embodiment, 3D sampler can read texture data differently based on a configured sample state and texture format associated with a given texture. In at least one embodiment, media sampler 3706A-3706F can perform similar read operations based on a type and format associated with media data. In at least one embodiment, each graphics sub-core 3701A-3701F can alternately include a unified 3D and media sampler. In at least one embodiment, threads executing on execution units within each of sub-cores 3701A-3701F can make use of shared local memory 3708A-3708F within each sub-core, to enable threads executing within a thread group to execute using a common pool of on-chip memory.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, portions or all of inference and/or training logic 1415 may be incorporated into graphics processor 3710. For example, in at least one embodiment, training and/or inferencing techniques described herein may use one or more of ALUs embodied in 3D pipeline 3710, graphics microcontroller 3738, geometry & fixed function pipeline 3714 and 3736, or other logic in FIG. 34. Moreover, in at least one embodiment, inferencing and/or training operations described herein may be done using logic other than logic illustrated in FIG. 14A or 14B. In at least one embodiment, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of graphics processor 3700 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIGS. 38A-38B illustrate thread execution logic 3800 including an array of processing elements of a graphics processor core according to at least one embodiment. FIG. 38A illustrates at least one embodiment, in which thread execution logic 3800 is used. FIG. 38B illustrates exemplary internal details of an execution unit, according to at least one embodiment.

As illustrated in FIG. 38A, in at least one embodiment, thread execution logic 3800 includes a shader processor 3802, a thread dispatcher 3804, instruction cache 3806, a scalable execution unit array including a plurality of execution units 3808A-3808N, a sampler 3810, a data cache 3812, and a data port 3814. In at least one embodiment a scalable execution unit array can dynamically scale by enabling or disabling one or more execution units (e.g., any of execution unit 3808A, 3808B, 3808C, 3808D, through 3808N-1 and 3808N) based on computational requirements of a workload, for example. In at least one embodiment, scalable execution units are interconnected via an interconnect fabric that links to each of execution unit. In at least one embodiment, thread execution logic 3800 includes one or more connections to memory, such as system memory or cache memory, through one or more of instruction cache 3806, data port 3814, sampler 3810, and execution units 3808A-3808N. In at least one embodiment, each execution unit (e.g., 3808A) is a stand-alone programmable general-purpose computational unit that is capable of executing multiple simultaneous hardware threads while processing multiple data elements in parallel for each thread. In at least one embodiment, array of execution units 3808A-3808N is scalable to include any number individual execution units.

In at least one embodiment, execution units 3808A-3808N are primarily used to execute shader programs. In at least one embodiment, shader processor 3802 can process various shader programs and dispatch execution threads associated with shader programs via a thread dispatcher 3804. In at least one embodiment, thread dispatcher 3804 includes logic to arbitrate thread initiation requests from graphics and media pipelines and instantiate requested threads on one or more execution units in execution units 3808A-3808N. For example, in at least one embodiment, a geometry pipeline can dispatch vertex, tessellation, or geometry shaders to thread execution logic for processing. In at least one embodiment, thread dispatcher 3804 can also process runtime thread spawning requests from executing shader programs.

In at least one embodiment, execution units 3808A-3808N support an instruction set that includes native support for many standard 3D graphics shader instructions, such that shader programs from graphics libraries (e.g., Direct 3D and OpenGL) are executed with a minimal translation. In at least one embodiment, execution units support vertex and geometry processing (e.g., vertex programs, geometry programs, vertex shaders), pixel processing (L, pixel shaders, fragment shaders) and general-purpose processing (e.g., compute and media shaders). In at least one embodiment, each of execution units 3808A-3808N, which include one or more arithmetic logic units (ALUs), is capable of multi-issue single instruction multiple data (SIMD) execution and multi-threaded operation enables an efficient execution environment despite higher latency memory accesses. In at least one embodiment, each hardware thread within each execution unit has a dedicated high-bandwidth register file and associated independent thread-state. In at least one embodiment, execution is multi-issue per clock to pipelines capable of integer, single and double precision floating point operations, SIMD branch capability, logical operations, transcendental operations, and other miscellaneous operations. In at least one embodiment, while waiting for data from memory or one of shared functions, dependency logic within execution units 3808A-3808N causes a waiting thread to sleep until requested data has been returned. In at least one embodiment, while a waiting thread is sleeping, hardware resources may be devoted to processing other threads. For example, in at least one embodiment, during a delay associated with a vertex shader operation, an execution unit can perform operations for a pixel shader, fragment shader, or another type of shader program, including a different vertex shader.

In at least one embodiment, each execution unit in execution units 3808A-3808N operates on arrays of data elements. In at least one embodiment, a number of data elements is “execution size,” or number of channels for an instruction. In at least one embodiment, an execution channel is a logical unit of execution for data element access, masking, and flow control within instructions. In at least one embodiment, a number of channels may be independent of a number of physical Arithmetic Logic Units (ALUs) or Floating Point Units (FPUs) for a particular graphics processor. In at least one embodiment, execution units 3808A-3808N support integer and floating-point data types.

In at least one embodiment, an execution unit instruction set includes SIMD instructions. In at least one embodiment, various data elements can be stored as a packed data type in a register and execution unit will process various elements based on data size of elements. For example, in at least one embodiment, when operating on a 256-bit wide vector, 256 bits of a vector are stored in a register and an execution unit operates on a vector as four separate 64-bit packed data elements (Quad-Word (QW) size data elements), eight separate 32-bit packed data elements (Double Word (DW) size data elements), sixteen separate 16-bit packed data elements (Word (W) size data elements), or thirty-two separate 8-bit data elements (byte (B) size data elements). However, in at least one embodiment, different vector widths and register sizes are possible.

In at least one embodiment, one or more execution units can be combined into a fused execution unit 3809A-3809N having thread control logic (3807A-3807N) that is common to fused EUs. In at least one embodiment, multiple EUs can be fused into an EU group. In at least one embodiment, each EU in fused EU group can be configured to execute a separate SIMD hardware thread. The number of EUs in a fused EU group can vary according to various embodiments. In at least one embodiment, various SIMD widths can be performed per-EU, including but not limited to SIMD8, SIMD16, and SIMD32. In at least one embodiment, each fused graphics execution unit 3809A-3809N includes at least two execution units. For example, in at least one embodiment, fused execution unit 3809A includes a first EU 3808A, second EU 3808B, and thread control logic 3807A that is common to first EU 3808A and second EU 3808B. In at least one embodiment, thread control logic 3807A controls threads executed on fused graphics execution unit 3809A, allowing each EU within fused execution units 3809A-3809N to execute using a common instruction pointer register.

In at least one embodiment, one or more internal instruction caches (e.g., 3806) are included in thread execution logic 3800 to cache thread instructions for execution units. In at least one embodiment, one or more data caches (e.g., 3812) are included to cache thread data during thread execution. In at least one embodiment, a sampler 3810 is included to provide texture sampling for 3D operations and media sampling for media operations. In at least one embodiment, sampler 3810 includes specialized texture or media sampling functionality to process texture or media data during sampling process before providing sampled data to an execution unit.

During execution, in at least one embodiment, graphics and media pipelines send thread initiation requests to thread execution logic 3800 via thread spawning and dispatch logic. In at least one embodiment, once a group of geometric objects has been processed and rasterized into pixel data, pixel processor logic (e.g., pixel shader logic, fragment shader logic, etc.) within shader processor 3802 is invoked to further compute output information and cause results to be written to output surfaces (e.g., color buffers, depth buffers, stencil buffers, etc.). In at least one embodiment, a pixel shader or fragment shader calculates values of various vertex attributes that are to be interpolated across a rasterized object. In at least one embodiment, pixel processor logic within shader processor 3802 then executes an application programming interface (API)-supplied pixel or fragment shader program. In at least one embodiment, to execute a shader program, shader processor 3802 dispatches threads to an execution unit (e.g., 3808A) via thread dispatcher 3804. In at least one embodiment, shader processor 3802 uses texture sampling logic in sampler 3810 to access texture data in texture maps stored in memory. In at least one embodiment, arithmetic operations on texture data and input geometry data compute pixel color data for each geometric fragment, or discards one or more pixels from further processing.

In at least one embodiment, data port 3814 provides a memory access mechanism for thread execution logic 3800 to output processed data to memory for further processing on a graphics processor output pipeline. In at least one embodiment, data port 3814 includes or couples to one or more cache memories (e.g., data cache 3812) to cache data for memory access via a data port.

As illustrated in FIG. 38B, in at least one embodiment, a graphics execution unit 3808 can include an instruction fetch unit 3837, a general register file array (GRF) 3824, an architectural register file array (ARF) 3826, a thread arbiter 3822, a send unit 3830, a branch unit 3832, a set of SIMD floating point units (FPUs) 3834, and In at least one embodiment a set of dedicated integer SIMD ALUs 3835. In at least one embodiment, GRF 3824 and ARF 3826 includes a set of general register files and architecture register files associated with each simultaneous hardware thread that may be active in graphics execution unit 3808. In at least one embodiment, per thread architectural state is maintained in ARF 3826, while data used during thread execution is stored in GRF 3824. In at least one embodiment, execution state of each thread, including instruction pointers for each thread, can be held in thread-specific registers in ARF 3826.

In at least one embodiment, graphics execution unit 3808 has an architecture that is a combination of Simultaneous Multi-Threading (SMT) and fine-grained Interleaved Multi-Threading (IMT). In at least one embodiment, architecture has a modular configuration that can be fine-tuned at design time based on a target number of simultaneous threads and number of registers per execution unit, where execution unit resources are divided across logic used to execute multiple simultaneous threads.

In at least one embodiment, graphics execution unit 3808 can co-issue multiple instructions, which may each be different instructions. In at least one embodiment, thread arbiter 3822 of graphics execution unit thread 3808 can dispatch instructions to one of send unit 3830, branch unit 3842, or SIMD FPU(s) 3834 for execution. In at least one embodiment, each execution thread can access 128 general-purpose registers within GRF 3824, where each register can store 32 bytes, accessible as a SIMD 8-element vector of 32-bit data elements. In at least one embodiment, each execution unit thread has access to 4 Kbytes within GRF 3824, although embodiments are not so limited, and greater or fewer register resources may be provided in other embodiments. In at least one embodiment, up to seven threads can execute simultaneously, although a number of threads per execution unit can also vary according to embodiments. In at least one embodiment, in which seven threads may access 4 Kbytes, GRF 3824 can store a total of 28 Kbytes. In at least one embodiment, flexible addressing modes can permit registers to be addressed together to build effectively wider registers or to represent strided rectangular block data structures.

In at least one embodiment, memory operations, sampler operations, and other longer-latency system communications are dispatched via “send” instructions that are executed by message passing send unit 3830. In at least one embodiment, branch instructions are dispatched to a dedicated branch unit 3832 to facilitate SIMD divergence and eventual convergence.

In at least one embodiment graphics execution unit 3808 includes one or more SIMD floating point units (FPU(s)) 3834 to perform floating-point operations. In at least one embodiment, FPU(s) 3834 also support integer computation. In at least one embodiment FPU(s) 3834 can SIMD execute up to M number of 32-bit floating-point (or integer) operations, or SIMD execute up to 2M 16-bit integer or 16-bit floating-point operations. In at least one embodiment, at least one of FPU(s) provides extended math capability to support high-throughput transcendental math functions and double precision 64-bit floating-point. In at least one embodiment, a set of 8-bit integer SIMD ALUs 3835 are also present, and may be specifically optimized to perform operations associated with machine learning computations.

In at least one embodiment, arrays of multiple instances of graphics execution unit 3808 can be instantiated in a graphics sub-core grouping (e.g., a sub-slice). In at least one embodiment execution unit 3808 can execute instructions across a plurality of execution channels. In at least one embodiment, each thread executed on graphics execution unit 3808 is executed on a different channel.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, portions or all of inference and/or training logic 1415 may be incorporated into execution logic 3800. Moreover, in at least one embodiment, inferencing and/or training operations described herein may be done using logic other than logic illustrated in FIG. 14A or 14B. In at least one embodiment, weight parameters may be stored in on-chip or off-chip memory and/or registers (shown or not shown) that configure ALUs of execution logic 3800 to perform one or more machine learning algorithms, neural network architectures, use cases, or training techniques described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 39 illustrates a parallel processing unit (“PPU”) 3900, according to at least one embodiment. In at least one embodiment, PPU 3900 is configured with machine-readable code that, if executed by PPU 3900, causes PPU 3900 to perform some or all of processes and techniques described throughout this disclosure. In at least one embodiment, PPU 3900 is a multi-threaded processor that is implemented on one or more integrated circuit devices and that utilizes multithreading as a latency-hiding technique designed to process computer-readable instructions (also referred to as machine-readable instructions or simply instructions) on multiple threads in parallel. In at least one embodiment, a thread refers to a thread of execution and is an instantiation of a set of instructions configured to be executed by PPU 3900. In at least one embodiment, PPU 3900 is a graphics processing unit (“GPU”) configured to implement a graphics rendering pipeline for processing three-dimensional (“3D”) graphics data in order to generate two-dimensional (“2D”) image data for display on a display device such as a liquid crystal display (“LCD”) device. In at least one embodiment, PPU 3900 is utilized to perform computations such as linear algebra operations and machine-learning operations. FIG. 39 illustrates an example parallel processor for illustrative purposes only and should be construed as a non-limiting example of processor architectures contemplated within scope of this disclosure and that any suitable processor may be employed to supplement and/or substitute for same.

In at least one embodiment, one or more PPUs 3900 are configured to accelerate High Performance Computing (“HPC”), data center, and machine learning applications. In at least one embodiment, PPU 3900 is configured to accelerate deep learning systems and applications including following non-limiting examples: autonomous vehicle platforms, deep learning, high-accuracy speech, image, text recognition systems, intelligent video analytics, molecular simulations, drug discovery, disease diagnosis, weather forecasting, big data analytics, astronomy, molecular dynamics simulation, financial modeling, robotics, factory automation, real-time language translation, online search optimizations, and personalized user recommendations, and more.

In at least one embodiment, PPU 3900 includes, without limitation, an Input/Output (“I/O”) unit 3906, a front-end unit 3910, a scheduler unit 3912, a work distribution unit 3914, a hub 3916, a crossbar (“Xbar”) 3920, one or more general processing clusters (“GPCs”) 3918, and one or more partition units (“memory partition units”) 3922. In at least one embodiment, PPU 3900 is connected to a host processor or other PPUs 3900 via one or more high-speed GPU interconnects (“GPU interconnects”) 3908. In at least one embodiment, PPU 3900 is connected to a host processor or other peripheral devices via an interconnect 3902. In at least one embodiment, PPU 3900 is connected to a local memory comprising one or more memory devices (“memory”) 3904. In at least one embodiment, memory devices 3904 include, without limitation, one or more dynamic random access memory (“DRAM”) devices. In at least one embodiment, one or more DRAM devices are configured and/or configurable as high-bandwidth memory (“HBM”) subsystems, with multiple DRAM dies stacked within each device.

In at least one embodiment, high-speed GPU interconnect 3908 may refer to a wire-based multi-lane communications link that is used by systems to scale and include one or more PPUs 3900 combined with one or more central processing units (“CPUs”), supports cache coherence between PPUs 3900 and CPUs, and CPU mastering. In at least one embodiment, data and/or commands are transmitted by high-speed GPU interconnect 3908 through hub 3916 to/from other units of PPU 3900 such as one or more copy engines, video encoders, video decoders, power management units, and other components which may not be explicitly illustrated in FIG. 39.

In at least one embodiment, I/O unit 3906 is configured to transmit and receive communications (e.g., commands, data) from a host processor (not illustrated in FIG. 39) over system bus 3902. In at least one embodiment, I/O unit 3906 communicates with host processor directly via system bus 3902 or through one or more intermediate devices such as a memory bridge. In at least one embodiment, I/O unit 3906 may communicate with one or more other processors, such as one or more of PPUs 3900 via system bus 3902. In at least one embodiment, I/O unit 3906 implements a Peripheral Component Interconnect Express (“PCIe”) interface for communications over a PCIe bus. In at least one embodiment, I/O unit 3906 implements interfaces for communicating with external devices.

In at least one embodiment, I/O unit 3906 decodes packets received via system bus 3902. In at least one embodiment, at least some packets represent commands configured to cause PPU 3900 to perform various operations. In at least one embodiment, I/O unit 3906 transmits decoded commands to various other units of PPU 3900 as specified by commands. In at least one embodiment, commands are transmitted to front-end unit 3910 and/or transmitted to hub 3916 or other units of PPU 3900 such as one or more copy engines, a video encoder, a video decoder, a power management unit, etc. (not explicitly illustrated in FIG. 39). In at least one embodiment, I/O unit 3906 is configured to route communications between and among various logical units of PPU 3900.

In at least one embodiment, a program executed by host processor encodes a command stream in a buffer that provides workloads to PPU 3900 for processing. In at least one embodiment, a workload comprises instructions and data to be processed by those instructions. In at least one embodiment, buffer is a region in a memory that is accessible (e.g., read/write) by both host processor and PPU 3900 a host interface unit may be configured to access buffer in a system memory connected to system bus 3902 via memory requests transmitted over system bus 3902 by I/O unit 3906. In at least one embodiment, host processor writes command stream to buffer and then transmits a pointer to start of command stream to PPU 3900 such that front-end unit 3910 receives pointers to one or more command streams and manages one or more command streams, reading commands from command streams and forwarding commands to various units of PPU 3900.

In at least one embodiment, front-end unit 3910 is coupled to scheduler unit 3912 that configures various GPCs 3918 to process tasks defined by one or more command streams. In at least one embodiment, scheduler unit 3912 is configured to track state information related to various tasks managed by scheduler unit 3912 where state information may indicate which of GPCs 3918 a task is assigned to, whether task is active or inactive, a priority level associated with task, and so forth. In at least one embodiment, scheduler unit 3912 manages execution of a plurality of tasks on one or more of GPCs 3918.

In at least one embodiment, scheduler unit 3912 is coupled to work distribution unit 3914 that is configured to dispatch tasks for execution on GPCs 3918. In at least one embodiment, work distribution unit 3914 tracks a number of scheduled tasks received from scheduler unit 3912 and work distribution unit 3914 manages a pending task pool and an active task pool for each of GPCs 3918. In at least one embodiment, pending task pool comprises a number of slots (e.g., 32 slots) that contain tasks assigned to be processed by a particular GPC 3918; active task pool may comprise a number of slots (e.g., 4 slots) for tasks that are actively being processed by GPCs 3918 such that as one of GPCs 3918 completes execution of a task, that task is evicted from active task pool for GPC 3918 and one of other tasks from pending task pool is selected and scheduled for execution on GPC 3918. In at least one embodiment, if an active task is idle on GPC 3918, such as while waiting for a data dependency to be resolved, then active task is evicted from GPC 3918 and returned to pending task pool while another task in pending task pool is selected and scheduled for execution on GPC 3918.

In at least one embodiment, work distribution unit 3914 communicates with one or more GPCs 3918 via XBar 3920. In at least one embodiment, XBar 3920 is an interconnect network that couples many of units of PPU 3900 to other units of PPU 3900 and can be configured to couple work distribution unit 3914 to a particular GPC 3918. In at least one embodiment, one or more other units of PPU 3900 may also be connected to XBar 3920 via hub 3916.

In at least one embodiment, tasks are managed by scheduler unit 3912 and dispatched to one of GPCs 3918 by work distribution unit 3914. GPC 3918 is configured to process task and generate results. In at least one embodiment, results may be consumed by other tasks within GPC 3918, routed to a different GPC 3918 via XBar 3920, or stored in memory 3904. In at least one embodiment, results can be written to memory 3904 via partition units 3922, which implement a memory interface for reading and writing data to/from memory 3904. In at least one embodiment, results can be transmitted to another PPU 3904 or CPU via high-speed GPU interconnect 3908. In at least one embodiment, PPU 3900 includes, without limitation, a number U of partition units 3922 that is equal to number of separate and distinct memory devices 3904 coupled to PPU 3900. In at least one embodiment, partition unit 3922 will be described in more detail herein in conjunction with FIG. 41.

In at least one embodiment, a host processor executes a driver kernel that implements an application programming interface (“API”) that enables one or more applications executing on host processor to schedule operations for execution on PPU 3900. In at least one embodiment, multiple compute applications are simultaneously executed by PPU 3900 and PPU 3900 provides isolation, quality of service (“QoS”), and independent address spaces for multiple compute applications. In at least one embodiment, an application generates instructions (e.g., in form of API calls) that cause driver kernel to generate one or more tasks for execution by PPU 3900 and driver kernel outputs tasks to one or more streams being processed by PPU 3900. In at least one embodiment, each task comprises one or more groups of related threads, which may be referred to as a warp. In at least one embodiment, a warp comprises a plurality of related threads (e.g., 32 threads) that can be executed in parallel. In at least one embodiment, cooperating threads can refer to a plurality of threads including instructions to perform task and that exchange data through shared memory. In at least one embodiment, threads and cooperating threads are described in more detail, in accordance with at least one embodiment, in conjunction with FIG. 41.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, deep learning application processor is used to train a machine learning model, such as a neural network, to predict or infer information provided to PPU 3900. In at least one embodiment, deep learning application processor 3900 is used to infer or predict information based on a trained machine learning model (e.g., neural network) that has been trained by another processor or system or by PPU 3900. In at least one embodiment, PPU 3900 may be used to perform one or more neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 40 illustrates a general processing cluster (“GPC”) 4000, according to at least one embodiment. In at least one embodiment, GPC 4000 is GPC 3918 of FIG. 39. In at least one embodiment, each GPC 4000 includes, without limitation, a number of hardware units for processing tasks and each GPC 4000 includes, without limitation, a pipeline manager 4002, a pre-raster operations unit (“PROP”) 4004, a raster engine 4008, a work distribution crossbar (“WDX”) 4016, a memory management unit (“MMU”) 4018, one or more Data Processing Clusters (“DPCs”) 4006, and any suitable combination of parts.

In at least one embodiment, operation of GPC 4000 is controlled by pipeline manager 4002. In at least one embodiment, pipeline manager 4002 manages configuration of one or more DPCs 4006 for processing tasks allocated to GPC 4000. In at least one embodiment, pipeline manager 4002 configures at least one of one or more DPCs 4006 to implement at least a portion of a graphics rendering pipeline. In at least one embodiment, DPC 4006 is configured to execute a vertex shader program on a programmable streaming multi-processor (“SM”) 4014. In at least one embodiment, pipeline manager 4002 is configured to route packets received from a work distribution unit to appropriate logical units within GPC 4000, in at least one embodiment, and some packets may be routed to fixed function hardware units in PROP 4004 and/or raster engine 4008 while other packets may be routed to DPCs 4006 for processing by a primitive engine 4012 or SM 4014. In at least one embodiment, pipeline manager 4002 configures at least one of DPCs 4006 to implement a neural network model and/or a computing pipeline.

In at least one embodiment, PROP unit 4004 is configured, in at least one embodiment, to route data generated by raster engine 4008 and DPCs 4006 to a Raster Operations (“ROP”) unit in partition unit 3922, described in more detail above in conjunction with FIG. 39. In at least one embodiment, PROP unit 4004 is configured to perform optimizations for color blending, organize pixel data, perform address translations, and more. In at least one embodiment, raster engine 4008 includes, without limitation, a number of fixed function hardware units configured to perform various raster operations, in at least one embodiment, and raster engine 4008 includes, without limitation, a setup engine, a coarse raster engine, a culling engine, a clipping engine, a fine raster engine, a tile coalescing engine, and any suitable combination thereof. In at least one embodiment, setup engine receives transformed vertices and generates plane equations associated with geometric primitive defined by vertices; plane equations are transmitted to coarse raster engine to generate coverage information (e.g., an x, y coverage mask for a tile) for primitive; output of coarse raster engine is transmitted to culling engine where fragments associated with primitive that fail a z-test are culled, and transmitted to a clipping engine where fragments lying outside a viewing frustum are clipped. In at least one embodiment, fragments that survive clipping and culling are passed to fine raster engine to generate attributes for pixel fragments based on plane equations generated by setup engine. In at least one embodiment, output of raster engine 4008 comprises fragments to be processed by any suitable entity such as by a fragment shader implemented within DPC 4006.

In at least one embodiment, each DPC 4006 included in GPC 4000 comprise, without limitation, an M-Pipe Controller (“MPC”) 4010; primitive engine 4012; one or more SMs 4014; and any suitable combination thereof. In at least one embodiment, MPC 4010 controls operation of DPC 4006, routing packets received from pipeline manager 4002 to appropriate units in DPC 4006. In at least one embodiment, packets associated with a vertex are routed to primitive engine 4012, which is configured to fetch vertex attributes associated with vertex from memory; in contrast, packets associated with a shader program may be transmitted to SM 4014.

In at least one embodiment, SM 4014 comprises, without limitation, a programmable streaming processor that is configured to process tasks represented by a number of threads. In at least one embodiment, SM 4014 is multi-threaded and configured to execute a plurality of threads (e.g., 32 threads) from a particular group of threads concurrently and implements a Single-Instruction, Multiple-Data (“SIMD”) architecture where each thread in a group of threads (e.g., a warp) is configured to process a different set of data based on same set of instructions. In at least one embodiment, all threads in group of threads execute same instructions. In at least one embodiment, SM 4014 implements a Single-Instruction, Multiple Thread (“SIMT”) architecture wherein each thread in a group of threads is configured to process a different set of data based on same set of instructions, but where individual threads in group of threads are allowed to diverge during execution. In at least one embodiment, a program counter, call stack, and execution state is maintained for each warp, enabling concurrency between warps and serial execution within warps when threads within warp diverge. In another embodiment, a program counter, call stack, and execution state is maintained for each individual thread, enabling equal concurrency between all threads, within and between warps. In at least one embodiment, execution state is maintained for each individual thread and threads executing same instructions may be converged and executed in parallel for better efficiency. At least one embodiment of SM 4014 are described in more detail herein.

In at least one embodiment, MMU 4018 provides an interface between GPC 4000 and memory partition unit (e.g., partition unit 3922 of FIG. 39) and MMU 4018 provides translation of virtual addresses into physical addresses, memory protection, and arbitration of memory requests. In at least one embodiment, MMU 4018 provides one or more translation lookaside buffers (“TLBs”) for performing translation of virtual addresses into physical addresses in memory.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, deep learning application processor is used to train a machine learning model, such as a neural network, to predict or infer information provided to GPC 4000. In at least one embodiment, GPC 4000 is used to infer or predict information based on a trained machine learning model (e.g., neural network) that has been trained by another processor or system or by GPC 4000. In at least one embodiment, GPC 4000 may be used to perform one or more neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

FIG. 41 illustrates a memory partition unit 4100 of a parallel processing unit (“PPU”), in a4lordance with at least one embodiment. In at least one embodiment, memory partition unit 4100 includes, without limitation, a Raster Operations (“ROP”) unit 4102; a level two (“L2”) cache 4104; a memory interface 4106; and any suitable combination thereof. memory interface 4106 is coupled to memory. memory interface 4106 may implement 32, 64, 128, 1024-bit data buses, or like, for high-speed data transfer. In at least one embodiment, PPU incorporates U memory interfaces 4106, one memory interface 4106 per pair of partition units 4100, where each pair of partition units 4100 is connected to a corresponding memory device. For example, in at least one embodiment, PPU may be connected to up to Y memory devices, such as high bandwidth memory stacks or graphics double-data-rate, version 5, synchronous dynamic random a4less memory (“GDDR5 SDRAM”).

In at least one embodiment, memory interface 4106 implements a high bandwidth memory second generation (“HBM2”) memory interface and Y equals half U. In at least one embodiment, HBM2 memory stacks are located on same physical package as PPU, providing substantial power and area savings compared with conventional GDDR5 SDRAM systems. In at least one embodiment, each HBM2 stack includes, without limitation, four memory dies and Y equals 4, with each HBM2 stack including two 128-bit channels per die for a total of 8 channels and a data bus width of 1024 bits. In at least one embodiment, memory supports Single-Error Correcting Double-Error Detecting (“SECDED”) Error Correction Code (“ECC”) to protect data. ECC provides higher reliability for compute applications that are sensitive to data corruption.

In at least one embodiment, PPU implements a multi-level memory hierarchy. In at least one embodiment, memory partition unit 4100 supports a unified memory to provide a single unified virtual address space for central processing unit (“CPU”) and PPU memory, enabling data sharing between virtual memory systems. In at least one embodiment frequency of a41esses by a PPU to memory located on other processors is traced to ensure that memory pages are moved to physical memory of PPU that is a4lessing pages more frequently. In at least one embodiment, high-speed GPU interconnect 3908 supports address translation services allowing PPU to directly a4less a CPU's page tables and providing full a4less to CPU memory by PPU.

In at least one embodiment, copy engines transfer data between multiple PPUs or between PPUs and CPUs. In at least one embodiment, copy engines can generate page faults for addresses that are not mapped into page tables and memory partition unit 4100 then services page faults, mapping addresses into page table, after which copy engine performs transfer. In at least one embodiment, memory is pinned (i.e., non-pageable) for multiple copy engine operations between multiple processors, substantially reducing available memory. In at least one embodiment, with hardware page faulting, addresses can be passed to copy engines without regard as to whether memory pages are resident, and copy process is transparent.

Data from memory 3904 of FIG. 39 or other system memory is fetched by memory partition unit 4100 and stored in L2 cache 4104, which is located on-chip and is shared between various GPCs, in a4lordance with at least one embodiment. Each memory partition unit 4100, in at least one embodiment, includes, without limitation, at least a portion of L2 cache associated with a corresponding memory device. In at least one embodiment, lower level caches are implemented in various units within GPCs. In at least one embodiment, each of SMs 4014 may implement a level one (“L1”) cache wherein L1 cache is private memory that is dedicated to a particular SM 4014 and data from L2 cache 4104 is fetched and stored in each of L1 caches for processing in functional units of SMs 4014. In at least one embodiment, L2 cache 4104 is coupled to memory interface 4106 and XBar 3920.

ROP unit 4102 performs graphics raster operations related to pixel color, such as color compression, pixel blending, and more, in at least one embodiment. ROP unit 4102, in at least one embodiment, implements depth testing in conjunction with raster engine 4008, receiving a depth for a sample location associated with a pixel fragment from culling engine of raster engine 4008. In at least one embodiment, depth is tested against a corresponding depth in a depth buffer for a sample location associated with fragment. In at least one embodiment, if fragment passes depth test for sample location, then ROP unit 4102 updates depth buffer and transmits a result of depth test to raster engine 4008. It will be appreciated that number of partition units 4100 may be different than number of GPCs and, therefore, each ROP unit 4102 can, in at least one embodiment, be coupled to each of GPCs. In at least one embodiment, ROP unit 4102 tracks packets received from different GPCs and determines which that a result generated by ROP unit 4102 is routed to through XBar 3920.

FIG. 42 illustrates a streaming multi-processor (“SM”) 4200, according to at least one embodiment. In at least one embodiment, SM 4200 is SM of FIG. 40. In at least one embodiment, SM 4200 includes, without limitation, an instruction cache 4202; one or more scheduler units 4204; a register file 4208; one or more processing cores (“cores”) 4210; one or more special function units (“SFUs”) 4212; one or more load/store units (“LSUs”) 4214; an interconnect network 4216; a shared memory/level one (“L1”) cache 4218; and any suitable combination thereof. In at least one embodiment, a work distribution unit dispatches tasks for execution on general processing clusters (“GPCs”) of parallel processing units (“PPUs”) and each task is allocated to a particular Data Processing Cluster (“DPC”) within a GPC and, if task is associated with a shader program, task is allocated to one of SMs 4200. In at least one embodiment, scheduler unit 4204 receives tasks from work distribution unit and manages instruction scheduling for one or more thread blocks assigned to SM 4200. In at least one embodiment, scheduler unit 4204 schedules thread blocks for execution as warps of parallel threads, wherein each thread block is allocated at least one warp. In at least one embodiment, each warp executes threads. In at least one embodiment, scheduler unit 4204 manages a plurality of different thread blocks, allocating warps to different thread blocks and then dispatching instructions from plurality of different cooperative groups to various functional units (e.g., processing cores 4210, SFUs 4212, and LSUs 4214) during each clock cycle.

In at least one embodiment, Cooperative Groups may refer to a programming model for organizing groups of communicating threads that allows developers to express granularity at which threads are communicating, enabling expression of richer, more efficient parallel decompositions. In at least one embodiment, cooperative launch APIs support synchronization amongst thread blocks for execution of parallel algorithms. In at least one embodiment, applications of conventional programming models provide a single, simple construct for synchronizing cooperating threads: a barrier across all threads of a thread block (e.g., syncthreads( ) function). However, in at least one embodiment, programmers may define groups of threads at smaller than thread block granularities and synchronize within defined groups to enable greater performance, design flexibility, and software reuse in form of collective group-wide function interfaces. In at least one embodiment, Cooperative Groups enables programmers to define groups of threads explicitly at sub-block (i.e., as small as a single thread) and multi-block granularities, and to perform collective operations such as synchronization on threads in a cooperative group. Programming model supports clean composition across software boundaries, so that libraries and utility functions can synchronize safely within their local context without having to make assumptions about convergence. In at least one embodiment, Cooperative Groups primitives enable new patterns of cooperative parallelism, including, without limitation, producer-consumer parallelism, opportunistic parallelism, and global synchronization across an entire grid of thread blocks.

In at least one embodiment, a dispatch unit 4206 is configured to transmit instructions to one or more of functional units and scheduler unit 4204 includes, without limitation, two dispatch units 4206 that enable two different instructions from same warp to be dispatched during each clock cycle. In at least one embodiment, each scheduler unit 4204 includes a single dispatch unit 4206 or a42itional dispatch units 4206.

In at least one embodiment, each SM 4200, in at least one embodiment, includes, without limitation, register file 4208 that provides a set of registers for functional units of SM 4200. In at least one embodiment, register file 4208 is divided between each of functional units such that each functional unit is allocated a dedicated portion of register file 4208. In at least one embodiment, register file 4208 is divided between different warps being executed by SM 4200 and register file 4208 provides temporary storage for operands connected to data paths of functional units. In at least one embodiment, each SM 4200 comprises, without limitation, a plurality of L processing cores 4210. In at least one embodiment, SM 4200 includes, without limitation, a large number (e.g., 128 or more) of distinct processing cores 4210. In at least one embodiment, each processing core 4210, in at least one embodiment, includes, without limitation, a fully-pipelined, single-precision, double-precision, and/or mixed precision processing unit that includes, without limitation, a floating point arithmetic logic unit and an integer arithmetic logic unit. In at least one embodiment, floating point arithmetic logic units implement IEEE 754-2008 standard for floating point arithmetic. In at least one embodiment, processing cores 4210 include, without limitation, 64 single-precision (32-bit) floating point cores, 64 integer cores, 32 double-precision (64-bit) floating point cores, and 8 tensor cores.

Tensor cores are configured to perform matrix operations in accordance with at least one embodiment. In at least one embodiment, one or more tensor cores are included in processing cores 4210. In at least one embodiment, tensor cores are configured to perform deep learning matrix arithmetic, such as convolution operations for neural network training and inferencing. In at least one embodiment, each tensor core operates on a 4×4 matrix and performs a matrix multiply and accumulate operation D=A×B+C, where A, B, C, and D are 4×4 matrices.

In at least one embodiment, matrix multiply inputs A and Bare 16-bit floating point matrices and accumulation matrices C and D are 16-bit floating point or 32-bit floating point matrices. In at least one embodiment, tensor cores operate on 16-bit floating point input data with 32-bit floating point accumulation. In at least one embodiment, 16-bit floating point multiply uses 64 operations and results in a full precision product that is then accumulated using 32-bit floating point a42ition with other intermediate products for a 4×4×4 matrix multiply. Tensor cores are used to perform much larger two-dimensional or higher dimensional matrix operations, built up from these smaller elements, in at least one embodiment. In at least one embodiment, an API, such as CUDA 9 C++ API, exposes specialized matrix load, matrix multiply and accumulate, and matrix store operations to efficiently use tensor cores from a CUDA-C++ program. In at least one embodiment, at CUDA level, warp-level interface assumes 16×16 size matrices spanning all 32 threads of warp.

In at least one embodiment, each SM 4200 comprises, without limitation, M SFUs 4212 that perform special functions (e.g., attribute evaluation, reciprocal square root, and like). In at least one embodiment, SFUs 4212 include, without limitation, a tree traversal unit configured to traverse a hierarchical tree data structure. In at least one embodiment, SFUs 4212 include, without limitation, a texture unit configured to perform texture map filtering operations. In at least one embodiment, texture units are configured to load texture maps (e.g., a 2D array of texels) from memory and sample texture maps to produce sampled texture values for use in shader programs executed by SM 4200. In at least one embodiment, texture maps are stored in shared memory/L1 cache 4218. In at least one embodiment, texture units implement texture operations such as filtering operations using mip-maps (e.g., texture maps of varying levels of detail), in accordance with at least one embodiment. In at least one embodiment, each SM 4200 includes, without limitation, two texture units.

Each SM 4200 comprises, without limitation, N LSUs 4214 that implement load and store operations between shared memory/L1 cache 4218 and register file 4208, in at least one embodiment. Each SM 4200 includes, without limitation, interconnect network 4216 that connects each of functional units to register file 4208 and LSU 4214 to register file 4208 and shared memory/L1 cache 4218 in at least one embodiment. In at least one embodiment, interconnect network 4216 is a crossbar that can be configured to connect any of functional units to any of registers in register file 4208 and connect LSUs 4214 to register file 4208 and memory locations in shared memory/L1 cache 4218.

In at least one embodiment, shared memory/L1 cache 4218 is an array of on-chip memory that allows for data storage and communication between SM 4200 and primitive engine and between threads in SM 4200, in at least one embodiment. In at least one embodiment, shared memory/L1 cache 4218 comprises, without limitation, 128 KB of storage capacity and is in path from SM 4200 to partition unit. In at least one embodiment, shared memory/L1 cache 4218, in at least one embodiment, is used to cache reads and writes. In at least one embodiment, one or more of shared memory/L1 cache 4218, L2 cache, and memory are backing stores.

Combining data cache and shared memory functionality into a single memory block provides improved performance for both types of memory accesses, in at least one embodiment. In at least one embodiment, capacity is used or is usable as a cache by programs that do not use shared memory, such as if shared memory is configured to use half of capacity, texture and load/store operations can use remaining capacity. Integration within shared memory/L1 cache 4218 enables shared memory/L1 cache 4218 to function as a high-throughput conduit for streaming data while simultaneously providing high-bandwidth and low-latency access to frequently reused data, in accordance with at least one embodiment. In at least one embodiment, when configured for general purpose parallel computation, a simpler configuration can be used compared with graphics processing. In at least one embodiment, fixed function graphics processing units are bypassed, creating a much simpler programming model. In general purpose parallel computation configuration, work distribution unit assigns and distributes blocks of threads directly to DPCs, in at least one embodiment. In at least one embodiment, threads in a block execute same program, using a unique thread ID in calculation to ensure each thread generates unique results, using SM 4200 to execute program and perform calculations, shared memory/L1 cache 4218 to communicate between threads, and LSU 4214 to read and write global memory through shared memory/L1 cache 4218 and memory partition unit. In at least one embodiment, when configured for general purpose parallel computation, SM 4200 writes commands that scheduler unit 4204 can use to launch new work on DPCs.

In at least one embodiment, PPU is included in or coupled to a desktop computer, a laptop computer, a tablet computer, servers, supercomputers, a smart-phone (e.g., a wireless, hand-held device), personal digital assistant (“PDA”), a digital camera, a vehicle, a head mounted display, a hand-held electronic device, and more. In at least one embodiment, PPU is embodied on a single semiconductor substrate. In at least one embodiment, PPU is included in a system-on-a-chip (“SoC”) along with one or more other devices such as additional PPUs, memory, a reduced instruction set computer (“RISC”) CPU, a memory management unit (“MMU”), a digital-to-analog converter (“DAC”), and like.

In at least one embodiment, PPU may be included on a graphics card that includes one or more memory devices. Graphics card may be configured to interface with a PCIe slot on a motherboard of a desktop computer. In at least one embodiment, PPU may be an integrated graphics processing unit (“iGPU”) included in chipset of motherboard.

Inference and/or training logic 1415 are used to perform inferencing and/or training operations associated with one or more embodiments. Details regarding inference and/or training logic 1415 are provided herein in conjunction with FIGS. 14A and/or 14B. In at least one embodiment, deep learning application processor is used to train a machine learning model, such as a neural network, to predict or infer information provided to SM 4200. In at least one embodiment, SM 4200 is used to infer or predict information based on a trained machine learning model (e.g., neural network) that has been trained by another processor or system or by SM 4200. In at least one embodiment, SM 4200 may be used to perform one or more neural network use cases described herein.

At least one embodiment may be constructed using the techniques described above. In at least one embodiment, the evaluation and generation networks of a robotic control system can be implemented using a processor as described above.

In at least one embodiment, a single semiconductor platform may refer to a sole unitary semiconductor-based integrated circuit or chip. In at least one embodiment, multi-chip modules may be used with increased connectivity which simulate on-chip operation, and make substantial improvements over utilizing a conventional central processing unit (“CPU”) and bus implementation. In at least one embodiment, various modules may also be situated separately or in various combinations of semiconductor platforms per desires of user.

In at least one embodiment, computer programs in form of machine-readable executable code or computer control logic algorithms are stored in main memory 2004 and/or secondary storage. Computer programs, if executed by one or more processors, enable system 2000 to perform various functions in accordance with at least one embodiment. Memory 2004, storage, and/or any other storage are possible examples of computer-readable media. In at least one embodiment, secondary storage may refer to any suitable storage device or system such as a hard disk drive and/or a removable storage drive, representing a floppy disk drive, a magnetic tape drive, a compact disk drive, digital versatile disk (“DVD”) drive, recording device, universal serial bus (“USB”) flash memory, etc. In at least one embodiment, architecture and/or functionality of various previous figures are implemented in context of CPU 2002; parallel processing system 2012; an integrated circuit capable of at least a portion of capabilities of both CPU 2002; parallel processing system 2012; a chipset (e.g., a group of integrated circuits designed to work and sold as a unit for performing related functions, etc.); and any suitable combination of integrated circuit(s).

In at least one embodiment, architecture and/or functionality of various previous figures are implemented in context of a general computer system, a circuit board system, a game console system dedicated for entertainment purposes, an application-specific system, and more. In at least one embodiment, computer system 2000 may take form of a desktop computer, a laptop computer, a tablet computer, servers, supercomputers, a smart-phone (e.g., a wireless, hand-held device), personal digital assistant (“PDA”), a digital camera, a vehicle, a head mounted display, a hand-held electronic device, a mobile phone device, a television, workstation, game consoles, embedded system, and/or any other type of logic.

In at least one embodiment, parallel processing system 2012 includes, without limitation, a plurality of parallel processing units (“PPUs”) 2014 and associated memories 2016. In at least one embodiment, PPUs 2014 are connected to a host processor or other peripheral devices via an interconnect 2018 and a switch 2020 or multiplexer. In at least one embodiment, parallel processing system 2012 distributes computational tasks across PPUs 2014 which can be parallelizable for example, as part of distribution of computational tasks across multiple graphics processing unit (“GPU”) thread blocks. In at least one embodiment, memory is shared and accessible (e.g., for read and/or write access) across some or all of PPUs 2014, although such shared memory may incur performance penalties relative to use of local memory and registers resident to a PPU 2014. In at least one embodiment, operation of PPUs 2014 is synchronized through use of a command such as_syncthreadso, wherein all threads in a block (e.g., executed across multiple PPUs 2014) to reach a certain point of execution of code before proc20ding.

Other variations are within spirit of present disclosure. Thus, while disclosed techniques are susceptible to various modifications and alternative constructions, certain illustrated embodiments thereof are shown in drawings and have been described above in detail. It should be understood, however, that there is no intention to limit disclosure to specific form or forms disclosed, but on contrary, intention is to cover all modifications, alternative constructions, and equivalents falling within spirit and scope of disclosure, as defined in appended claims.

Use of terms “a” and “an” and “the” and similar referents in context of describing disclosed embodiments (especially in context of following claims) are to be construed to cover both singular and plural, unless otherwise indicated herein or clearly contradicted by context, and not as a definition of a term. Terms “comprising,” “having,” “including,” and “containing” are to be construed as open-ended terms (meaning “including, but not limited to,”) unless otherwise noted. term “connected,” when unmodified and referring to physical connections, is to be construed as partly or wholly contained within, attached to, or joined together, even if there is something intervening. Recitation of ranges of values herein are merely intended to serve as a shorthand method of referring individually to each separate value falling within range, unless otherwise indicated herein and each separate value is incorporated into specification as if it were individually recited herein. Use of term “set” (e.g., “a set of items”) or “subset” unless otherwise noted or contradicted by context, is to be construed as a nonempty collection comprising one or more members. Further, unless otherwise noted or contradicted by context, term “subset” of a corresponding set does not necessarily denote a proper subset of corresponding set, but subset and corresponding set may be equal.

Conjunctive language, such as phrases of form “at least one of A, B, and C,” or “at least one of A, B and C,” unless specifically stated otherwise or otherwise clearly contradicted by context, is otherwise understood with context as used in general to present that an item, term, etc., may be either A or B or C, or any nonempty subset of set of A and B and C. For instance, in illustrative example of a set having three members, conjunctive phrases “at least one of A, B, and C” and “at least one of A, Band C” refer to any of following sets: {A}, {B}, {C}, {A, B}, {A, C}, {B, C}, {A, B, C}. Thus, such conjunctive language is not generally intended to imply that certain embodiments require at least one of A, at least one of B and at least one of C each to be present. In addition, unless otherwise noted or contradicted by context, term “plurality” indicates a state of being plural (e.g., “a plurality of items” indicates multiple items). Number of items in a plurality is at least two, but can be more when so indicated either explicitly or by context. Further, unless stated otherwise or otherwise clear from context, phrase “based on” means “based at least in part on” and not “based solely on.”

Operations of processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. In at least one embodiment, a process such as those processes described herein (or variations and/or combinations thereof) is performed under control of one or more computer systems configured with executable instructions and is implemented as code (e.g., executable instructions, one or more computer programs or one or more applications) executing collectively on one or more processors, by hardware or combinations thereof. In at least one embodiment, code is stored on a computer-readable storage medium, for example, in form of a computer program comprising a plurality of instructions executable by one or more processors. In at least one embodiment, a computer-readable storage medium is a non-transitory computer-readable storage medium that excludes transitory signals (e.g., a propagating transient electric or electromagnetic transmission) but includes non-transitory data storage circuitry (e.g., buffers, cache, and queues) within transceivers of transitory signals. In at least one embodiment, code (e.g., executable code or source code) is stored on a set of one or more non-transitory computer-readable storage media having stored thereon executable instructions (or other memory to store executable instructions) that, when executed (i.e., as a result of being executed) by one or more processors of a computer system, cause computer system to perform operations described herein. Set of non-transitory computer-readable storage media, in at least one embodiment, comprises multiple non-transitory computer-readable storage media and one or more of individual non-transitory storage media of multiple non-transitory computer-readable storage media lack all of code while multiple non-transitory computer-readable storage media collectively store all of code. In at least one embodiment, executable instructions are executed such that different instructions are executed by different processors for example, a non-transitory computer-readable storage medium store instructions and a main central processing unit (“CPU”) executes some of instructions while a graphics processing unit (“GPU”) executes other instructions. In at least one embodiment, different components of a computer system have separate processors and different processors execute different subsets of instructions.

Accordingly, in at least one embodiment, computer systems are configured to implement one or more services that singly or collectively perform operations of processes described herein and such computer systems are configured with applicable hardware and/or software that enable performance of operations. Further, a computer system that implements at least one embodiment of present disclosure is a single device and, in another embodiment, is a distributed computer system comprising multiple devices that operate differently such that distributed computer system performs operations described herein and such that a single device does not perform all operations.

Use of any and all examples, or exemplary language (e.g., “such as”) provided herein, is intended merely to better illuminate embodiments of disclosure and does not pose a limitation on scope of disclosure unless otherwise claimed. No language in specification should be construed as indicating any non-claimed element as essential to practice of disclosure.

All references, including publications, patent applications, and patents, cited herein are hereby incorporated by reference to same extent as if each reference were individually and specifically indicated to be incorporated by reference and were set forth in its entirety herein.

In description and claims, terms “coupled” and “connected,” along with their derivatives, may be used. It should be understood that these terms may be not intended as synonyms for each other. Rather, in particular examples, “connected” or “coupled” may be used to indicate that two or more elements are in direct or indirect physical or electrical contact with each other. “Coupled” may also mean that two or more elements are not in direct contact with each other, but yet still co-operate or interact with each other.

Unless specifically stated otherwise, it may be appreciated that throughout specification terms such as “processing,” “computing,” “calculating,” “determining,” or like, refer to action and/or processes of a computer or computing system, or similar electronic computing device, that manipulate and/or transform data represented as physical, such as electronic, quantities within computing system's registers and/or memories into other data similarly represented as physical quantities within computing system's memories, registers or other such information storage, transmission or display devices.

Ina similar manner, term “processor” may refer to any device or portion of a device that processes electronic data from registers and/or memory and transform that electronic data into other electronic data that may be stored in registers and/or memory. As non-limiting examples, “processor” may be a CPU or a GPU. A “computing platform” may comprise one or more processors. As used herein, “software” processes may include, for example, software and/or hardware entities that perform work over time, such as tasks, threads, and intelligent agents. Also, each process may refer to multiple processes, for carrying out instructions in sequence or in parallel, continuously or intermittently. Terms “system” and “method” are used herein interchangeably insofar as system may embody one or more methods and methods may be considered a system.

In present document, references may be made to obtaining, acquiring, receiving, or inputting analog or digital data into a subsystem, computer system, or computer-implemented machine. Process of obtaining, acquiring, receiving, or inputting analog and digital data can be accomplished in a variety of ways such as by receiving data as a parameter of a function call or a call to an application programming interface. In some implementations, process of obtaining, acquiring, receiving, or inputting analog or digital data can be accomplished by transferring data via a serial or parallel interface. In another implementation, process of obtaining, acquiring, receiving, or inputting analog or digital data can be accomplished by transferring data via a computer network from providing entity to acquiring entity. References may also be made to providing, outputting, transmitting, sending, or presenting analog or digital data. In various examples, process of providing, outputting, transmitting, sending, or presenting analog or digital data can be accomplished by transferring data as an input or output parameter of a function call, a parameter of an application programming interface or interprocess communication mechanism.

Although discussion above sets forth example implementations of described techniques, other architectures may be used to implement described functionality, and are intended to be within scope of this disclosure. Furthermore, although specific distributions of responsibilities are defined above for purposes of discussion, various functions and responsibilities might be distributed and divided in different ways, depending on circumstances.

Furthermore, although subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that subject matter claimed in appended claims is not necessarily limited to specific features or acts described. Rather, specific features and acts are disclosed as exemplary forms of implementing the claims.

Claims

1. A computer-implemented method comprising: obtaining tactile sensor information from a robotic appendage that is manipulating an object in the real world;generating a plurality of simulations of the robotic appendage manipulating the object, individual simulations having different poses for the object;determine a plurality of costs where each cost of the plurality of costs corresponds to a respective simulation of the plurality of simulations, and each cost of the plurality of costs is based at least in part on differences between the tactile sensor information and simulated tactile sensor information generated by the respective simulation of the plurality of simulations;identifying an individual simulation of the plurality of simulations based at least in part on the cost;determining a pose of the object in the real world based at least in part on a pose of the object in the identified individual simulation; andproviding the pose of the object to a robotic control system that controls a robot to perform a task based at least in part on the pose of the object.

2. The computer-implemented method of claim 1, further comprising updating one or more physical parameters of the plurality of simulations to reduce a difference between a simulation and an observation in the real world.

3. The computer-implemented method of claim 1, wherein the plurality of simulations are implemented using a GPU-Accelerated physics simulator.

4. The computer-implemented method of claim 1, wherein the individual simulation is identified by identifying an individual simulation that generates simulated tactile sensor information most similar to the tactile sensor information.

5. The computer-implemented method of claim 1, further comprising: obtaining an initial pose estimation of the object; andgenerating a plurality of possible poses to be applied to the plurality of simulations, the plurality of possible poses generated by modifying the initial pose estimation.

6. The computer-implemented method of claim 5, wherein the initial pose estimation of the object is determined based on an image of the object obtained before the object is grasped by the robotic appendage.

7. The computer-implemented method of claim 1, wherein the individual simulation is identified by identifying an individual simulation with a lowest associated cost.

8. The computer-implemented method of claim 1, wherein the tactile sensor information includes a 2-dimensional array of force values for each digit of the robotic appendage.

9. A system comprising: one or more processors; andcomputer-readable memory storing executable instructions that, as a result of being executed by the one or more processors, cause the system to: obtain data describing forces on a robotic appendage that is grasping an object;generate simulations of the robotic appendage grasping the object, individual simulations of the simulations having different poses for the object;determine a plurality of values where each value of the plurality of values corresponds to a respective simulation of the plurality of simulations, and each value of the plurality of values is based at least in part on differences between the forces and simulated tactile simulated forces generated by the respective simulation of the plurality of simulations;identify an individual simulation of the simulations based at least in part on the value; anddetermine a pose of the object in the real world based at least in part on a pose of the object in the identified individual simulation.

10. The system of claim 9, wherein the executable instructions cause the system to further update one or more physical parameters of the simulations to reduce a difference between a state of a simulation and an observed state in the real world.

11. The system of claim 9, wherein the one or more processors include a graphics processing unit.

12. The system of claim 9, wherein the individual simulation is identified by identifying an individual simulation that generates simulated data most closely corresponding to the data.

13. The system of claim 9, wherein the executable instructions cause the system to further: obtain an initial pose of the object; andgenerate a plurality of poses to be applied to objects in the simulations, the plurality of possible poses generated by perturbing the initial pose.

14. The system of claim 13, wherein the initial pose of the object is determined using an image of the object obtained before the object is grasped by the robotic appendage.

15. The system of claim 9, wherein: the value is a measure of difference between the forces and the simulated forces; andthe individual simulation is identified by identifying an individual simulation with a lowest associated value.

16. The system of claim 9, wherein the data is tactile sensor information generated by tactile force sensor on each digit of the robotic appendage.

17. Computer-readable media storing instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to: obtain data describing forces on a robotic appendage that is grasping an object;perform simulations of the robotic appendage grasping the object, individual simulations of the simulations having different poses for the object;determine a plurality of values where each value of the plurality of values corresponds to a respective simulation of the plurality of simulations, and each value of the plurality of values is based at least in part on differences between the forces and simulated tactile simulated forces generated by the respective simulation of the plurality of simulations;identify an individual simulation of the simulations based at least in part on the value; anddetermine a pose of the object in the real world based at least in part on a pose of the object in the identified individual simulation.

18. The computer-readable media of claim 17, wherein the instructions cause the computer system to further update one or more parameters of the simulations to reduce a difference between a state of a simulation and an observed state in the real world.

19. The computer-readable media of claim 17, wherein the one or more processors include a multi-core graphics processing unit.

20. The computer-readable media of claim 17, wherein the simulations are performed in parallel using a plurality of processors.

21. The computer-readable media of claim 17, wherein the instructions cause the computer system to further: obtain an initial pose of the object; andgenerate a plurality of poses to be applied to objects in the simulations, the plurality of possible poses generated by perturbing the initial pose.

22. The computer-readable media of claim 21, wherein the initial pose of the object is determined using an image of the object obtained with a depth camera.

23. The computer-readable media of claim 17, wherein: the value is a measure of difference between the forces and the simulated forces; andthe individual simulation is identified by identifying an individual simulation with a lowest associated value.

24. The computer-readable media of claim 17, wherein the data is tactile sensor information generated by tactile force sensor on the robotic appendage.

25. A robot comprising: an arm that includes one or more articulated members connected via one or more servo motors;a robotic appendage connected to the arm, the robotic appendage having one or more tactile force sensors;one or more processors; andthe computer-readable media of claim 17 connected to the one or more processors.