The disclosure generally relates to estimating heights of aerial robots and, more specifically, to robots that use different sensors to estimate heights accurately.
For aerial robots such as drones to be autonomous, aerial robots need to navigate through the environment without colliding with objects. Estimating the height of the robot at any time instance is important for the robot's navigation and collision avoidance, especially in an indoor setting. Conventionally, an aerial robot may be equipped with a barometer to determine the pressure change in various altitudes in order for the aerial robot to estimate the height. However, the measurements obtained from the barometer are often not sensitive enough to produce highly accurate height estimates. Also, pressure change in an indoor setting is either insufficiently significant or even unmeasurable. Hence, estimating heights for aerial robots can be challenging.
Embodiments relate to an aerial robot that may include a distance sensor and visual inertial sensor. Embodiments also related to a method for the robot to perform height estimates using the distance sensor and the visual inertial sensor. The method may include determining a first height estimate of the aerial robot relative to a first region with a first surface level using data from a distance sensor of the aerial robot. The method may also include controlling the flight of the aerial robot over at least a part of the first region based on the first estimated height. The method may further include determining that the aerial robot is in a transition region between the first region and a second region with a second surface level different from the first surface level. The method may further include determining a second height estimate of the aerial robot using data from a visual inertial sensor of the aerial robot. The method may further include controlling the flight of the aerial robot using the second height estimate in the transition region. The aerial robot may include one or more processors and memory for storing instructions for performing the height estimate method.
The figures depict, and the detailed description describes, various non-limiting embodiments for purposes of illustration only.
The figures (FIGs.) and the following description relate to preferred embodiments by way of illustration only. One of skill in the art may recognize alternative embodiments of the structures and methods disclosed herein as viable alternatives that may be employed without departing from the principles of what is disclosed.
Reference will now be made in detail to several embodiments, examples of which are illustrated in the accompanying figures. It is noted that wherever practicable similar or like reference numbers may be used in the figures and may indicate similar or like functionality. The figures depict embodiments of the disclosed system (or method) for purposes of illustration only. One skilled in the art will readily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles described herein.
Embodiments relate to an aerial robot that navigates an environment with a level flight by accurately estimating the height of the robot using a combination of a distance sensor and a visual inertial sensor. The distance sensor and the visual inertial sensor may use different methods to estimate heights. Data generated from the two sensors may be used to compensate each other to provide an accurate height estimate. In some embodiments, the aerial robot may use the distance sensor to estimate the heights when the aerial robot travels over leveled surfaces. The aerial robot may also monitor the bias between the data from the two different sensors. At a transition region between two leveled surfaces, the aerial robot may switch to the visual inertial sensor. The aerial robot may adjust the data from the visual inertial sensor using the monitored biased.
A storage site 110 may be any suitable facility that stores, sells, or displays inventories such as goods, merchandise, groceries, articles and collections. Example storage sites 110 may include warehouses, inventory sites, bookstores, shoe stores, outlets, other retail stores, libraries, museums, etc. A storage site 110 may include a number of regularly shaped structures. Regularly shaped structures may be structures, fixtures, equipment, furniture, frames, shells, racks, or other suitable things in the storage site 110 that have a regular shape or outline that can be readily identifiable, whether the things are permanent or temporary, fixed or movable, weight-bearing or not. The regularly shaped structures are often used in a storage site 110 for storage of inventory. For example, racks (including metallic racks, shells, frames, or other similar structures) are often used in a warehouse for the storage of goods and merchandise. However, not all regularly shaped structures may need to be used for inventory storage. A storage site 110 may include a certain layout that allows various items to be placed and stored systematically. For example, in a warehouse, the racks may be grouped by sections and separated by aisles. Each rack may include multiple pallet locations that can be identified using a row number and a column number. A storage site may include high racks and low racks, which may, in some case, largely carry most of the inventory items near the ground level.
A storage site 110 may include one or more robots 120 that are used to keep track of the inventory and to manage the inventory in the storage site 110. For the ease of reference, the robot 120 may be referred to in a singular form, even though more than one robot 120 may be used. Also, in some embodiments, there can be more than one type of robot 120 in a storage site 110. For example, some robots 120 may specialize in scanning inventory in the storage site 110, while other robots 120 may specialize in moving items. A robot 120 may also be referred to as an autonomous robot, an inventory cycle-counting robot, an inventory survey robot, an inventory detection robot, or an inventory management robot. An inventory robot may be used to track inventory items, move inventory items, and carry out other inventory management tasks. The degree of autonomy may vary from embodiments to embodiments. For example, in some embodiments, the robot 120 may be fully autonomous so that the robot 120 automatically performs assigned tasks. In another embodiment, the robot 120 may be semi-autonomous such that it can navigate through the storage site 110 with minimal human commands or controls. In some embodiments, no matter what the degree of autonomy it has, a robot 120 may also be controlled remotely and may be switched to a manual mode. The robot 120 may take various forms such as an aerial drone, a ground robot, a vehicle, a forklift, and a mobile picking robot.
A base station 130 may be a device for the robot 120 to return and, for an aerial robot, to land. The base station 130 may include more than one return site. The base station 130 may be used to repower the robot 120. Various ways to repower the robot 120 may be used in different embodiments. For example, in some embodiments, the base station 130 serves as a battery-swapping station that exchanges batteries on a robot 120 as the robot arrives at the base station to allow the robot 120 to quickly resume duty. The replaced batteries may be charged at the base station 130, wired or wirelessly. In another embodiment, the base station 130 serves as a charging station that has one or more charging terminals to be coupled to the charging terminal of the robot 120 to recharge the batteries of the robot 120. In yet another embodiment, the robot 120 may use fuel for power and the base station 130 may repower the robot 120 by filling its fuel tank.
The base station 130 may also serve as a communication station for the robot 120. For example, for certain types of storage sites 110 such as warehouses, network coverage may not be present or may only be present at certain locations. The base station 130 may communicate with other components in the system environment 100 using wireless or wired communication channels such as Wi-Fi or an Ethernet cable. The robot 120 may communicate with the base station 130 when the robot 120 returns to the base station 130. The base station 130 may send inputs such as commands to the robot 120 and download data captured by the robot 120. In embodiments where multiple robots 120 are used, the base station 130 may be equipped with a swarm control unit or algorithm to coordinate the movements among the robots. The base station 130 and the robot 120 may communicate in any suitable ways such as radio frequency, Bluetooth, near-field communication (NFC), or wired communication. While, in some embodiments, the robot 120 mainly communicates to the base station, in other embodiments the robot 120 may also have the capability to directly communicate with other components in the system environment 100. In some embodiments, the base station 130 may serve as a wireless signal amplifier for the robot 120 to directly communicate with the network 180.
The inventory management system 140 may be a computing system that is operated by the administrator (e.g., a company that owns the inventory, a warehouse management administrator, a retailer selling the inventory) using the storage site 110. The inventory management system 140 may be a system used to manage the inventory items. The inventory management system 140 may include a database that stores data regarding inventory items and the items' associated information, such as quantities in the storage site 110, metadata tags, asset type tags, barcode labels and location coordinates of the items. The inventory management system 140 may provide both front-end and back-end software for the administrator to access a central database and point of reference for the inventory and to analyze data, generate reports, forecast future demands, and manage the locations of the inventory items to ensure items are correctly placed. An administrator may rely on the item coordinate data in the inventory management system 140 to ensure that items are correctly placed in the storage site 110 so that the items can be readily retrieved from a storage location. This prevents an incorrectly placed item from occupying a space that is reserved for an incoming item and also reduces time to locate a missing item at an outbound process.
The computing server 150 may be a server that is tasked with analyzing data provided by the robot 120 and provide commands for the robot 120 to perform various inventory recognition and management tasks. The robot 120 may be controlled by the computing server 150, the user device 170, or the inventory management system 140. For example, the computing server 150 may direct the robot 120 to scan and capture pictures of inventory stored at various locations at the storage site 110. Based on the data provided by the inventory management system 140 and the ground truth data captured by the robot 120, the computing server 150 may identify discrepancies in two sets of data and determine whether any items may be misplaced, lost, damaged, or otherwise should be flagged for various reasons. In turn, the computing server 150 may direct a robot 120 to remedy any potential issues such as moving a misplaced item to the correct position. In some embodiments, the computing server 150 may also generate a report of flagged items to allow site personnel to manually correct the issues.
The computing server 150 may include one or more computing devices that operate at different locations. For example, a part of the computing server 150 may be a local server that is located at the storage site 110. The computing hardware such as the processor may be associated with a computer on site or may be included in the base station 130. Another part of the computing server 150 may be a cloud server that is geographically distributed. The computing server 150 may serve as a ground control station (GCS), provide data processing, and maintain end-user software that may be used in a user device 170. A GCS may be responsible for the control, monitor and maintenance of the robot 120. In some embodiments, GCS is located on-site as part of the base station 130. The data processing pipeline and end-user software server may be located remotely or on-site.
The computing server 150 may maintain software applications for users to manage the inventory, the base station 130, and the robot 120. The computing server 150 and the inventory management system 140 may or may not be operated by the same entity. In some embodiments, the computing server 150 may be operated by an entity separated from the administrator of the storage site. For example, the computing server 150 may be operated by a robotic service provider that supplies the robot 120 and related systems to modernize and automate a storage site 110. The software application provided by the computing server 150 may take several forms. In some embodiments, the software application may be integrated with or as an add-on to the inventory management system 140. In another embodiment, the software application may be a separate application that supplements or replaces the inventory management system 140. In some embodiments, the software application may be provided as software as a service (SaaS) to the administrator of the storage site 110 by the robotic service provider that supplies the robot 120.
The data store 160 includes one or more storage units such as memory that takes the form of non-transitory and non-volatile computer storage medium to store various data that may be uploaded by the robot 120 and inventory management system 140. For example, the data stored in data store 160 may include pictures, sensor data, and other data captured by the robot 120. The data may also include inventory data that is maintained by the inventory management system 140. The computer-readable storage medium is a medium that does not include a transitory medium such as a propagating signal or a carrier wave. The data store 160 may take various forms. In some embodiments, the data store 160 communicates with other components by the network 180. This type of data store 160 may be referred to as a cloud storage server. Example cloud storage service providers may include AWS, AZURE STORAGE, GOOGLE CLOUD STORAGE, etc. In another embodiment, instead of a cloud storage server, the data store 160 is a storage device that is controlled and connected to the computing server 150. For example, the data store 160 may take the form of memory (e.g., hard drives, flash memories, discs, ROMs, etc.) used by the computing server 150 such as storage devices in a storage server room that is operated by the computing server 150.
The user device 170 may be used by an administrator of the storage site 110 to provide commands to the robot 120 and to manage the inventory in the storage site 110. For example, using the user device 170, the administrator can provide task commands to the robot 120 for the robot to automatically complete the tasks. In one case, the administrator can specify a specific target location or a range of storage locations for the robot 120 to scan. The administrator may also specify a specific item for the robot 120 to locate or to confirm placement. Examples of user devices 170 include personal computers (PCs), desktop computers, laptop computers, tablet computers, smartphones, wearable electronic devices such as smartwatches, or any other suitable electronic devices.
The user device 170 may include a user interface 175, which may take the form of a graphical user interface (GUI). Software application provided by the computing server 150 or the inventory management system 140 may be displayed as the user interface 175. The user interface 175 may take different forms. In some embodiments, the user interface 175 is part of a front-end software application that includes a GUI displayed at the user device 170. In one case, the front-end software application is a software application that can be downloaded and installed at user devices 170 via, for example, an application store (e.g., App Store) of the user device 170. In another case, the user interface 175 takes the form of a Web interface of the computing server 150 or the inventory management system 140 that allows clients to perform actions through web browsers. In another embodiment, user interface 175 does not include graphical elements but communicates with the computing server 150 or the inventory management system 140 via other suitable ways such as command windows or application program interfaces (APIs).
The communications among the robot 120, the base station 130, the inventory management system 140, the computing server 150, the data store 160, and the user device 170 may be transmitted via a network 180, for example, via the Internet. In some embodiments, the network 180 uses standard communication technologies and/or protocols. Thus, the network 180 can include links using technologies such as Ethernet, 802.11, worldwide interoperability for microwave access (WiMAX), 3G, 4G, LTE, 5G, digital subscriber line (DSL), asynchronous transfer mode (ATM), InfiniBand, PCI Express, etc. Similarly, the networking protocols used on the network 180 can include multiprotocol label switching (MPLS), the transmission control protocol/Internet protocol (TCP/IP), the user datagram protocol (UDP), the hypertext transport protocol (HTTP), the simple mail transfer protocol (SMTP), the file transfer protocol (FTP), etc. The data exchanged over the network 180 can be represented using technologies and/or formats including the hypertext markup language (HTML), the extensible markup language (XML), etc. In addition, all or some of the links can be encrypted using conventional encryption technologies such as secure sockets layer (SSL), transport layer security (TLS), virtual private networks (VPNs), Internet protocol security (IPsec), etc. The network 180 also includes links and packet switching networks such as the Internet. In some embodiments, two computing servers, such as computing server 150 and inventory management system 140, may communicate through APIs. For example, the computing server 150 may retrieve inventory data from the inventory management system 140 via an API.
The image sensor 210 captures images of an environment of a storage site for navigation, localization, collision avoidance, object recognition and identification, and inventory recognition purposes. A robot 120 may include more than one image sensors 210 and more than one type of such image sensors 210. For example, the robot 120 may include a digital camera that captures optical images of the environment for the state estimator 235. For example, data captured by the image sensor 210 may also be provided to the VIO unit 236 that may be included in the state estimator 235 for localization purposes such as to determine the position and orientation of the robot 120 with respect to an inertial frame, such as a global frame whose location is known and fixed. The robot 120 may also include a stereo camera that includes two or more lenses to allow the image sensor 210 to capture three-dimensional images through stereoscopic photography. For each image frame, the stereo camera may generate pixel values such as in red, green, and blue (RGB) and point cloud data that includes depth information. The images captured by the stereo camera may be provided to visual reference engine 240 for object recognition purposes. The image sensor 210 may also be another type of image sensor such as a light detection and ranging (LIDAR) sensor, an infrared camera, and 360-degree depth cameras. The image sensor 210 may also capture pictures of labels (e.g., barcodes) on items for inventory cycle-counting purposes. In some embodiments, a single stereo camera may be used for various purposes. For example, the stereo camera may provide image data to the visual reference engine 240 for object recognition. The stereo camera may also be used to capture pictures of labels (e.g., barcodes). In some embodiments, the robot 120 includes a rotational mount such as a gimbal that allows the image sensor 210 to rotate in different angles and to stabilize images captured by the image sensor 210. In some embodiments, the image sensor 210 may also capture data along the path for the purpose of mapping the storage site.
The robot 120 includes one or more processors 215 and one or more memories 220 that store one or more sets of instructions. The one or more sets of instructions, when executed by one or more processors, cause the one or more processors to carry out processes that are implemented as one or more software engines. Various components, such as FCU 225 and state estimator 235, of the robot 120 may be implemented as a combination of software and hardware (e.g., sensors). The robot 120 may use a single general processor to execute various software engines or may use separate more specialized processors for different functionalities. In some embodiments, the robot 120 may use a general-purpose computer (e.g., a CPU) that can execute various instruction sets for various components (e.g., FCU 225, visual reference engine 240, state estimator 235, planner 250). The general-purpose computer may run on a suitable operating system such as LINUX, ANDROID, etc. For example, in some embodiments, the robot 120 may carry a smartphone that includes an application used to control the robot. In another embodiment, the robot 120 includes multiple processors that are specialized in different functionalities. For example, some of the functional components such as FCU 225, visual reference engine 240, state estimator 235, and planner 250 may be modularized and each includes its own processor, memory, and a set of instructions. The robot 120 may include a central processor unit (CPU) to coordinate and communicate with each modularized component. Hence, depending on embodiments, a robot 120 may include a single processor or multiple processors 215 to carry out various operations. The memory 220 may also store images and videos captured by the image sensor 210. The images may include images that capture the surrounding environment and images of the inventory such as barcodes and labels.
The flight control unit (FCU) 225 may be a combination of software and hardware, such as inertial measurement unit (IMU) 230 and other sensors, to control the movement of the robot 120. For ground robot 120, the flight control unit 225 may also be referred to as a microcontroller unit (MCU). The FCU 225 relies on information provided by other components to control the movement of the robot 120. For example, the planner 250 determines the path of the robot 120 from a starting point to a destination and provides commands to the FCU 225. Based on the commands, the FCU 225 generates electrical signals to various mechanical parts (e.g., actuators, motors, engines, wheels) of the robot 120 to adjust the movement of the robot 120. The precise mechanical parts of the robots 120 may depend on the embodiments and the types of robots 120.
The IMU 230 may be part of the FCU 225 or may be an independent component. The IMU 230 may include one or more accelerometers, gyroscopes, and other suitable sensors to generate measurements of forces, linear accelerations, and rotations of the robot 120. For example, the accelerometers measure the force exerted on the robot 120 and detect the linear acceleration. Multiple accelerometers cooperate to detect the acceleration of the robot 120 in the three-dimensional space. For instance, a first accelerometer detects the acceleration in the x-direction, a second accelerometer detects the acceleration in the y-direction, and a third accelerometer detects the acceleration in the z-direction. The gyroscopes detect the rotations and angular acceleration of the robot 120. Based on the measurements, a processor 215 may obtain the estimated localization of the robot 120 by integrating the translation and rotation data of the IMU 230 with respect to time. The IMU 230 may also measure the orientation of the robot 120. For example, the gyroscopes in the IMU 230 may provide readings of the pitch angle, the roll angle, and the yaw angle of the robot 120.
The state estimator 235 may correspond to a set of software instructions stored in the memory 220 that can be executed by the processor 215. The state estimator 235 may be used to generate localization information of the robot 120 and may include various sub-components for estimating the state of the robot 120. For example, in some embodiments, the state estimator 235 may include a visual-inertial odometry (VIO) unit 236 and an height estimator 238. In other embodiments, other modules, sensors, and algorithms may also be used in the state estimator 235 to determine the location of the robot 120.
The VIO unit 236 receives image data from the image sensor 210 (e.g., a stereo camera) and measurements from IMU 230 to generate localization information such as the position and orientation of the robot 120. The localization data obtained from the double integration of the acceleration measurements from the IMU 230 is often prone to drift errors. The VIO unit 236 may extract image feature points and tracks the feature points in the image sequence to generate optical flow vectors that represent the movement of edges, boundaries, surfaces of objects in the environment captured by the image sensor 210. Various signal processing techniques such as filtering (e.g., Wiener filter, Kalman filter, bandpass filter, particle filter) and optimization, and data/image transformation may be used to reduce various errors in determining localization information. The localization data generated by the VIO unit 236 may include an estimate of the pose of the robot 120, which may be expressed in terms of the roll angle, the pitch angle, and the yaw angle of the robot 120.
The height estimator 238 may be a combination of software and hardware that are used to determine the absolute height and relative height (e.g., distance from an object that lies on the floor) of the robot 120. The height estimator 238 may include a downward distance sensor 239 that may measure the height relative to the ground or to an object underneath the robot 120. The distance sensor 239 may be electromagnetic wave based, laser based, optics based, sonar based, ultrasonic based, or another suitable signal based. For example, the distance sensor 239 may be a laser range finder, a lidar range finder, a sonar range finder, an ultrasonic range finder, or a radar. A range finder may include one or more emitters that emit signals (e.g., infrared, laser, sonar, etc.) and one or more sensors that detect the round trip time of the signal reflected by an object. In some embodiments, the robot 120 may be equipped with a single emitter range finder. The height estimator 238 may also receive data from the VIO unit 236 that may estimate the height of the robot 120, but usually in a less accurate fashion compared to a distance sensor 239. The height estimator 238 may include software algorithms to combine data generated by the distance sensor 239 and the data generated by the VIO unit 236 as the robot 120 flies over various objects and inventory that are placed on the floor or other horizontal levels. The data generated by the height estimator 238 may be used for collision avoidance and finding a target location. The height estimator 238 may set a global maximum altitude to prevent the robot 120 from hitting the ceiling. The height estimator 238 also provides information regarding how many rows in the rack are below the robot 120 for the robot 120 to locate a target location. The height data may be used in conjunction with the count of rows that the robot 120 has passed to determine the vertical level of the robot 120. The height estimation will be discussed in further detail with reference to
The visual reference engine 240 may correspond to a set of software instructions stored in the memory 220 that can be executed by the processor 215. The visual reference engine 240 may include various image processing algorithm and location algorithm to determine the current location of the robot 120, to identify the objects, edges, and surfaces of the environment near the robot 120, and to determine an estimated distance and orientation (e.g., yaw) of the robot 120 relative to a nearby surface of an object. The visual reference engine 240 may receive pixel data of a series of images and point cloud data from the image sensor 210. The location information generated by the visual reference engine 240 may include distance and yaw from an object and center offset from a target point (e.g., a midpoint of a target object).
The visual reference engine 240 may include one or more algorithms and machine learning models to create image segmentations from the images captured by the image sensor 210. The image segmentation may include one or more segments that separate the frames (e.g., vertical or horizontal bars of racks) or outlines of regularly shaped structures appearing in the captured images from other objects and environments. The algorithms used for image segmentation may include a convolutional neural network (CNN). In performing the segmentation, other image segmentation algorithms such as edge detection algorithms (e.g., Canny operator, Laplacian operator, Sobel operator, Prewitt operator), corner detection algorithms, Hough transform, and other suitable feature detection algorithms may also be used.
The visual reference engine 240 also performs object recognition (e.g., object detection and further analyses) and keeps track of the relative movements of the objects across a series of images. The visual reference engine 240 may track the number of regularly shaped structures in the storage site 110 that are passed by the robot 120. For example, the visual reference engine 240 may identify a reference point (e.g., centroid) of a frame of a rack and determine if the reference point passes a certain location of the images across a series of images (e.g., whether the reference point passes the center of the images). If so, the visual reference engine 240 increments the number of regularly shaped structures that have been passed by the robot 120.
The robot 120 may use various components to generate various types of location information (including location information relative to nearby objects and localization information). For example, in some embodiments, the state estimator 235 may process the data from the VIO unit 236 and the height estimator 238 to provide localization information to the planner 250. The visual reference engine 240 may count the number of regularly shaped structures that the robot 120 has passed to determine a current location. The visual reference engine 240 may generate location information relative to nearby objects. For example, when the robot 120 reaches a target location of a rack, the visual reference engine 240 may use point cloud data to reconstruct a surface of the rack and use the depth data from the point cloud to determine more accurate yaw and distance between the robot 120 and the rack. The visual reference engine 240 may determine a center offset, which may correspond to the distance between the robot 120 and the center of a target location (e.g., the midpoint of a target location of a rack). Using the center offset information, the planner 250 controls the robot 120 to move to the target location and take a picture of the inventory in the target location. When the robot 120 changes direction (e.g., rotations, transitions from horizontal movement to vertical movement, transitions from vertical movement to horizontal movement, etc.), the center offset information may be used to determine the accurate location of the robot 120 relative to an object.
The planner 250 may correspond to a set of software instructions stored in the memory 220 that can be executed by the processor 215. The planner 250 may include various routing algorithms to plan a path of the robot 120 as the robot travels from a first location (e.g., a starting location, the current location of the robot 120 after finishing the previous journey) to a second location (e.g., a target destination). The robot 120 may receive inputs such as user commands to perform certain actions (e.g., scanning of inventory, moving an item, etc.) at certain locations. The planner 250 may include two types of routes, which corresponds to a spot check and a range scan. In a spot check, the planner 250 may receive an input that includes coordinates of one or more specific target locations. In response, the planner 250 plans a path for the robot 120 to travel to the target locations to perform an action. In a range scan, the input may include a range of coordinates corresponding to a range of target locations. In response, the planner 250 plans a path for the robot 120 to perform a full scan or actions for the range of target locations.
The planner 250 may plan the route of the robot 120 based on data provided by the visual reference engine 240 and the data provided by the state estimator 235. For example, the visual reference engine 240 estimates the current location of the robot 120 by tracking the number of regularly shaped structures in the storage site 110 passed by the robot 120. Based on the location information provided by the visual reference engine 240, the planner 250 determines the route of the robot 120 and may adjust the movement of the robot 120 as the robot 120 travels along the route.
The planner 250 may also include a fail-safe mechanism in the case where the movement of the robot 120 has deviated from the plan. For example, if the planner 250 determines that the robot 120 has passed a target aisle and traveled too far away from the target aisle, the planner 250 may send signals to the FCU 225 to try to remedy the path. If the error is not remedied after a timeout or within a reasonable distance, or the planner 250 is unable to correctly determine the current location, the planner 250 may direct the FCU to land or to stop the robot 120.
Relying on various location information, the planner 250 may also include algorithms for collision avoidance purposes. In some embodiments, the planner 250 relies on the distance information, the yaw angle, and center offset information relative to nearby objects to plan the movement of the robot 120 to provide sufficient clearance between the robot 120 and nearby objects. Alternatively, or additionally, the robot 120 may include one or more depth cameras such as a 360-degree depth camera set that generates distance data between the robot 120 and nearby objects. The planner 250 uses the location information from the depth cameras to perform collision avoidance.
The communication engine 255 and the I/O interface 260 are communication components to allow the robot 120 to communicate with other components in the system environment 100. A robot 120 may use different communication protocols, wireless or wired, to communicate with an external component such as the base station 130. Example communication protocols may include Wi-Fi, Bluetooth, NFC, USB, etc. that couple the robot 120 to the base station 130. The robot 120 may transmit various types of data, such as image data, flight logs, location data, inventory data, and robot status information. The robot 120 may also receive inputs from an external source to specify the actions that need to be performed by the robot 120. The commands may be automatically generated or manually generated by an administrator. The communication engine 255 may include algorithms for various communication protocols and standards, encoding, decoding, multiplexing, traffic control, data encryption, etc. for various communication processes. The I/O interface 260 may include software and hardware component such as hardware interface, antenna, and so forth for communication.
The robot 120 also includes a power source 265 used to power various components and the movement of the robot 120. The power source 265 may be one or more batteries or a fuel tank. Example batteries may include lithium-ion batteries, lithium polymer (LiPo) batteries, fuel cells, and other suitable battery types. The batteries may be placed inside permanently or may be easily replaced. For example, batteries may be detachable so that the batteries may be swapped when the robot 120 returns to the base station 130.
While
Continuing to refer to
The base station 130 includes one or more processors 270 and one or more memories 275 that include one or more set of instructions for causing the processors 270 to carry out various processes that are implemented as one or more software modules. The base station 130 may provide inputs and commands to the robot 120 for performing various inventory management tasks. The base station 130 may also include an instruction set for performing swarm control among multiple robots 120. Swarm control may include task allocation, routing and planning, coordination of movements among the robots to avoid collisions, etc. The base station 130 may serve as a central control unit to coordinate the robots 120. The memory 275 may also include various sets of instructions for performing analysis of data and images downloaded from a robot 120. The base station 130 may provide various degrees of data processing from raw data format conversion to a full data processing that generates useful information for inventory management. Alternatively, or additionally, the base station 130 may directly upload the data downloaded from the robot 120 to a data store, such as the data store 160. The base station 130 may also provide operation, administration, and management commands to the robot 120. In some embodiments, the base station 130 can be controlled remotely by the user device 170, the computing server 150, or the inventory management system 140.
The base station 130 may also include various types of I/O interfaces 280 for communications with the robot 120 and to the Internet. The base station 130 may communicate with the robot 120 continuously using a wireless protocol such as Wi-Fi or Bluetooth. In some embodiments, one or more components of the robot 120 in
The repowering unit 285 includes components that are used to detect the power level of the robot 120 and to repower the robot 120. Repowering may be done by swapping the batteries, recharging the batteries, re-filling the fuel tank, etc. In some embodiments, the base station 130 includes mechanical actuators such as robotic arms to swap the batteries on the robot 120. In another embodiment, the base station 130 may serve as the charging station for the robot 120 through wired charging or inductive charging. For example, the base station 130 may include a landing or resting pad that has an inductive coil underneath for wirelessly charging the robot 120 through the inductive coil in the robot. Other suitable ways to repower the robot 120 is also possible.
In accordance with some embodiments, the computer receives 310 a configuration of a storage site 110. The storage site 110 may be a warehouse, a retail store, or another suitable site. The configuration information of the storage site 110 may be uploaded to the robot 120 for the robot to navigate through the storage site 110. The configuration information may include a total number of the regularly shaped structures in the storage site 110 and dimension information of the regularly shaped structures. The configuration information provided may take the form of a computer-aided design (CAD) drawing or another type of file format. The configuration may include the layout of the storage site 110, such as the rack layout and placement of other regularly shaped structures. The layout may be a 2-dimensional layout. The computer extracts the number of sections, aisles, and racks and the number of rows and columns for each rack from the CAD drawing by counting those numbers as appeared in the CAD drawing. The computer may also extract the height and the width of the cells of the racks from the CAD drawing or from another source. In some embodiments, the computer does not need to extract the accurate distances between a given pair of racks, the width of each aisle, or the total length of the racks. Instead, the robot 120 may measure dimensions of aisles, racks, and cells from a depth sensor data or may use a counting method performed by the planner 250 in conjunction with the visual reference engine 240 to navigate through the storage site 110 by counting the number of rows and columns the robot 120 has passed. Hence, in some embodiments, the accurate dimensions of the racks may not be needed.
Some configuration information may also be manually inputted by an administrator of the storage site 110. For example, the administrator may provide the number of sections, the number of aisles and racks in each section, and the size of the cells of the racks. The administrator may also input the number of rows and columns of each rack.
Alternatively, or additionally, the configuration information may also be obtained through a mapping process such as a pre-flight mapping or a mapping process that is conducted as the robot 120 carries out an inventory management task. For example, for a storage site 110 that newly implements the automated management process, an administrator may provide the size of the navigable space of the storage site for one or more mapping robots to count the numbers of sections, aisles, rows and columns of the regularly shaped structures in the storage site 110. Again, in some embodiments, the mapping or the configuration information does not need to measure the accurate distance among racks or other structures in the storage site 110. Instead, a robot 120 may navigate through the storage site 110 with only a rough layout of the storage site 110 by counting the regularly shaped structures along the path in order to identify a target location. The robotic system may gradually perform mapping or estimation of scales of various structures and locations as the robot 120 continues to perform various inventory management tasks.
The computer receives 320 inventory management data for inventory management operations at the storage site 110. Certain inventory management data may be manually inputted by an administrator while other data may be downloaded from the inventory management system 140. The inventory management data may include scheduling and planning for inventory management operations, including the frequency of the operations, time window, etc. For example, the management data may specify that each location of the racks in the storage site 110 is to be scanned every predetermined period (e.g., every day) and the inventory scanning process is to be performed in the evening by the robot 120 after the storage site is closed. The data in the inventory management system 140 may provide the barcodes and labels of items, the correct coordinates of the inventory, information regarding racks and other storage spaces that need to be vacant for incoming inventory, etc. The inventory management data may also include items that need to be retrieved from the storage site 110 (e.g., items on purchase orders that need to be shipped) for each day so that the robot 120 may need to focus on those items.
The computer generates 330 a plan for performing inventory management. For example, the computer may generate an automatic plan that includes various commands to direct the robot 120 to perform various scans. The commands may specify a range of locations that the robot 120 needs to scan or one or more specific locations that the robot 120 needs to go. The computer may estimate the time for each scanning trip and design the plan for each operation interval based on the available time for the robotic inventory management. For example, in certain storage sites 110, robotic inventory management is not performed during the business hours.
The computer generates 340 various commands to operate one or more robots 120 to navigate the storage site 110 according to the plan and the information derived from the configuration of the storage site 110. The robot 120 may navigate the storage site 110 by at least visually recognizing the regularly shaped structures in the storage sites and counting the number of regularly shaped structures. In some embodiments, in addition to the localization techniques such as VIO used, the robot 120 counts the number of racks, the number of rows, and the number of columns that it has passed to determine its current location along a path from a starting location to a target location without knowing the accurate distance and direction that it has traveled.
The scanning of inventory or other inventory management tasks may be performed autonomously by the robot 120. In some embodiments, a scanning task begins at a base station at which the robot 120 receives 342 an input that includes coordinates of target locations in the storage site 110 or a range of target locations. The robot 120 departs 344 from the base station 130. The robot 120 navigates 346 through the storage site 110 by visually recognizing regularly shaped structures. For example, the robot 120 tracks the number of regularly shaped structures that are passed by the robot 120. The robot 120 makes turns and translation movements based on the recognized regularly shaped structures captured by the robot's image sensor 210. Upon reaching the target location, the robot 120 may align itself with a reference point (e.g., the center location) of the target location. At the target location, the robot 120 captures 348 data (e.g., measurements, pictures, etc.) of the target location that may include the inventory item, barcodes, and labels on the boxes of the inventory item. If the initial command before the departure of the robot 120 includes multiple target locations or a range of target locations, the robot 120 continues to the next target locations by moving up, down, or sideways to the next location to continue to scanning operation.
Upon completion of a scanning trip, the robot 120 returns 350 to the base station 130 by counting the number of regularly shaped structures that the robot 120 has passed, in a reversed direction. The robot 120 may potentially recognize the structures that the robot has passed when the robot 120 travels to the target location. Alternatively, the robot 120 may also return to the base station 130 by reversing the path without any count. The base station 130 repowers the robot 120. For example, the base station 130 provides the next commands for the robot 120 and swaps 352 the battery of the robot 120 so that the robot 120 can quickly return to service for another scanning trip. The used batteries may be charged at the base station 130. The base station 130 also may download the data and images captured by the robot 120 and upload the data and images to the data store 160 for further process. Alternatively, the robot 120 may include a wireless communication component to send its data and images to the base station 130 or directly to the network 180.
The computer performs 360 analyses of the data and images captured by the robot 120. For example, the computer may compare the barcodes (including serial numbers) in the images captured by the robot 120 to the data stored in the inventory management system 140 to identify if any items are misplaced or missing in the storage site 110. The computer may also determine other conditions of the inventory. The computer may generate a report to display at the user interface 175 for the administrator to take remedial actions for misplaced or missing inventory. For example, the report may be generated daily for the personnel in the storage site 110 to manually locate and move the misplaced items. Alternatively, or additionally, the computer may generate an automated plan for the robot 120 to move the misplaced inventory. The data and images captured by the robot 120 may also be used to confirm the removal or arrival of inventory items.
Referring to
The processors of the robot 120, such as the one executing the planner 250, control 520 the robot 120 to the target location 474 along a path 470. The path 470 may be determined based on the coordinate of the target location 474. The robot 120 may turn so that the image sensor 210 is facing the regularly shaped structures (e.g., the racks). The movement of the robot 120 to the target location 474 may include traveling to a certain aisle, taking a turn to enter the aisle, traveling horizontally to the target column, traveling vertically to the target row, and turning to the right angle facing the target location 474 to capture a picture of inventory items on the pallet 460.
As the robot 120 moves to the target location 474, the robot 120 captures 530 images of the storage site 110 using the image sensor 210. The images captured may be in a sequence of images. The robot 120 receives the images captured by the image sensor 210 as the robot 120 moves along the path 470. The images may capture the objects in the environment, including the regularly shaped structures such as the racks. For example, the robot 120 may use the algorithms in the visual reference engine 240 to visually recognize the regularly shaped structures.
The robot 120 analyzes 540 the images captured by the image sensor 210 to determine the current location of the robot 120 in the path 470 by tracking the number of regularly shaped structures in the storage site passed by the robot 120. The robot 120 may use various image processing and object recognition techniques to identify the regularly shaped structures and to track the number of structures that the robot 120 has passed. Referring to the path 470 shown in
The failure to maintain a level flight could bring various challenges to the navigation of the aerial robot 602. For example, the type of unwanted change in height shown in
In some embodiments, the aerial robot 610 is equipped with a distance sensor (e.g., the distance sensor 239) and a visual inertial sensor (e.g., the VIO unit 236). The aerial robot 610 may rely on the fusion of analyses of the distance sensor and visual inertial sensor to navigate the aerial robot 610 to maintain a level flight, despite the change in the surface levels in regions 604, 606, and 608. Again, the first region 604 may correspond to the floor and the second and third regions 606 and 608 may correspond to obstacles on the floor (e.g., objects on the floor, or pallets and inventory items placed on the floor in the setting of a storage site).
The aerial robot 610 may use data from both sensors to compensate for and adjust data of each other for determining a vertical height estimate regardless of whether the aerial robot 610 is traveling over the first region 604, the second region 606, or the third region 608. A distance sensor may return highly accurate measurements (with errors within feet, sometimes inches, or even smaller errors) of distance readings based on the round-trip time of the signal transmitted from the distance sensor's transmitter and reflected by a nearby surface at which the transmitter is pointing. However, the distance readings from the distance sensor may be affected by nearby environment changes such as the presence of an obstacle that elevates the surface at which the distance sensor's transmitter is pointing. Also, the orientation of the distance sensor may also not be directly pointing downward due to the orientation of the aerial robot 610. For example, in
A visual inertial sensor (e.g., the VIO unit 236), or simply an inertial sensor, may be less susceptible to environmental changes such as the presence of obstacles in the second and third regions 606 and 608. An inertial sensor may also simply be an inertial sensor such as the IMU 230 or include the visual element such as the VIO unit 236. An inertial sensor provides localization data of the aerial robot 610 based on the accelerometers and gyroscopes in an IMU. Since the IMU is internal to the aerial robot 610, the localization data is not measured relative to a nearby object or surface. Thus, the data is usually also not affected by a nearby object or surface. However, the position data (including a vertical height estimate) generated from an inertial sensor is often obtained by twice integrating, with respect to time, the acceleration data obtained from the accelerometers of an IMU. The localization data is prone to drift and could become less accurate as the aerial robot 610 travels a relatively long distance.
The aerial robot 610 may use data from a visual inertial sensor to compensate the data generated by the distance sensor in regions of transitions that are associated with a change in surface levels. In some embodiments, in regions of transitions, such as regions 640, 642, 644, and 646, the data from the distance sensor may become unstable due to sudden changes in the surface levels. The aerial robot 610 may temporarily switch to the visual inertial senor to estimate its vertical height. After the transition regions, the aerial robot 610 may revert to the distance sensor. Relying on both types of sensor data, the aerial robot 610 may travel in a relatively level manner (relatively at the same horizontal level), as illustrated in
In some embodiments, the aerial robot 610 may determine 650 a first height estimate 630 of the aerial robot 610 relative to a first region 604 with a first surface level 605 using data from the distance sensor. For example, the data from the distance sensor may take the form of a time series of distance readings from the distance sensor. For a particular instance, a processor of the aerial robot 610 may receive a distance reading from the data of the distance sensor. The processor may also receive a pose of the aerial robot 610. The pose may include a pitch angle 620, a roll angle 622, and a yaw angle. In some embodiments, the aerial robot 610 may use one or more angles related to the pose to determine the first height estimate 630 from the distance reading adjusted by the pitch angle 620 the roll angle 622. For example, the processor may use one or more trigonometry relationship to convert the distance reading to the first height estimate 630.
The processor controls 655 the flight of the aerial robot 610 over at least a part of the first region based on the first estimated height 630. As the aerial robot 610 travels over the first region 604, the readings from the distance sensor should be relatively stable. The aerial robot 610 may also monitor the data of the visual inertial sensor. The data of the visual inertial sensor may also be a time series of readings of localization data that include readings of height estimates. The readings of distance data from the distance sensor may be generated by, for example, a laser range finder while the readings of location data in the z-direction from the visual inertial sensor may be generated by double integrating the z-direction accelerometer's data with respect to time. Since the two sensors estimate the height using different sources and methods, the readings from the two sensors may not agree. In addition, the readings from the visual inertial sensor may also be affected by drifts. The aerial robot 610 may monitor the readings from the visual inertial sensors and determine a bias between the readings form the visual inertial sensor and the readings from the distance sensor. The bias may be the difference between the two readings.
The processor determines 660 that the aerial robot 610 is in a transition region 640 between the first region 604 and a second region 606 with a second surface level 607 that is different from the first surface level 605. A transition region may be a region where the surface levels are changing. The transition region may indicate the presence of an obstacle on the ground level, such as an object that prevents the distance sensor's signal from reaching the ground. For example, in the setting of a storage site, the transition region may be at the boundary of a pallet or an inventory item placed on the floor.
In various embodiments, a transition region and its size may be defined differently, depending on the implementation of the height estimation algorithm. In some embodiments, the transition region may be defined based on a predetermined length in the horizontal direction. For example, the transition region may be a fixed length after the distance sensor detects a sudden change in distance readings. In another embodiment, the transition region may be defined based on a duration of time. For example, the transition region may be a time duration after the distance sensor detects a sudden change in distance readings. The time may be a predetermined period or a relative period determined based on the speed of the aerial robot 610 in the horizontal direction.
In yet another embodiment, the transition region may be defined as a region in which the processor becomes uncertain that the aerial robot 610 is in a leveled region. For example, the aerial robot 610 may include, in its memory, one or more probabilistic models that determine the likelihood that the aerial robot 610 is traveling in a leveled region. The likelihood may be determined based on the readings of the distance data from the distance sensor, which should be relatively stable when the aerial robot 610 is traveling over a leveled region. If the likelihood that the aerial robot 610 is traveling in a leveled region is below a threshold value, the processor may determine that the aerial robot 610 is in a transition region. For example, in some embodiments, the processor may determine a first likelihood that the aerial robot 610 is in the first region 604. The processor may determine a second likelihood that the aerial robot 610 is in the second region 606. The processor may determine that the aerial robot is the transition region 640 based on the first likelihood and the second likelihood. For instance, if both the first likelihood indicates that the aerial robot 610 is unlikely to be in the first region 604 and the second likelihood indicates that the aerial robot 610 is unlikely to be in the second region 606, the process may determine that the aerial robot 610 is in the transition region 640.
In yet another embodiment, the transition region may be defined based on the presence of an obstacle. For example, the processor may determine whether an obstacle is present based on the distance readings from the distance sensors. The processor may determine an average of distance readings from the data of the distance sensor, such as an average of the time series distance data from a period preceding the latest value. The processor may determine a difference between the average and a particular distance reading at a particular instance, such as the latest instance. In response to the difference being larger than a threshold, the processor may determine that an obstacle likely is present at the particular instance because there is a sudden change in distance reading that is rather significant. The processor may, in turn, determine that the aerial robot 610 has entered a transition region until the readings from the distance sensor become stable again.
In yet another embodiment, the transition may be defined based on any suitable combinations of criteria mentioned above or another criterion that is not explicitly discussed.
The processor determines 665 a second height estimate 632 of the aerial robot 610 using data from the visual inertial sensor for at least a part of the duration in which the aerial robot 610 is in the transition region 640. At the transition region 640, the sudden change in surface levels from the first surface level 605 to the second surface level 607 prevents the distance senor from accurately determining the second height estimate 632 because the signal of the distance sensor cannot penetrate an obstacle and travel to the first surface level 605. Instead of using the data of the distance sensor, the aerial robot 610 switches to the data of the visual inertial sensor. However, as explained above, there may be biases between the readings of the distance sensor and the readings of the visual inertial sensor. The processor may determine the visual inertial bias. For example, the visual inertial bias may be determined from an average of the readings of the visual inertial sensor from a period preceding the transition region 640, such as the period during which the aerial robot 610 is in the first region 604. In determining the second height estimate 632, the processor receives a reading from the data of the visual inertial sensor. The processor determines the second height estimate 632 using the reading adjusted by the visual inertial bias.
The processor controls 670 the flight of the aerial robot 610 using the second height estimate 632 in the transition region 640. The size of the transition region 640 may depend on various factors as discussed in step 660. When traveling in the transition region 640 or immediately after the transition region 640, the processor may determine a distance sensor bias. For example, in the transition region, the visual inertial sensor may be providing the second height estimate 632 while the distance sensor may be providing a distance reading D because the signal of the distance sensor is reflected at the second surface level 607. As such, the distance sensor bias may be the difference between the second height estimate 632 and the distance reading D, which is approximately equal to the difference between the first surface level 605 and the second surface level 607.
Based on one or more factors that define a transition region as discussed above in step 660, the processor may determine that the aerial robot 610 has exited a transition region. For example, the processor determines 675 that the aerial robot 610 is in the second region 606 for more than a threshold period of time. The threshold period of time may be of a predetermined length or may be measured based on the stability of the data of the distance sensor. The processor reverts 680 to using the data from the distance sensor to determine a third height estimate 634 of the aerial robot 610 during which the aerial robot 610 is in the second region 606. In using the data of the distance sensor to determine the third height estimate 634, the processor may adjust the data using the distance sensor bias. For example, the processor may add the distance sensor bias to the distance readings from the distance sensor.
The aerial robot 610 may continue to travel to the third region 608 and back to the second region 606 via the transition region 642 and the transition region 644. The aerial robot 610 may repeat the process of switching between the data from the distance sensor and the data from the visual inertial sensor and monitoring the various biases between the two sets of data.
The height estimate algorithm 700 may include various functions for making different determinations. For example, the height estimate algorithm 700 may include an obstacle detection function 710, a downward status detection function 720, a visual inertial bias correction function 730, a distance sensor bias correction function 740, and a sensor selection and publication function 750. In various embodiments, the height estimate algorithm 700 may include different, fewer, or additional functions. Functions may also be combined or further separated. The determinations made by each function may also be distributed among various functions in a different manner described in
The flow described in the height estimate algorithm 700 may correspond to a particular instance in time. The processor of an aerial robot 610 may repeat the height estimate algorithm 700 to generate one or more time series of data. The height estimate algorithm 700 may receive distance sensor data 760, pose data 770, and visual inertial data 780 as inputs and generate the height estimate 790 as the output. The distance sensor data 760 may include mr, which may be the distance reading from a distance sensor, such as the distance reading as indicated by line 624 shown in
The obstacle detection function 710 may determine whether an obstacle is detected based on the pose data 770 {circumflex over (z)}, {circumflex over (ϕ)} and {circumflex over (θ)}, and the distance sensor data 760 mr. For example, the obstacle detection function 710 may determine whether the distance reading from the distance data 760 and the distance reading calculated from the pose data 770 agree (e.g., the absolute difference or square difference between the two readings is less than or larger than a threshold). If the two data sources agree, the obstacle detection function 710 may generate a first label as the output of the obstacle detection function 710. The first label denotes that an obstacle is not detected. If the two data sources do not agree, the obstacle detection function 710 may generate a second label as the output, which denotes that an obstacle is detected. The obstacle detection function 710 may be represented by the following mathematical equations. 1G may be the output of the obstacle detection function 710.
where,
d=(mr−
r
={circumflex over (z)}/(cos({circumflex over (ϕ)})*sin({circumflex over (θ)}))
The downward status detection function 720 may include one or more probabilities model to determine the likelihood P(H1) that the aerial robot 610 is flying over a first region (e.g., the floor) and the likelihood P(H2) that the aerial robot 610 is flying over a second region (e.g., on top of an obstacle). The downward status detection function 720 assigns a state S to the aerial robot 610. The state may correspond to the first region, the second region, or a transition region. For example, if the likelihood P(H1) and likelihood P(H2) indicate that the aerial robot 610 is neither in the first region nor the second region, the downward status detection function 720 assigns that the aerial robot 610 is in the transition region. The downward status detection function 720 may be represented by the following mathematical equations.
where
The visual inertial bias correction function 730 monitors the averaged bias of the visual inertial data 780 mv relative to the distance sensor data 760 mr. As discussed above, data from a visual inertial sensor is prone to errors from drifts. The data from the visual inertial sensor may also have a constant bias compared to the data from the distance sensor. The aerial robot 610 monitors the visual inertial data 780 and determines the average of the visual inertial data 780 over a period of time. The average may be used to determine the visual inertial bias and corrects the visual inertial data 780 based on the bias. The visual inertial bias correction function 730 may be represented by the following mathematical equations. bz(k) denotes the visual inertial bias and MA denotes a moving average. {circumflex over (m)}v,z(k) denotes the adjusted visual inertial data.
{circumflex over (m)}
v,z(k)=MA(mv,z(k−n:k))
b
z(k)={circumflex over (m)}v,z(k)−mr(k)cos(ϕ))cos(θ)
{hacek over (m)}
v,z(k)={circumflex over (m)}v,z(k)−bz(k)
The distance sensor bias correction function 740 compensates the distance sensor data 760 from the distance sensor when the aerial robot 610 is flying over an obstacle. The values of the distance sensor data 760 may become smaller than the actual height because signals from the distance sensor are unable to reach the ground due to the presence of an obstacle. The distance sensor bias correction function 740 makes the adjustment when the aerial robot 610 reverts to using the distance sensor to estimate height after a transition region. The distance sensor bias correction function 740 may be represented by the following mathematical equations. br(k) denotes the distance sensor bias and {circumflex over (m)}r(k) denotes the adjusted distance sensor data.
If S=1 and ts=1<ε, (on obstacle)
{hacek over (m)}r(k)=mr(k)−br(k)
where
The sensor selection and publication function 750 selects the sensor used in various situations and generate the final determination of the height estimate z. For example, in one embodiment, if the aerial robot 610 is in the first region, the aerial robot 610 uses the distance sensor data 760 to determine the height estimate z. If the aerial robot 610 is in the transition region, the aerial robot 610 uses the visual inertial data 780. If the aerial robot 610 is in the second region (e.g., on top of an obstacle) after the transition region within a threshold period of time, the aerial robot 610 may also use the visual inertial data 780. Afterward, the aerial robot 610 reverts to using the distance sensor data 760. The sensor selection and publication function 750 may be represented by the following pseudocode.
The height estimate algorithm 700 provides an example of estimating heights of an aerial robot that may be implemented at a site that has a layer of obstacles. In various embodiments, similar principles may be expanded for multiple layers of obstacles.
Shortly after the transition region 794 and within the threshold ε 796, the aerial robot 610 may determine that the distance sensor data 760 may become stable again. In this period, the aerial robot 610 may continue to use the visual inertial data 780 and may run the distance sensor bias correction function 740 to determine a compensation value that should be added to the distance sensor data 760 to account for the depth of the obstacle. When the aerial robot 610 is in the second region 798 (e.g., on top of the obstacle) and the aerial robot 610 also determines that it is ready to switch back to the distance sensor (e.g., the data of the distance sensor is stable again), the aerial robot 610 uses the distance sensor data 760 to estimate the height again, with an adjustment by the distance sensor bias. The aerial robot 610 also runs the visual inertial bias correction function 730 again to monitor the bias of the visual inertial data 780. The process may continue in a similar manner as the aerial robot 610 travel across different surface levels.
In various embodiments, a wide variety of machine learning techniques may be used. Examples include different forms of supervised learning, unsupervised learning, and semi-supervised learning such as decision trees, support vector machines (SVMs), regression, Bayesian networks, and genetic algorithms. Deep learning techniques such as neural networks, including convolutional neural networks (CNN), recurrent neural networks (RNN) and long short-term memory networks (LSTM), may also be used. For example, various object recognitions performed by visual reference engine 240, localization, and other processes may apply one or more machine learning and deep learning techniques.
In various embodiments, the training techniques for a machine learning model may be supervised, semi-supervised, or unsupervised. In supervised learning, the machine learning models may be trained with a set of training samples that are labeled. For example, for a machine learning model trained to classify objects, the training samples may be different pictures of objects labeled with the type of objects. The labels for each training sample may be binary or multi-class. In training a machine learning model for image segmentation, the training samples may be pictures of regularly shaped objects in various storage sites with segments of the images manually identified. In some cases, an unsupervised learning technique may be used. The samples used in training are not labeled. Various unsupervised learning technique such as clustering may be used. In some cases, the training may be semi-supervised with training set having a mix of labeled samples and unlabeled samples.
A machine learning model may be associated with an objective function, which generates a metric value that describes the objective goal of the training process. For example, the training may intend to reduce the error rate of the model in generating predictions. In such a case, the objective function may monitor the error rate of the machine learning model. In object recognition (e.g., object detection and classification), the objective function of the machine learning algorithm may be the training error rate in classifying objects in a training set. Such an objective function may be called a loss function. Other forms of objective functions may also be used, particularly for unsupervised learning models whose error rates are not easily determined due to the lack of labels. In image segmentation, the objective function may correspond to the difference between the model's predicted segments and the manually identified segments in the training sets. In various embodiments, the error rate may be measured as cross-entropy loss, L1 loss (e.g., the sum of absolute differences between the predicted values and the actual value), L2 loss (e.g., the sum of squared distances).
Referring to
The order of layers and the number of layers of the CNN 800 in
A machine learning model may include certain layers, nodes, kernels and/or coefficients. Training of a neural network, such as the CNN 800, may include forward propagation and backpropagation. Each layer in a neural network may include one or more nodes, which may be fully or partially connected to other nodes in adjacent layers. In forward propagation, the neural network performs the computation in the forward direction based on outputs of a preceding layer. The operation of a node may be defined by one or more functions. The functions that define the operation of a node may include various computation operations such as convolution of data with one or more kernels, pooling, recurrent loop in RNN, various gates in LSTM, etc. The functions may also include an activation function that adjusts the weight of the output of the node. Nodes in different layers may be associated with different functions.
Each of the functions in the neural network may be associated with different coefficients (e.g. weights and kernel coefficients) that are adjustable during training. In addition, some of the nodes in a neural network may also be associated with an activation function that decides the weight of the output of the node in forward propagation. Common activation functions may include step functions, linear functions, sigmoid functions, hyperbolic tangent functions (tanh), and rectified linear unit functions (ReLU). After an input is provided into the neural network and passes through a neural network in the forward direction, the results may be compared to the training labels or other values in the training set to determine the neural network's performance. The process of prediction may be repeated for other images in the training sets to compute the value of the objective function in a particular training round. In turn, the neural network performs backpropagation by using gradient descent such as stochastic gradient descent (SGD) to adjust the coefficients in various functions to improve the value of the objective function.
Multiple rounds of forward propagation and backpropagation may be performed. Training may be completed when the objective function has become sufficiently stable (e.g., the machine learning model has converged) or after a predetermined number of rounds for a particular set of training samples. The trained machine learning model can be used for performing prediction, object detection, image segmentation, or another suitable task for which the model is trained.
By way of example,
The structure of a computing machine described in
By way of example, a computing machine may be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular telephone, a smartphone, a web appliance, a network router, an internet of things (IoT) device, a switch or bridge, or any machine capable of executing instructions 924 that specify actions to be taken by that machine. Further, while only a single machine is illustrated, the term “machine” shall also be taken to include any collection of machines that individually or jointly execute instructions 924 to perform any one or more of the methodologies discussed herein.
The example computer system 900 includes one or more processors (generally, processor 902) (e.g., a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), one or more application-specific integrated circuits (ASICs), one or more radio-frequency integrated circuits (RFICs), or any combination of these), a main memory 904, and a non-volatile memory 906, which are configured to communicate with each other via a bus 908. The computer system 900 may further include graphics display unit 910 (e.g., a plasma display panel (PDP), a liquid crystal display (LCD), a projector, or a cathode ray tube (CRT)). The computer system 900 may also include alphanumeric input device 912 (e.g., a keyboard), a cursor control device 914 (e.g., a mouse, a trackball, a joystick, a motion sensor, or other pointing instrument), a storage unit 916, a signal generation device 918 (e.g., a speaker), and a network interface device 920, which also are configured to communicate via the bus 908.
The storage unit 916 includes a computer-readable medium 922 on which is stored instructions 924 embodying any one or more of the methodologies or functions described herein. The instructions 924 may also reside, completely or at least partially, within the main memory 904 or within the processor 902 (e.g., within a processor's cache memory) during execution thereof by the computer system 900, the main memory 904 and the processor 902 also constituting computer-readable media. The instructions 924 may be transmitted or received over a network 926 via the network interface device 920.
While computer-readable medium 922 is shown in an example embodiment to be a single medium, the term “computer-readable medium” should be taken to include a single medium or multiple media (e.g., a centralized or distributed database, or associated caches and servers) able to store instructions (e.g., instructions 924). The computer-readable medium may include any medium that is capable of storing instructions (e.g., instructions 924) for execution by the machine and that cause the machine to perform any one or more of the methodologies disclosed herein. The computer-readable medium may include, but not be limited to, data repositories in the form of solid-state memories, optical media, and magnetic media. The computer-readable medium does not include a transitory medium such as a signal or a carrier wave.
Certain embodiments are described herein as including logic or a number of components, engines, modules, or mechanisms. Engines may constitute either software modules (e.g., code embodied on a computer-readable medium) or hardware modules. A hardware engine is a tangible unit capable of performing certain operations and may be configured or arranged in a certain manner. In example embodiments, one or more computer systems (e.g., a standalone, client or server computer system) or one or more hardware engines of a computer system (e.g., a processor or a group of processors) may be configured by software (e.g., an application or application portion) as a hardware engine that operates to perform certain operations as described herein.
In various embodiments, a hardware engine may be implemented mechanically or electronically. For example, a hardware engine may comprise dedicated circuitry or logic that is permanently configured (e.g., as a special-purpose processor, such as a field programmable gate array (FPGA) or an application-specific integrated circuit (ASIC)) to perform certain operations. A hardware engine may also comprise programmable logic or circuitry (e.g., as encompassed within a general-purpose processor or another programmable processor) that is temporarily configured by software to perform certain operations. It will be appreciated that the decision to implement a hardware engine mechanically, in dedicated and permanently configured circuitry, or temporarily configured circuitry (e.g., configured by software) may be driven by cost and time considerations.
The various operations of example methods described herein may be performed, at least partially, by one or more processors, e.g., processor 902, that are temporarily configured (e.g., by software) or permanently configured to perform the relevant operations. Whether temporarily or permanently configured, such processors may constitute processor-implemented engines that operate to perform one or more operations or functions. The engines referred to herein may, in some example embodiments, comprise processor-implemented engines.
The performance of certain of the operations may be distributed among the one or more processors, not only residing within a single machine, but deployed across a number of machines. In some example embodiments, the one or more processors or processor-implemented modules may be located in a single geographic location (e.g., within a home environment, an office environment, or a server farm). In other example embodiments, the one or more processors or processor-implemented modules may be distributed across a number of geographic locations.
Upon reading this disclosure, those of skill in the art will appreciate still additional alternative structural and functional designs for a similar system or process through the disclosed principles herein. Thus, while particular embodiments and applications have been illustrated and described, it is to be understood that the disclosed embodiments are not limited to the precise construction and components disclosed herein. Various modifications, changes, and variations, which will be apparent to those skilled in the art, may be made in the arrangement, operation and details of the method and apparatus disclosed herein without departing from the spirit and scope defined in the appended claims.
The present application claims the benefit of U.S. Provisional Patent Application 63/274,448, filed on Nov. 1, 2021, which is hereby incorporated by reference in its entirety.
Number | Date | Country | |
---|---|---|---|
63274448 | Nov 2021 | US |