Methods and apparatus for autocalibration of a wearable electrode sensor system

Abstract

Methods and systems used in calibrating the position and/or orientation of a wearable device configured to be worn on a wrist or forearm of a user, the method comprises sensing a plurality of neuromuscular signals from the user using a plurality of sensors arranged on the wearable device, and providing the plurality of neuromuscular signals and/or signals derived from the plurality of neuromuscular signals as inputs to one or more trained autocalibration models, determining based, at least in part, on the output of the one or more trained autocalibration models, a current position and/or orientation of the wearable device on the user, and generating a control signal based, at least in part, on the current position and/or orientation of the wearable device on the user and the plurality of neuromuscular signals.

Description

BACKGROUND

Some smart wearable devices that detect user-generated signals can be worn by users in various orientations and positions on the body. A common issue for these types of devices is that signal detection can be negatively affected if the wearable device is worn at a location or in an orientation for which the device was not optimized for performance.

SUMMARY

Some embodiments are directed to a system for calibrating the position and/or orientation of a wearable device configured to be worn on a wrist or forearm of a user. The system comprises a plurality of sensors arranged on the wearable device, wherein the plurality of sensors are configured to continuously sense a plurality of neuromuscular signals from the user, and at least one computer processor. The at least one computer processor is programmed to provide the plurality of neuromuscular signals and/or signals derived from the plurality of neuromuscular signals as inputs to one or more trained autocalibration models, determine based, at least in part, on the output of the one or more trained autocalibration models, a current position and/or orientation of the wearable device on the user, and generate a control signal based, at least in part, on the current position and/or orientation of the wearable device on the user and the plurality of neuromuscular signals.

In one aspect, the at least one computer processor is programmed to determine the current position and/or orientation of the wearable device on the user without the user performing a particular pose or gesture during sensing of the plurality of neuromuscular signals.

In another aspect, the at least one computer processor is programmed to process the sensed plurality of neuromuscular signals prior to providing the processed neuromuscular signals to the one or more autocalibration models.

In another aspect, the at least one programmed processor is further programmed to generate calibrated neuromuscular signals based, at least in part, on the current position and/or orientation of the wearable device and the plurality of neuromuscular signals, and generating a control signal comprises generating a control signal based, at least in part, on the calibrated neuromuscular signals.

In another aspect, the at least one programmed processor is further programmed to select or modify one or more inference models based, at least in part, on the current position and/or orientation of the wearable device, and provide the plurality of neuromuscular signals as input to the selected or modified one or more inference models, and generating a control signal is further based, at least in part, on an output of the selected or modified one or more inference models.

In another aspect, the one or more autocalibration models used to determine the current position and/or orientation of the wearable device on the user include a neural network.

In another aspect, the neural network is an LSTM neural network.

In another aspect, the LSTM neural network includes at least one pooling layer.

In another aspect, the at least one pooling layer comprises a max pooling layer.

In another aspect, the at least one pooling layer provides rotation invariance of ±1 sensor locations.

In another aspect, the one or more autocalibration models are trained to identify the orientation and/or location of the wearable device based on neuromuscular signals obtained from a plurality of users.

In another aspect, the neuromuscular signals obtained from a plurality of users were obtained as each of the plurality of users performed a plurality of hand gestures and/or poses while the wearable device was oriented and/or positioned in multiple different configurations on a forearm and/or wrist of the user.

In another aspect, the plurality of neuromuscular sensors comprises a plurality of electromyography (EMG) sensors.

In another aspect, the at least one computer processor is further programmed to identify that at least a portion of the wearable device has moved on the wrist or forearm of the user, and determine the current position and/or orientation of the wearable device on the user in response to identifying that at least a portion of the wearable device has moved on the wrist or forearm of the user.

In another aspect, identifying the at least a portion of the wearable device has moved on the wrist or forearm of the user comprises determining that the wearable device has rotated around and/or translated along the user's wrist or forearm.

In another aspect, identifying the at least a portion of the wearable device has moved on the wrist or forearm of the user comprises detecting one or more movement artifacts in the sensed plurality of neuromuscular signals.

In another aspect, the system further comprises at least one inertial measurement unit (IMU) sensor, and identifying the at least a portion of the wearable device has moved on the wrist or forearm of the user is based, at least in part, on at least one signal sensed by the at least one IMU sensor.

In another aspect, determining the current position and/or orientation of wearable device comprises determining a rotation of the wearable device relative to a virtual reference orientation of the wearable device.

In another aspect, the control signal is a control signal to control one or more operations of or within a virtual reality system or an augmented reality system.

Some embodiments are directed to a method for calibrating the position and/or orientation of a wearable band on a user. The method comprises sensing a plurality of neuromuscular signals from the user using a plurality of sensors arranged on the wearable device, providing the plurality of neuromuscular signals and/or signals derived from the plurality of neuromuscular signals as inputs to one or more trained autocalibration models, determining based, at least in part, on the output of the one or more trained autocalibration models, a current position and/or orientation of the wearable device on the user, and generating a control signal based, at least in part, on the current position and/or orientation of the wearable device on the user and the plurality of neuromuscular signals.

In one aspect, determining the current position and/or orientation of the wearable device comprises determining the current position and/orientation of the wearable device without the user performing a particular pose or gesture during sensing of the plurality of neuromuscular signals.

In another aspect, the method further comprises processing the sensed plurality of neuromuscular signals prior to providing the processed neuromuscular signals to the one or more autocalibration models.

In another aspect, the method further comprises generating calibrated neuromuscular signals based, at least in part, on the current position and/or orientation of the wearable device and the plurality of neuromuscular signals, and generating a control signal comprises generating a control signal based, at least in part, on the calibrated neuromuscular signals. In another aspect, the method further comprises selecting or modifying one or more inference models based, at least in part, on the current position and/or orientation of the wearable device, and providing the plurality of neuromuscular signals as input to the selected or modified one or more inference models, wherein generating a control signal is further based, at least in part, on an output of the selected or modified one or more inference models.

In another aspect, the one or more autocalibration models used to determine the current position and/or orientation of the wearable device on the user include a neural network. In another aspect, the one or more autocalibration models are trained to identify the orientation and/or location of the wearable device based on neuromuscular signals obtained from a plurality of users.

In another aspect, the neuromuscular signals obtained from a plurality of users were obtained as each of the plurality of users performed a plurality of hand gestures and/or poses while the wearable device was oriented and/or positioned in multiple different configurations on a forearm and/or wrist of the user.

In another aspect, determining the current position and/or orientation of wearable device comprises determining a rotation of the wearable device relative to a virtual reference orientation of the wearable device.

In another aspect, the method further comprises identifying that at least a portion of the wearable device has moved on the wrist or forearm of the user, and determining the current position and/or orientation of the wearable device on the user in response to identifying that at least a portion of the wearable device has moved on the wrist or forearm of the user.

In another aspect, identifying the at least a portion of the wearable device has moved on the wrist or forearm of the user comprises determining that the wearable device has rotated around and/or translated along the user's wrist or forearm.\

In another aspect, identifying the at least a portion of the wearable device has moved on the wrist or forearm of the user comprises detecting one or more movement artifacts in the sensed plurality of neuromuscular signals.

In another aspect, the method further comprises identifying the at least a portion of the wearable device has moved on the wrist or forearm of the user is based, at least in part, on at least one signal sensed by at least one inertial measurement unit (IMU) sensor.

Some embodiments are directed to a method for training an inference model for autocalibration of a wearable device. The method comprises receiving neuromuscular signals sensed from a plurality of users as the wearable device was worn by each of the users while performing a plurality of hand gestures and/or poses, wherein for each of the plurality of hand gestures and/or poses, the wearable device was oriented or positioned differently on the user, labeling the received neuromuscular signals for each of the plurality of hand gestures and/or poses based on the orientation or position of the wearable device during sensing of the corresponding neuromuscular signals, and training one or more autocalibration models based, at least in part, on the labeled neuromuscular signals.

In one aspect, the method further comprises extracting from the received neuromuscular signals, a plurality of templates, where each of the plurality of templates corresponds to one of the plurality of hand gestures and/or poses, and generating calibrated neuromuscular signals based, at least on the extracted plurality of templates, and training the one or more autocalibration models comprises training the one or more autocalibration models based, at least in part, on the calibrated neuromuscular signals.

In another aspect, the method further comprises predicting a plurality of rotation offsets of the wearable device using a plurality of simulations of different orientations.

In another aspect, the one or more autocalibration models include a neural network.

In another aspect, the neural network is an LSTM neural network.

Some embodiments are directed to a non-transitory computer-readable storage medium encoded with a plurality of processor-executable instructions that, when executed by one or more computer processors perform one or more of the foregoing methods.

It should be appreciated that all combinations of the foregoing concepts and additional concepts discussed in greater detail below (provided such concepts are not mutually inconsistent) are contemplated as being part of the inventive subject matter disclosed herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the inventive subject matter disclosed herein.

BRIEF DESCRIPTION OF DRAWINGS

Various non-limiting embodiments of the technology will be described with reference to the following figures. It should be appreciated that the figures are not necessarily drawn to scale.

FIG. 1 illustrates a wearable system with sixteen EMG sensors arranged circumferentially around a band configured to be worn around a user's lower arm or wrist, in accordance with some embodiments of the technology described herein;

FIGS. 2A and 2B schematically illustrate a computer-based system that includes a wearable portion and a dongle portion, respectively, in accordance with some embodiments of the technology described herein;

FIG. 3 is a plot illustrating an example distribution of outputs generated by an autocalibration model trained in accordance with some embodiments. The distribution of outputs is generated across a dataset with data collected from different users;

FIG. 4 shows predicted values output from an autocalibration model trained in accordance with some embodiments. The predicted value are averaged across time which result in model predictions with greater confidence values;

FIG. 5 shows a plot of the accuracy of an autocalibration model trained in accordance with some embodiments. The accuracy is expressed in an Area Under the Receiver Operating Characteristic Curve (AUC);

FIG. 6 shows a plot of a correlation between a baseline model and an augmented model for autocalibrating sensors of a wearable device in accordance with some embodiments;

FIG. 7 shows a plot of correlations of augmented inference models trained with electrode invariances set to ±1, ±2, and ±3 along with the baseline model in accordance with some embodiments;

FIG. 8 schematically illustrates a visualization of an energy plot for a single EMG electrode on a wearable device at a first position and a second position in which the wearable device has been rotated;

FIG. 9 illustrates a flowchart of a process for offline training of an autocalibration model based on neuromuscular signals recorded from a plurality of users in accordance with some embodiments; and

FIG. 10 illustrates a flowchart of a process for using the trained autocalibration model to calibrate the position and/or orientation of sensors on a wearable device in accordance with some embodiments.

DETAILED DESCRIPTION

The disclosed methods and apparatus herein relate to autocalibration of sensor systems such that the systems perform well when the sensors are worn by users in various orientations and/or locations on a body part In some embodiments of the technology described herein, sensor signals may be used to predict information about a position and/or a movement of one or more portions of a user's body (e.g., a leg, an arm, and/or a hand), which may be represented as a multi-segment articulated rigid-body system with joints connecting the multiple segments of the rigid-body system. For example, in the case of a hand movement, signals sensed by wearable neuromuscular sensors placed at locations on the user's body (e.g., the user's arm and/or wrist) may be provided as input to one or more inference models trained to predict estimates of the position (e.g., absolute position, relative position, orientation) and the force(s) associated with a plurality of rigid segments in a computer-based musculoskeletal representation associated with a hand, for example, when the user performs one or more hand movements. The combination of position information and force information associated with segments of the musculoskeletal representation associated with a hand may be referred to herein as a “handstate” of the musculoskeletal representation. As a user performs different movements, a trained inference model may interpret neuromuscular signals sensed by the wearable neuromuscular sensors into position and force estimates (handstate information) that are used to update the musculoskeletal representation.

In certain embodiments, processed neuromuscular signals and/or neuromuscular signal patterns (e.g., EMG signals and/or EMG signal patterns) are collected from multiple users, and the processed signals and/or signal patterns are used to generate generalized models to predict the orientation and/or location of an armband containing EMG sensors. For example, spatiotemporal patterns of EMG signals can be obtained as users perform different handstate configurations with or without system supervision (e.g., gestures or poses comprising various pinches and wrist movements in the four cardinal directions) when the armband is positioned on the forearm and/or oriented in various ways, and these signals can be used to train generalized inference models to predict the user's gestures based on detected EMG signals. In certain embodiments, the systems and methods disclosed herein comprise identifying a “virtual” reference electrode and mapping the specific orientation of an armband to an offset value associated with an electrode and the “virtual” reference electrode.

As used herein, the term “gestures” may refer to a static or dynamic configuration of one or more body parts including a position of the one or more body parts and forces associated with the configuration. For example, gestures may include discrete gestures, such as placing or pressing the palm of a hand down on a solid surface, or grasping a ball, or pinching two fingers together (e.g., to form a pose); or continuous gestures, such as waving a finger back and forth, grasping and throwing a ball, rotating a wrist in a direction; or a combination of discrete and continuous gestures. Gestures may include covert gestures that may be imperceptible to another person, such as slightly tensing a joint by co-contracting opposing muscles or using sub-muscular activations, or “off-manifold” activations. In training an inference model, gestures may be defined using an application configured to prompt a user to perform the gestures or, alternatively, gestures may be arbitrarily defined by a user. The gestures performed by the user may include symbolic gestures (e.g., gestures mapped to other gestures, interactions, or commands, for example, based on a gesture vocabulary that specifies the mapping). In some cases, hand and arm gestures may be symbolic and used to communicate according to cultural standards.

Following autocalibration of the system(s), in various embodiments, a number of muscular activation states of a user may be identified from the recorded and/or detected signals and/or information based on the signals, to enable improved selection and/or control of objects in the user's environment when those objects are configured to receive control signals. The control of the objects can be performed directly from a neuromuscular activity device or indirectly via another system such as an augmented reality (AR) system or any extended or cross reality system (XR system or environment), including but not limited to mixed reality (MR), virtual reality (VR), etc.

The description set forth below is intended as a description of various configurations of the subject technology and is not intended to represent the only configurations in which the subject technology may be practiced. The appended drawings are incorporated herein and constitute a part of the detailed description. The detailed description includes specific details for the purpose of providing a thorough understanding of the subject technology. It, however, will be clear and apparent that the subject technology is not limited to the specific details set forth herein and may be practiced without these specific details. In some instances, well-known structures and components are shown in block diagram form to avoid obscuring the concepts of the subject technology.

The terms “computer”, “processor”, “computer processor”, “compute device” or the like should be expansively construed to cover any kind of electronic device with data processing capabilities including, by way of non-limiting example, a digital signal processor (DSP), a microcontroller, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or any other electronic computing device comprising one or more processors of any kind, or any combination thereof.

As used herein, the phrase “for example,” “such as”, “for instance” and variants thereof describe non-limiting embodiments of the presently disclosed subject matter. Reference in the specification to “one case”, “some cases”, “other cases”, or variants thereof means that a particular feature, structure or characteristic described in connection with the embodiment(s) is included in at least one embodiment of the presently disclosed subject matter. Thus the appearance of the phrase “one case”, “some cases”, “other cases” or variants thereof does not necessarily refer to the same embodiment(s).

It is appreciated that, unless specifically stated otherwise, certain features of the presently disclosed subject matter, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the presently disclosed subject matter, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable sub-combination.

FIG. 1 illustrates a band system with sixteen EMG sensors arranged circumferentially around an elastic band 102 configured to be worn around a user's lower arm. For example, FIG. 1 shows EMG sensors 101 arranged circumferentially (e.g., symmetrically spaced) around elastic band 102. It should be appreciated that any suitable number of neuromuscular sensors may be used and the number and arrangement of neuromuscular sensors used may depend on the particular application for which the wearable device is used. For example, a wearable armband or wristband may be used to generate control information for controlling an augmented reality system, a robot, controlling a vehicle, scrolling through text, controlling a virtual avatar, or any other suitable control task. Further, the band system can be configured to be worn around other parts of a user's body, such as their thigh or calf, for example.

In some embodiments, sensors 101 only includes a plurality of neuromuscular or muscular sensors or electrodes (e.g., EMG electrodes/sensors, MMG electrodes/sensors, SMG electrodes/sensors, etc.). In other embodiments, sensors 101 includes a plurality of neuromuscular sensors and at least one “auxiliary” sensor configured to continuously record a plurality of auxiliary signals. Examples of auxiliary sensors include, but are not limited to, other sensors such as inertial measurement unit (IMU) sensors, microphones, imaging devices (e.g., a camera), radiation based sensors for use with a radiation-generation device (e.g., a laser-scanning device), or other types of sensors such as a heart-rate monitor.

In some embodiments, the output of one or more of the sensors may be optionally processed using hardware signal processing circuitry (e.g., to perform amplification, filtering, and/or rectification). In other embodiments, at least some signal processing of the output of the sensors may be performed in software. Thus, signal processing of signals sensed by the sensors may be performed in hardware, software, or by any suitable combination of hardware and software, as aspects of the technology described herein are not limited in this respect. A non-limiting example of a signal-processing procedure used to process recorded data from the sensors 101 is discussed in more detail below in connection with FIGS. 2A and 2B.

FIGS. 2A and 2B illustrate a schematic diagram with internal components of a wearable system with sixteen sensors (e.g., EMG sensors), in accordance with some embodiments of the technology described herein. As shown, the wearable system includes a wearable portion 210 (FIG. 2A) and a dongle portion 220 (FIG. 2B). Although not illustrated, the dongle portion 220 is in communication with the wearable portion 210 (e.g., via Bluetooth or another suitable short range wireless communication technology). As shown in FIG. 2A, the wearable portion 210 includes the sensors 101, examples of which are described above in connection with FIG. 1. The sensors 101 provide output (e.g., sensed signals) to an analog front end 230, which performs analog processing (e.g., noise reduction, filtering, etc.) on the sensed signals. Processed analog signals produced by the analog front end 230 are then provided to an analog-to-digital converter 232, which converts the processed analog signals to digital signals that can be processed by one or more computer processors. An example of a computer processor that may be used in accordance with some embodiments is a microcontroller (MCU) 234. As shown in FIG. 2A, the MCU 234 may also receive inputs from other sensors (e.g., an IMU 240) and from a power and battery module 242. As will be appreciated, the MCU 234 may receive data from other devices not specifically shown. A processing output by the MCU 234 may be provided to an antenna 250 for transmission to the dongle portion 220, shown in FIG. 2B. The dongle portion 220 includes an antenna 252 that communicates with the antenna 250 of the wearable portion 210. Communication between the antennas 250 and 252 may occur using any suitable wireless technology and protocol, non-limiting examples of which include radiofrequency signaling and Bluetooth. As shown, the signals received by the antenna 252 of the dongle portion 220 may be provided to a host computer for further processing, for display, and/or for effecting control of a particular physical or virtual object or objects (e.g., to perform a control operation in an AR or VR environment)

Although the examples provided with reference to FIGS. 1, 2A, and 2B are discussed in the context of interfaces with EMG sensors, it is to be understood that the wearable systems described herein can also be implemented with other types of sensors, including, but not limited to, mechanomyography (MMG) sensors, sonomyography (SMG) sensors, and electrical impedance tomography (EIT) sensors.

As described above in connection with FIGS. 2A and 2B, the wearable system may include one or more computer processors (e.g., MCU 234) programmed to communicate with the sensors (e.g., neuromuscular sensors 101 and/or IMU sensor(s) 240). For example, signals recorded by one or more of the sensors may be provided to the processor(s), which may be programmed to execute one or more trained inference models or machine learning models that process signals captured by the sensors.

In some embodiments, the trained inference model(s) may comprise a neural network and, for example, may comprise a recurrent neural network. In some embodiments, the recurrent neural network may be a long short-term memory (LSTM) neural network. It should be appreciated, however, that the recurrent neural network is not limited to be an LSTM neural network and may have any other suitable architecture. For example, in some embodiments, the recurrent neural network may be a fully recurrent neural network, a gated recurrent neural network, a recursive neural network, a Hopfield neural network, an associative memory neural network, an Elman neural network, a Jordan neural network, an echo state neural network, a second order recurrent neural network, and/or any other suitable type of recurrent neural network. In other embodiments, neural networks that are not recurrent neural networks may be used. For example, deep neural networks, convolutional neural networks, and/or feedforward neural networks, may be used. In some implementations, the inference model(s) can comprise an unsupervised machine learning model, i.e., users are not required to perform a predetermined set of handstate configurations for which an inference model was previously trained to predict or identify. In other embodiments, the inference model(s) can comprise one or more supervised machine learning models wherein users performed specific gestures or handstate configurations in response to instructions, and the detected and processed signals from the electrodes or sensors have been associated with the performed gestures or handstate configurations.

In some embodiments, one or more inference models (e.g., a neural network as discussed above) can be implemented to classify sets of EMG signals or patterns characterized by unique spatio-temporal patterns that vary in amplitude for different amounts of forces generated by users' muscles or motor units. The processed EMG signals and/or patterns can be associated with a manner or way a user wears the band system during signal sensing (e.g., the orientation and/or positioning of the band on user's forearm). Accordingly, in one embodiment, the inference model(s) can associate one or more unique EMG signals or patterns with a specific orientation and/or location of the neuromuscular armband on the user (e.g., in a manner in which the electrodes are in contact with certain areas of the user's skin). Likewise, the inference model(s) can associate one or more unique EMG signals or patterns with a rotated orientation of the armband and/or with a re-located positioning of the armband when the user moves the armband from a lower forearm position to an upper forearm position (or the other way around, i.e., from an upper forearm position to a lower forearm position). Thus, the inference model(s) can be trained to associate various armband rotations and/or positional offsets with detected and processed EMG signals or signal patterns. Once the system can identify the specific orientation and/or positioning of the band on the user, the system can select and apply one or more previously-trained inference model(s) at that specific orientation and/or positioning in order to predict, for example, user handstate configurations with a higher level of accuracy compared to other inference model(s) that may have been previously trained using different orientations and/or positions of the band on users. Differently stated, in certain embodiments, the armband system can be calibrated via selection of particular inference model(s) such that the system adapts to any rotation and/or arm position offset without interfering with the user experience, also referred to herein as autocalibration. In other embodiments, the detected offset in orientation and/or relative difference in location of the band on the user compared to the orientation and/or positioning of the band used to train one or more inference models can be used to “normalize” the orientation and/or location of the band for input into the inference model(s). A virtual reference electrode (e.g., the “O” electrode) may be defined, and a rotation offset relative to the virtual reference electrode may be used to normalize the orientation and/or location of the band. For example, it may be determined based on the detected and processed EMG signals that the band has a rotational offset of three electrodes from the virtual reference electrode. In such an instance, the detected EMG signals can be further processed to account for this particular offset (either before or after being input into the trained inference models).

In some implementations, the armband system can be autocalibrated such that, the armband system adapts to users that may have injured or missing muscles, different adipose tissue or fat, and other anatomic variables. Although discussed below in the context of multiple trained inference models, it is appreciated that the embodiments discussed herein can in some instances be implemented as a single or sole trained inference model or machine learning model. It is also appreciated that at least some of the inference models may be trained from data collected from multiple users. For instance, data can be collected from recorded EMG signals of multiple users while they perform one or more handstate configurations, e.g., poses or gestures.

In one embodiment, the inference model that classifies EMG patterns can be created as follows: 1) build a new inference model/experiment class that takes as input(s) a set of preprocessed EMG signals; 2) generate training data by randomly applying a rotation offset to the preprocessed EMG signals; 3) produce positive labels when the augmented offset is 0, and null otherwise; 4) calibrate the training data to have calibrated data at offset=O; and 5) train an inference model using the calibrated training data, and evaluate the performance of the trained inference model by testing different rotation offsets.

An example of a distribution of outputs generated by a trained inference model across a dataset with data collected from different users is shown in FIG. 3. It can be appreciated that on average the trained model outputs show higher confidence values when the training data is calibrated (i.e., offset=0) compared to when the training data is not calibrated. In this example, the trained model produces a prediction for every 80 millisecond (ms) chunk of collected EMG data, however, other time intervals can be analogously used. Also, in this example, a binary classifier is used to determine whether the band is calibrated at the “O” electrode in the preferred orientation. FIG. 4 shows predicted values averaged across time which result in predictions with greater confidence values. It can be appreciated that by applying a smoothing factor of 10 seconds the predicted offset can be acquired with more accuracy e.g., an accuracy at the ±1 electrode range. FIG. 5 shows the accuracy of the model expressed in Area Under the Receiver Operating Characteristic Curve (AUC).

In some implementations, an electrode invariance factor associated with a small offset error can be provided as an input to an inference model, e.g., during the training phase. Such an inference model is also referred herein as an augmented inference model. The augmented inference model can be implemented by, for instance, expanding a batch of training data and creating a new dimension for each sample in a training data batch. The new dimension includes concatenations of different versions of the same feature vector with different electrode rotations. For example, given a feature vector of dimension 384, an invariance of ±1 electrode can be set by providing as input to the augmented inference model a feature tensor of dimension 3×384. Such a feature tensor includes the extra dimension containing the same feature vector with the three rotation offsets [−1, 0, I]. Then, the augmented inference model may be initiated with a first input dense layer and this extra dimension may be reduced by pooling across the extra dimension, for instance, by taking the maximum activation over the extra dimension. Accordingly, the augmented inference model applies the same set of coefficients (for the first dense layer) for each of the three offsets, then selects for each filter the largest activation over the three proposed offsets. FIG. 6 illustrates a correlation between a baseline model and the augmented inference model. It should be appreciated that the baseline model is more sensitive to the offset errors, as it sharply drops performance when is not calibrated (i.e., offset !=0). FIG. 7 illustrates correlations of augmented inference models trained with electrode invariances set to ±1, ±2, and ±3 along with the baseline model.

In another embodiment, the autocalibration process comprises numerically aligning the band position (e.g., rotation, location on the forearm, and/or orientation) via common reference point(s) across different users and their forearms and/or wrists, so that the system can identify virtual reference point(s) for each electrode and can consistently represent the same part of the forearm and/or wrist irrespective of the band position. For example, this can be done by applying an electrode permutation (e.g., reversing the order of the electrodes depending on the band orientation as applied to the forearm, or as applied to the left or right forearms) and applying a circular rotation to the electrodes. FIG. 8 schematically shows a visualization of an energy plot for a single EMG electrode “O” at a first position 810, and at a second position 820 in which the EMG electrode is rotated three positions. The disclosed systems and methods herein can utilize the difference(s) in such energy plots in order to achieve the autocalibration results as described herein.

In some embodiments, the user need not perform a specific calibration gesture to register the band position. Instead, an autocalibration model can be trained to recognize and predict the specific electrode permutation that generated a given set of EMG signals or patterns. Data augmentation techniques can be implemented on data sets obtained in a supervised machine learning fashion. For example, offline EMG data can be collected from a set of users who performed a set of canonical gestures or poses comprising various pinches and wrist movements in the four cardinal directions to register the band rotations based on the collected and processed EMG data associated with those gestures or poses. With this process, the system can generate labels to train a model (e.g., an autocalibration model) to recognize different simulated band rotations and/or orientations (e.g., data augmentation). The number of gestures or poses upon which the autocalibration model is trained can vary, but may comprise approximately eight gestures. In this embodiment, the autocalibration task can be regarded as a classification task, and more collected and labeled data may lead to better precision on the predicted offset of the electrode(s). During this offline training, the band can be rotated and additional data can be collected from users in each of the rotated positions. For example, if the band has 16 electrode channels, data can be collected from the same user or across multiple users at each of 16 different orientations. Some embodiments employ an autocalibration model to predict that the wearable system is rotated into one of multiple (e.g., 16) positions.

FIG. 9 shows an example of a flowchart for offline training of an autocalibration model in accordance with some embodiments. In act 910 EMG data is sensed as one or more users are performing multiple gestures or poses with different band orientations. As discussed above, any suitable number of gestures or poses (e.g., 8 gestures or poses) may be used. The process then proceeds to act 912, where templates are extracted for each gesture or pose, wherein each of the templates corresponds to spatio-temporal patterns of the sensed EMG signals. The process then proceeds to act 914, where the extracted templates are used to generate calibrated EMG data. Any given electrode channel can be designated the “O” channel electrode at which the system is set to by default process and interpret EMG inputs to predict handstate configurations.

In act 916, each of the band orientations during data collection is assigned an offset value depending on the degree of rotation of the armband when the data was collected. In this way, each labeled data set collected and processed for the specific gestures and/or poses can be further be labeled with an offset value depending on the band orientation and associated simulation position can be generated for the band. The labeled data for calibrated EMG signals, which have been assigned to gestures and/or poses, and EMG signals identified based on specific electrode channel orientation(s) can be input into an autocalibration model in act 920 to train the model. In some embodiments, the autocalibration model is a neural network (such as an LSTM, for example), and the autocalibration model is trained to predict the orientation of the armband based on received EMG signal data. Once properly trained, the process proceeds to act 922 where the autocalibration model can be implemented for any user to predict the orientation and/or rotation offset of the band (e.g., with respect to the “0” channel) based on received EMG data from the user and simulations of the band positioning and/or orientation based on previously-collected user EMG data. In some embodiments, the model may be trained such that it can predict the degree of rotation or offset to an error of +/−1 electrode channel. After the autocalibration model has been suitably trained to predict the orientation of the band within an error of +/−1 electrode channel, the autocalibration model can be used “online” to predict the band orientation on any given user during the user's performance of one or more motor actions or intended motor actions.

By contrast to techniques that require the user to perform a specific gesture or pose for calibration, some embodiments use autocalibration to automatically detect the position of the wearable device on a body part using an autocalibration model that does not require the user to perform a specific calibration gesture or pose. Such embodiments permit continuous or periodic autocalibration of the device as the user wears the device without requiring the user to stop what they are doing and perform a particular calibration gesture or pose. FIG. 10 illustrates a flowchart for online usage of a trained autocalibration model in accordance with some embodiments. In act 1010, EMG data is sensed from a user wearing the wearable device (e.g., the wearable device shown in FIG. 1). In act 1012, the sensed EMG data is provided to a trained autocalibration model (e.g., the autocalibration model trained in act 920 of the process in FIG. 9). The process of FIG. 10 then proceeds to act 1014 where the output of the model is used to predict a position and/or orientation of the wearable device by, for example, classifying the EMG signal or signal patterns into one of multiple (e.g., 16) rotation positions. The process then proceeds to act 1016 where a correction based on the predicted output of the autocalibration model is applied to the sensed EMG data to generate calibrated EMG data. In other embodiments, the sensed EMG data is not corrected, and the sensed EMG data is provided as input to selected inferential models trained at the detected orientation(s) and/or position(s) or provided as input to models that have been modified to interpret the sensed EMG data and accurately output handstate configurations, poses, and/or gestures.

As described above, in some embodiments, the autocalibration model may be implemented as an LSTM neural network. In certain embodiments, it is desirable to address the +/−1 electrode channel error rate for the rotation offset (see FIG. 6 showing an augmented model performing equally well for +/−1 electrode offset, but the baseline model performing not as well for a +/−1 electrode offset). This can be done by implementing a pooling layer within the neural network as shown schematically in act 1020 of FIG. 10. For example, the pooling layer may be a max pooling layer where the max pooling layer can be configured to determine the location of the “0” electrode channel (e.g., based on a probabilistic calculation for each of the three electrode locations), so the system can virtually label a specific electrode channel as the “0” reference channel out of a possible set of three channels initially identified by the autocalibration model. It should be understood that other types of pooling layers can be used that are well known in the art, e.g., average or mean pooling layers.

In certain embodiments, the band detects and analyzes EMG signals or EMG signal patterns either continuously for a certain period of time or at discrete time points. In either case, during this defined period of time or at discrete time points, the autocalibration model(s) can be used to determine the orientation and/or positioning of the band on the user's body part, e.g., forearm. In certain embodiments, the band detects and processes EMG signals or EMG signal patterns initially upon the user putting the band on and/or in response to detecting band movement subsequent to initial placement of the band. For example, the system may identify that at least a portion of the wearable device (e.g., one or more EMG sensors) has moved, e.g., rotated around and/or translated along the wrist or forearm of the user, and in response to determining that the at least a portion of the wearable device (e.g., the band) has moved, the system may perform autocalibration by determining a current position and/or orientation of the wearable device in accordance with one or more of the autocalibration techniques described herein. By automatically restarting calibration after identifying movement of the wearable device, re-calibration can be performed in an efficient and timely manner.

Movement of the wearable device on the user can be detected in any suitable manner. In some embodiments, detection of movement of the wearable device is based on one or more detected movement artifacts identified in the neuromuscular signals sensed by the plurality of neuromuscular sensors. In other embodiments that include at least one inertial measurement unit (IMU) sensor, movement of the wearable device may be detected based at least in part, on at least one signal sensed by the at least one IMU sensor. In yet other embodiments, detection of movement of the wearable device may be based, at least in part, on signals from multiple types of sensors (e.g., at least one EMG sensor and at least one IMU sensor) and/or sensors included on or external to the wearable device (e.g., at least one camera sensor).

Using the techniques described herein, the system may be able to calibrate the orientation and/or position of the armband in less than 10 seconds based on one or more sampling rates or frequencies. In other embodiments, due to settling time of the EMG electrodes (e.g., from skin-electrode interactions) after band movements, which can lead to an initially degraded signal quality, it may take a little longer to get a reliable estimation of the band position. For example, the system can initiate an autocalibration sequence upon detection of band movement, and the sequence can run for approximately 30 seconds to get a more reliable prediction of band orientation and/or position. Shorter time periods can be used to run the autocalibration sequences provided that the settling time can be reduced and little to no band movement is detected during the autocalibration sequence. In certain embodiments, the autocalibration sequence can be initiated and re-run upon the detection of any band movements no matter how slight in order to maximize the accuracy of the predictions associated with EMG signals or EMG signal patterns.

In other embodiments, the techniques described herein can be used to identify a specific position on the user's body part, e.g., how far up the band is sitting on the user's forearm as it relates to the distances between the band and the user's elbow joint or wrist joint. Similar to the embodiment described above (e.g., in FIG. 9) in which a generalized, offline model is generated and trained from labeled EMG signal or EMG signal pattern data with users performing different gestures and/or poses at different orientations of the band, in other embodiments additional data can be collected from users performing gestures and/or poses at different or the same orientations of the band but at different positions on the user's body part (e.g., along portions of the forearm). In this way, one or more generalized models can be trained to predict not only the specific orientation (e.g., rotation) of the band, but also its positioning on the user's body part (e.g., relative to a reference point such as the elbow joint or wrist joint). Such embodiments can be used to better predict user handstate configurations given potential variability in EMG signals or EMG signal patterns depending on the specific location of the band on the user. Given neuroanatomical constraints and specific user neuroanatomy, it may be advantageous to analyze EMG signals on more distal or more proximal points of the user's body part (e.g., along different portions of the forearm or wrist). In certain embodiments, the autocalibration model can detect a specific positioning of the band on the user's forearm, and generate an input signal to an interface to instruct the user to reposition the band in a more distal or more proximal location depending on the specific outcome desired (e.g., to move the band farther up the arm so that relatively more EMG data can be collected or farther down the arm because certain input models to be used were trained on collected data from users with the band closer to their wrists).

While various embodiments have been described above, it should be understood that they have been presented by way of example only, and not limitation. Where methods and/or schematics described above indicate certain events and/or flow patterns occurring in certain order, the ordering of certain events and/or flow patterns may be modified. While the embodiments have been particularly shown and described, it will be understood that various changes in form and details may be made. Additionally, certain of the steps may be performed concurrently in a parallel process when possible, as well as performed sequentially as described above. Although various embodiments have been described as having particular features and/or combinations of components, other embodiments are possible having any combination or sub-combination of any features and/or components from any of the embodiments described herein. Furthermore, although various embodiments are described as having a particular entity associated with a particular compute device, in other embodiments different entities can be associated with other and/or different compute devices.

It is intended that the systems and methods described herein can be performed by software (stored in memory and/or executed on hardware), hardware, or a combination thereof. Hardware modules may include, for example, a general-purpose processor, a field programmable gates array (FPGA), and/or an application specific integrated circuit (ASIC). Software modules (executed on hardware) can be expressed in a variety of software languages (e.g., computer code), including Unix utilities, C, C++, Java™, JavaScript, Ruby, SQL, SAS®, Python, Fortran, the R programming language/software environment, Visual Basic™, and other object-oriented, procedural, or other programming language and development tools. Examples of computer code include, but are not limited to, micro-code or micro-instructions, machine instructions, such as produced by a compiler, code used to produce a web service, and files containing higher-level instructions that are executed by a computer using an interpreter. Additional examples of computer code include, but are not limited to, control signals, encrypted code, and compressed code. Each of the devices described herein can include one or more processors as described above.

Some embodiments described herein relate to devices with a non-transitory computer-readable medium (also can be referred to as a non-transitory processor-readable medium or memory) having instructions or computer code thereon for performing various computer-implemented operations. The computer-readable medium (or processor-readable medium) is non-transitory in the sense that it does not include transitory propagating signals per se (e.g., a propagating electromagnetic wave carrying information on a transmission medium such as space or a cable). The media and computer code (also can be referred to as code) may be those designed and constructed for the specific purpose or purposes. Examples of non-transitory computer-readable media include, but are not limited to: magnetic storage media such as hard disks, floppy disks, and magnetic tape; optical storage media such as Compact Disc/Digital Video Discs (CD/DVDs), Compact Disc-Read Only Memories (CD-ROMs), and holographic devices; magneto-optical storage media such as optical disks; carrier wave signal processing modules; and hardware devices that are specially configured to store and execute program code, such as Application-Specific Integrated Circuits (ASICs), Programmable Logic Devices (PLDs), Read-Only Memory (ROM) and Random-Access Memory (RAM) devices. Other embodiments described herein relate to a computer program product, which can include, for example, the instructions and/or computer code discussed herein.

Processor-executable instructions can be in many forms, such as program modules, executed by one or more compute devices, and can include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular data types, and the functionality can be combined and/or distributed as appropriate for various embodiments. Data structures can be stored in processor-readable media in a number of suitable forms. For simplicity of illustration, data structures may be shown to have fields that are related through location in the data structure. Such relationships may likewise be achieved by assigning storage for the fields with locations in a processor-readable medium that conveys relationship(s) between the fields. However, any suitable mechanism/tool can be used to establish a relationship between information in fields of a data structure, including through the use of pointers, tags or other mechanisms/tools that establish relationship between data elements.

Various disclosed concepts can be embodied as one or more methods, of which examples have been provided. The acts performed as part of a particular method can be ordered in any suitable way. Accordingly, embodiments can be constructed in which acts are performed in an order different than illustrated/discussed, which can include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments. The use of flow diagrams is not meant to be limiting with respect to the order of operations performed. The herein described subject matter sometimes illustrates different components contained within, or connected with, different other components. It is to be understood that such depicted architectures are merely exemplary, and that in fact many other architectures can be implemented which achieve the same functionality. In a conceptual sense, any arrangement of components to achieve the same functionality is effectively “associated” such that the desired functionality is achieved. Hence, any two components herein combined to achieve a particular functionality can be seen as “associated with” each other such that the desired functionality is achieved, irrespective of architectures or intermediate components. Likewise, any two components so associated can also be viewed as being “operably connected,” or “operably coupled,” to each other to achieve the desired functionality, and any two components capable of being so associated can also be viewed as being “operably couplable,” to each other to achieve the desired functionality. Specific examples of operably couplable include but are not limited to physically malleable and/or physically interacting components and/or wirelessly interactable and/or wirelessly interacting components and/or logically interacting and/or logically interactable components.

All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and/or ordinary meanings of the defined terms.

The indefinite articles “a” and “an,” as used herein in the specification and in the claims, unless clearly indicated to the contrary, should be understood to mean “at least one.”

The phrase “and/or,” as used herein in the specification and in the claims, should be understood to mean “either or both” of the elements so conjoined, i.e., elements that are conjunctively present in some cases and disjunctively present in other cases. Multiple elements listed with “and/or” should be construed in the same fashion, i.e., “one or more” of the elements so conjoined. Other elements may optionally be present other than the elements specifically identified by the “and/or” clause, whether related or unrelated to those elements specifically identified. Thus, as a non-limiting example, a reference to “A and/or B”, when used in conjunction with open-ended language such as “comprising” can refer, in one embodiment, to A only (optionally including elements other than B); in another embodiment, to B only (optionally including elements other than A); in yet another embodiment, to both A and B (optionally including other elements); etc.

As used herein in the specification and in the claims, “or” should be understood to have the same meaning as “and/or” as defined above. For example, when separating items in a list, “or” or “and/or” shall be interpreted as being inclusive, i.e., the inclusion of at least one, but also including more than one, of a number or list of elements, and, optionally, additional unlisted items. Only terms clearly indicated to the contrary, such as “only one of” or “exactly one of,” or, when used in the claims, “consisting of,” will refer to the inclusion of exactly one element of a number or list of elements. In general, the term “or” as used herein shall only be interpreted as indicating exclusive alternatives (i.e. “one or the other but not both”) when preceded by terms of exclusivity, such as “either,” “one of,” “only one of,” or “exactly one of.” “Consisting essentially of,” when used in the claims, shall have its ordinary meaning as used in the field of patent law.

As used herein in the specification and in the claims, the phrase “at least one,” in reference to a list of one or more elements, should be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each and every element specifically listed within the list of elements and not excluding any combinations of elements in the list of elements. This definition also allows that elements may optionally be present other than the elements specifically identified within the list of elements to which the phrase “at least one” refers, whether related or unrelated to those elements specifically identified. Thus, as a non-limiting example, “at least one of A and B” (or, equivalently, “at least one of A or B,” or, equivalently “at least one of A and/or B”) can refer, in one embodiment, to at least one, optionally including more than one, A, with no B present (and optionally including elements other than B); in another embodiment, to at least one, optionally including more than one, B, with no A present (and optionally including elements other than A); in yet another embodiment, to at least one, optionally including more than one, A, and at least one, optionally including more than one, B (and optionally including other elements); etc.

In the claims, as well as in the specification above, all transitional phrases such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” “composed of,” and the like are to be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases “consisting of” and “consisting essentially of” shall be closed or semi-closed transitional phrases, respectively, as set forth in the United States Patent Office Manual of Patent Examining Procedures, Section 2111.03.

Claims

1. A system comprising: a plurality of sensors configured to continuously sense neuromuscular signals from a body part of a user; andat least one computer processor programmed to: determine based, at least in part, on the neuromuscular signals sensed from the body part of the user, a current position and/or orientation of the plurality of sensors on the user's body part;present a user interface to the user indicating a current position and/or orientation of the plurality of sensors; andgenerate a control signal based, at least in part, on the current position and/or orientation of the plurality of sensors on the user's body part, wherein the control signal causes the user interface to instruct the user to variably reposition the plurality of sensors to a different position on the user's body part based on the determined current position and/or orientation and further based on which position and/or orientation of the user's body part is associated with a specific set of training data generated from a specific input model.

2. The system of claim 1, wherein the plurality of sensors comprises a wearable device configured to be worn on a wrist or forearm of the user.

3. The system of claim 2, further comprising providing the plurality of neuromuscular signals as inputs to one or more trained autocalibration models to calibrate the position and/or orientation of the wearable device.

4. The system of claim 1, wherein the current position and/or orientation are determined based on one or more specified neuromuscular signals or patterns that indicate a distal or proximal position of the user's body part on which the plurality of sensors is located.

5. The system of claim 4, wherein the control signal causes the user interface to instruct the user to reposition the plurality of sensors to a more distal or more proximal location on the user's body part based on the determined current position and/or orientation.

6. The system of claim 5, wherein the instruction to reposition the plurality of sensors to a more distal or more proximal location on the user's body part is based on and is specific to the neuroanatomy of the user's body part.

7. The system of claim 4, further comprising: determining that an increased amount of data is needed from the plurality of sensors; andcausing the user interface, via the control signal, to instruct the user to reposition the plurality of sensors to a more proximal location on the user's body part.

8. The system of claim 4, further comprising: determining that additional training data is available for the plurality of sensors at a distal location on the user's body part; andcausing the user interface, via the control signal, to instruct the user to reposition the plurality of sensors to a more distal location on the user's body part.

9. The system of claim 1, wherein the at least one computer processor is programmed to determine the current position and/or orientation of the plurality of sensors on the user without the user performing a particular pose or gesture during sensing of the plurality of neuromuscular signals.

10. The system of claim 1, wherein the at least one programmed processor is further programmed to: generate, using an autocalibration model, one or more calibrated neuromuscular signals based, at least in part, on the current position and/or orientation of the plurality of sensors, wherein generating the control signal comprises generating a control signal based, at least in part, on the calibrated neuromuscular signals.

11. The system of claim 10, wherein the autocalibration model used to generate the calibrated neuromuscular signals comprises a neural network.

12. The system of claim 1, wherein the control signal is configured to control one or more operations of or within a virtual reality system or an augmented reality system.

13. A computer-implemented method comprising: determining based, at least in part, on one or more neuromuscular signals sensed from a body part of a user wearing a plurality of sensors, a current position and/or orientation of the plurality of sensors on the user's body part;presenting a user interface to the user indicating a current position and/or orientation of the plurality of sensors; andgenerating a control signal based, at least in part, on the current position and/or orientation of the plurality of sensors on the user's body part, wherein the control signal causes the user interface to variably instruct the user to reposition the plurality of sensors to a different position on the user's body part based on the determined current position and/or orientation and further based on which position and/or orientation of the user's body part is associated with a specific set of training data generated from a specific input model.

14. The method of claim 13, wherein the plurality of sensors comprises a wearable device configured to be worn on a wrist or forearm of the user.

15. The method of claim 14, further comprising providing the plurality of neuromuscular signals as inputs to one or more trained autocalibration models to calibrate the position and/or orientation of the wearable device.

16. The method of claim 13, wherein the current position and/or orientation are determined based on one or more specified neuromuscular signals or patterns that indicate a distal or proximal position of the user's body part on which the plurality of sensors is located.

17. The method of claim 16, wherein the control signal causes the user interface to instruct the user to reposition the plurality of sensors to a more distal or more proximal location on the user's body part based on the determined current position and/or orientation.

18. The method of claim 17, wherein the instruction to reposition the plurality of sensors to a more distal or more proximal location on the user's body part is based on and is specific to the neuroanatomy of the user's body part.

19. The method of claim 16, further comprising: determining that an increased amount of data is needed from the plurality of sensors; andcausing the user interface, via the control signal, to instruct the user to reposition the plurality of sensors to a more proximal location on the user's body part.

20. A wearable device comprising: a plurality of sensors configured to continuously sense neuromuscular signals from a body part of a user; andat least one computer processor programmed to:determine based, at least in part, on the neuromuscular signals sensed from the body part of the user, a current position and/or orientation of the plurality of sensors on the user's body part;present a user interface to the user indicating a current position and/or orientation of the plurality of sensors; andgenerate a control signal based, at least in part, on the current position and/or orientation of the plurality of sensors on the user's body part, wherein the control signal causes the user interface to variably instruct the user to reposition the plurality of sensors to a different position on the user's body part based on the determined current position and/or orientation and further based on which position and/or orientation of the user's body part is associated with a specific set of training data generated from a specific input model.