Biomechanical and physiological state estimation for task agnostic wearable robot control and human monitoring

Patent metadata

Field	Value
Publication number	US-12496246-B1
Application number	US-202418945287-A
Country	US
Kind code	B1
Filing date	Nov 12, 2024
Priority date	Nov 12, 2024
Publication date	Dec 16, 2025
Grant date	Dec 16, 2025

Abstract

Exemplary task-agnostic exoskeleton control system and method are disclosed that are task-agnostic utilizing instantaneous estimates of biological joint moments from deep neural networks to assist the user movements. The exemplary control system employs multiple body sections and joints in-the-loop estimation to provide multi-joint assistance operation, e.g., within an autonomous, clothing-integrated exoskeleton. The exemplary control system may deploy a deep domain adaptation (DDA) method configured to translate human movement data between a simulated sensor domain and a real sensor domain.

First claim

What is claimed:

1 . A system comprising:
a wearable device having one or more sensors; a processor; and a memory having instructions stored thereon for a controller having a joint moment estimator, a muscle moment estimator, or a force estimator, wherein the processor is configured to execute the instructions, which include:
receiving, via a control loop, sensor data from the one or more sensors of the wearable device;
determining, via a trained AI model, at least one of a physical state, muscle extension moments, or muscle extension forces using the sensor data as a direct input to the trained AI model;
outputting, from the trained AI model, and via the control loop, joint moment, muscle moment, or force estimates from the physical state, muscle extension moments, or muscle extension forces for operating the wearable device based on the determined at least one of the physical state, muscle extension moments, or muscle extension forces; and
determining a desired torque assistance value by:
scaling joint moments estimates by an assistance offset value; and
delaying scaled joint moments estimates by a pre-defined offset value.

2 . The system of claim 1 , wherein the one or more sensors comprise a plurality of exoskeleton sensors and the trained AI model was trained using time-synchronized exoskeleton sensor data and labels, where the sensor data is received from the plurality of exoskeleton sensors and labels come from joint moment measurements or estimates for a number of cyclic and non-cyclic human activities.

3 . The system of claim 2 , wherein the cyclic and non-cyclic human activities include at least 5 motions selected from the group consisting of: level ground walk, standing in various poses, push and pull, turn, change of direction, sit and stand, tug of war, jump across, toe and heel walk, to stand up from a seated position, vertical jump, lunge, walk, skipping, medicine ball toss, lift a weight, step up, butt kicks, walk backward,
wherein the cyclic and non-cyclic human activities were measured using hip position and configuration, shank position and configuration, thigh position and configuration, trunk and foot position and configuration, and knee position and configuration.

4 . The system of claim 2 , wherein the cyclic activities include walking and running at different configurations.

5 . The system of claim 2 , wherein the non-cyclic activities include impedance action and unstructured action, wherein the impedance action includes at least one of standing in various poses, lunges, sit and stand, tug of war, medicine ball toss, step up, jump across, jump in place, lift and place weight, and squats, and wherein the unstructured action includes at least one of start and stop, change of direction, step over, turn, meander, twisting side to side at a waist of a human while standing without moving the human's feet, push and pull recovery, and systematic rhythmic bodily exercises performed without an apparatus.

6 . The system of claim 1 , wherein the controller is implemented in a remote co-processor.

7 . The system of claim 1 , wherein the wearable device comprises an actuator and an inertial measurement unit (IMU), wherein the one or more sensors comprise an insole sensor, and wherein the joint moment estimator or the force estimator is implemented in a remote co-processor to determine total hip and knee flexion/extension moment values using measured actuator data, measured IMU data, and measured pressure insole data, as the sensor data, acquired from the wearable device.

8 . The system of claim 1 , wherein the execution of the instructions by the processor further causes the processor to determine the desired torque assistance value further comprises:
modulating a sharpness and a flatness of peaks of the delayed joint moment estimates.

9 . The system of claim 1 , wherein the trained AI model comprises a temporal convolutional neural network (TCN).

10 . The system of claim 1 , wherein the wearable device comprises a robotic device or prosthetic device.

11 . A method comprising:
receiving, via a control loop, sensor data from one or more sensors of a wearable device; determining, via a trained AI model, at least one of a physical state, muscle extension moments, or muscle extension forces using the sensor data as a direct input to the trained AI model; outputting, from the trained AI model, and via the control loop, joint moment, muscle moment, or force estimates from the physical state, muscle extension moments, or muscle extension forces for operating the wearable device based on the determined at least one of the physical state, muscle extension moments, or muscle extension forces; and determining a desired torque assistance value by scaling biological joint moments estimates by an assistance offset value, and delaying the scaled biological joint moment estimates by a pre-defined offset value.

12 . The system of claim 1 , wherein the controller is configured by:
receiving biomechanical data; generating simulated sensor data using a translator network to simulate sensors for a transferred device using the received biomechanical data generated using physics-based simulation software, where the simulation is performed at body segments for positions, for forces, and/or for velocities constrained or defined on the same body segment for a target device to provide paired data of (i) the sensor data of the target device and (ii) the biomechanical data, wherein the paired data is used for training of at least one neural network; and training the at least one neural network, wherein the neural network is configured to map the simulated data associated with the biomechanical data to real data associated with the sensor data.

13 . The system of claim 12 , wherein the at least one neural network forms in part:
a first generative adversarial network (GAN) configured to generate the real data associated with the sensor data; or a second GAN configured to generate simulated data associated with the biomechanical data.

14 . The system of claim 13 , wherein the training of the at least one neural network employs supervised loss, cycle consistency loss, GAN loss, and reconstruction loss,
wherein the reconstruction loss is computed using error value normalized by sensor modality weighted equally in reconstruction, and wherein the cycle consistency loss is computed by matching (i) simulated data associated with the biomechanical data that was passed through both a sim-to-real translator and then a real-to-sim translator to (ii) the original data.

15 . The method of claim 11 , wherein the wearable device comprises a robotic device or prosthetic device.

16 . The method of claim 11 , wherein the one or more sensors comprise a plurality of exoskeleton sensors and the trained AI model was trained using time-synchronized exoskeleton sensor data and labels, where the sensor data is received from the plurality of exoskeleton sensors and labels come from joint moment measurements or estimates for a number of cyclic and non-cyclic human activities.

17 . The method of claim 16 , wherein the time-synchronized exoskeleton sensor data and labels for training the AI model is comprised of a set of tasks that promote task generalization, which were selected by a task selection algorithm.

18 . The method of claim 16 , further comprising selecting one or more of the plurality of exoskeleton sensors to provide the exoskeleton sensor data.

19 . The method of claim 11 , further comprising: wherein determin

Assignees

• Georgia Tech Res Inst

Inventors

• Molinaro Dean
• Young Aaron
• Scherpereel Keaton

Classifications

• A61H1/024

  Knee (A61H1/0255 takes precedence) · CPC title

• A61H2201/165

  Wearable interfaces · CPC title

• A61H2230/625

  used as a control parameter for the apparatus · CPC title

• A61H3/00 Primary

  Appliances for aiding patients or disabled persons to walk about (apparatus for helping babies to walk A47D13/04 {; applying electrical currents by contact electrodes for stimulating motor muscles, e.g. walking assistance A61N1/36003}) · CPC title

• A61H2201/5007

  computer controlled · CPC title