IP Library › Granted Patent US 12,569,358
Granted Patent B2
US 12,569,358 · App. 17/535,298 · Granted Mar 10, 2026

Predicting user preference with AI to control a new generation of wearable assistive technologies

Inventors: Elliott J. Rouse (Ann Arbor, MI); Ung Hee Lee (Ann Arbor, MI); Varun Shetty (Ann Arbor, MI)
A61F2/70G05B13/0265A61F2002/6614A61F2002/704
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,569,358
App. No.
17/535,298
Granted
Mar 10, 2026
Kind
B2
Abstract

Techniques are provided for using machine learning methods to predict user preferences with respect to robotic assistive prostheses and/or custom tune the robotic assistive prostheses for individual users based on the predicted user preferences. A training biomechanical dataset, including historical biomechanical sensor data for a robotic assistive prosthetic device operating using a plurality of tuning settings at a plurality of speeds, and a training user preference dataset including historical user tuning preference data for the robotic assistive prosthetic device for each respective tuning setting and speed, are generated. A machine learning model is trained using the training biomechanical dataset and the training user preference dataset. The trained machine learning model is applied to new biomechanical sensor data associated with a particular user to predict the user's tuning preferences for the robotic assistive prosthetic device, and the settings of the device are automatically modified based on the predicted tuning preferences.

Claims (35)

1 . A computer-implemented method, comprising:

generating, by one or more processors of a system including a robotic assistive device, a training biomechanical dataset, including historical biomechanical sensor data for a plurality of users operating the robotic assistive device using a plurality of tuning settings at a plurality of speeds;

generating, by the one or more processors, a training user preference dataset including historical user tuning preference data for the plurality of users for each respective tuning setting and speed of the robotic assistive device;

training, by the one or more processors, using the training biomechanical dataset and the training user preference dataset, a robotic assistive device tuning preference machine learning model to predict user tuning preferences for the robotic assistive device based on biomechanical sensor data for the robotic assistive device associated with a user;

obtaining, by the one or more processors, new biomechanical sensor data, associated with a particular user, for the robotic assistive device operating using the plurality of tuning settings at the plurality of speeds;

applying, by the one or more processors, the trained robotic assistive device tuning preference machine learning model to the new biomechanical sensor data associated with the particular user to predict user tuning preference data for the robotic assistive device; and

automatically modifying, by the one or more processors, one or more tuning settings for the robotic assistive prosthetic device based on the predicted user tuning preference data, wherein automatically modifying the one or more tuning settings for the robotic assistive device includes causing a motor to adjust one or more mechanical properties of the robotic assistive device.

2 . The computer-implemented method of claim 1 , wherein at least a portion of the new biomechanical sensor data is captured by onboard sensors of the robotic assistive device.

3 . The computer-implemented method of claim 1 , wherein at least a portion of the historical biomechanical sensor data is captured by onboard sensors of the robotic assistive device.

4 . The computer-implemented method of claim 1 , wherein the new biomechanical sensor data and the historical biomechanical sensor data each include one or more of: bilateral three degree-of-freedom (3-DOF) hip angle data, sagittal knee angle data, sagittal ankle angle data, bilateral 3-DOF hip moment data, sagittal knee moment data, and sagittal ankle moment data, bilateral 3-DOF ground reaction force, anterior center of pressure (COP) data, posterior COP data, mediolateral COP data, 3-DOF position of pelvis data, 3-DOF position of bilateral foot data, 3-DOF position of shank data, 3-DOF position of thigh data, 3-DOF velocity of pelvis data, 3-DOF velocity of bilateral foot data, 3-DOF velocity of shank data, 3-DOF velocity of thigh data, 3-DOF acceleration of pelvis data, 3-DOF acceleration of bilateral foot data, 3-DOF acceleration of shank data, or 3-DOF acceleration of thigh data.

5 . The computer-implemented method of claim 1 , wherein new biomechanical sensor data and historical sensor data are captured for each stride of the robotic assistive device.

6 . A system comprising:

a robotic assistive device;

one or more processors; and

a non-transitory memory storing computer-readable instructions that, when executed by the one or more processors, cause the one or more processors to:

generate a training biomechanical dataset, including historical biomechanical sensor data for a plurality of users operating the robotic assistive device using a plurality of tuning settings at a plurality of speeds;

generate a training user preference dataset including historical user tuning preference data for the plurality of users for each respective tuning setting and speed of the robotic assistive device;

train, using the training biomechanical dataset and the training user preference dataset, a robotic assistive device tuning preference machine learning model to predict user tuning preferences for the robotic assistive device based on biomechanical sensor data for the robotic assistive device associated with a user;

obtain new biomechanical sensor data, associated with a particular user, for the robotic assistive device operating using the plurality of tuning settings at the plurality of speeds;

apply the trained robotic assistive device tuning preference machine learning model to the new biomechanical sensor data associated with the particular user to predict user tuning preference data for the robotic assistive device; and

automatically modify one or more tuning settings for the robotic assistive device based on the predicted user tuning preference data, wherein automatically modifying the one or more tuning settings for the robotic assistive device includes causing a motor to adjust one or more mechanical properties of the robotic assistive device.

7 . The system of claim 6 , wherein the robotic assistive device includes one or more onboard sensors configured to capture at least a portion of the new biomechanical sensor data.

8 . The system of claim 6 , wherein the robotic assistive device includes one or more onboard sensors configured to capture at least a portion of the historical biomechanical sensor data.

9 . The system of claim 6 , wherein the new biomechanical sensor data and the historical biomechanical sensor data each include one or more of: bilateral three degree-of-freedom (3-DOF) hip angle data, sagittal knee angle data, sagittal ankle angle data, bilateral 3-DOF hip moment data, sagittal knee moment data, and sagittal ankle moment data, bilateral 3-DOF ground reaction force, anterior center of pressure (COP) data, posterior COP data, mediolateral COP data, 3-DOF position of pelvis data, 3-DOF position of bilateral foot data, 3-DOF position of shank data, 3-DOF position of thigh data, 3-DOF velocity of pelvis data, 3-DOF velocity of bilateral foot data, 3-DOF velocity of shank data, 3-DOF velocity of thigh data, 3-DOF acceleration of pelvis data, 3-DOF acceleration of bilateral foot data, 3-DOF acceleration of shank data, or 3-DOF acceleration of thigh data.

10 . The system of claim 6 , wherein new biomechanical sensor data and historical sensor data are captured for each stride of the robotic assistive prosthetic device.

11 . A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of a system including a robotic assistive device, cause the one or more processors to:

generate a training biomechanical dataset, including historical biomechanical sensor data for a plurality of users operating the robotic assistive device using a plurality of tuning settings at a plurality of speeds;

generate a training user preference dataset including historical user tuning preference data for the plurality of users for each respective tuning setting and speed of the robotic assistive device;

train, using the training biomechanical dataset and the training user preference dataset, a robotic assistive device tuning preference machine learning model to predict user tuning preferences for the robotic assistive device based on biomechanical sensor data for the robotic assistive device associated with a user;

obtain new biomechanical sensor data, associated with a particular user, for the robotic assistive device operating using the plurality of tuning settings at the plurality of speeds;

apply the trained robotic assistive device tuning preference machine learning model to the new biomechanical sensor data associated with the particular user to predict user tuning preference data for the robotic assistive device; and

automatically modify one or more tuning settings for the robotic assistive device based on the predicted user tuning preference data, wherein automatically modifying the one or more tuning settings for the robotic assistive device includes causing a motor to adjust one or more mechanical properties of the robotic assistive device.

12 . The non-transitory computer-readable medium of claim 11 , wherein at least a portion of the new biomechanical sensor data is captured by onboard sensors of the robotic assistive device.

13 . The non-transitory computer-readable medium of claim 11 , wherein at least a portion of the historical biomechanical sensor data is captured by onboard sensors of the robotic assistive device.

14 . The non-transitory computer-readable medium of claim 11 , wherein the new biomechanical sensor data and the historical biomechanical sensor data each include one or more of: bilateral three degree-of-freedom (3-DOF) hip angle data, sagittal knee angle data, sagittal ankle angle data, bilateral 3-DOF hip moment data, sagittal knee moment data, and sagittal ankle moment data, bilateral 3-DOF ground reaction force, anterior center of pressure (COP) data, posterior COP data, mediolateral COP data, 3-DOF position of pelvis data, 3-DOF position of bilateral foot data, 3-DOF position of shank data, 3-DOF position of thigh data, 3-DOF velocity of pelvis data, 3-DOF velocity of bilateral foot data, 3-DOF velocity of shank data, 3-DOF velocity of thigh data, 3-DOF acceleration of pelvis data, 3-DOF acceleration of bilateral foot data, 3-DOF acceleration of shank data, or 3-DOF acceleration of thigh data.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 21, 2021
From: ROUSE, ELLIOTT J.; LEE, UNG HEE; SHETTY, VARUN
To: THE REGENTS OF THE UNIVERSITY OF MICHIGAN
Reel/Frame 058447/0858 →
Continuity (2)
Provisional Application 63118556 · Nov 25, 2020
Related Publication 20220160523A1 · May 26, 2022
References Cited (43)
US 11278235B2 · Herr · 2022 [cited by examiner]
US 20160331557A1 · Tong · 2016 [cited by examiner]
US 20220160523A1 · Rouse · 2022 [cited by examiner]
US 20220257389A1 · Herr · 2022 [cited by examiner]
Ye J. et al. “Functional electrical stimulation based on a pelvis support robot for gait rehabilitation of hemiplegic patients after stroke,” [cited by applicant]
Johnson DC et al. “The evolution of gait in childhood and adolescent cerebral palsy.” J Pediatr Orthop. 1997;17: 392-396. [cited by applicant]
Waters RL et al. “Energy cost of walking of amputees: the influence of level of amputation.” J Bone Joint Surg Am. 1976;58: 42-46. [cited by applicant]
Naschitz JE et al. “Why traumatic leg amputees are at increased risk for cardiovascular diseases.” QJM. 2008;101: 251-259. [cited by applicant]
Breakey JW. Body Image: The Lower-Limb Amputee. J Prosthet Orthot. 1997;9: 58. [cited by applicant]
Rouse EJ, Mooney LM, Martinez-Villalpando EC, Herr HM. Clutchable series-elastic actuator: design of a robotic knee prosthesis for minimum energy consumption. IEEE Int Conf Rehabil Robot. 2013;2013: 6650383. [cited by applicant]
Zhang J, Fiers P, Witte KA, Jackson RW, Poggensee KL, Atkeson CG, et al. Human-in-the-loop optimization of exoskeleton assistance during walking. Science. 2017;356: 1280-1284. [cited by applicant]
Mooney LM, Rouse EJ, Herr HM. Autonomous exoskeleton reduces metabolic cost of human walking during load carriage. Journal of NeuroEngineering and Rehabilitation. 2014. p. 80. doi:10.1186/1743-0003-11-80. [cited by applicant]
Tucker MR, Olivier J, Pagel A, Bleuler H, Bouri M, Lambercy O, et al. Control strategies for active lower extremity prosthetics and orthotics: a review. J Neuroeng Rehabil. 2015;12: 1. [cited by applicant]
Tyler R. Clites, Max K. Shepherd, Kimberly A. Ingraham, and Elliott J. Rouse. Patient Preference in the Selection of Prosthetic Joint Stiffness. Proc IEEE RAS EMBS Int Conf Biomed Robot Biomechatron. Dec. 2020. [cited by applicant]
Shepherd MK, Rouse EJ. Comparing preference of ankle-foot stiffness in below-knee amputees and prosthetists. Scientific Reports. 2020. doi:10.1038/s41598-020-72131-2. [cited by applicant]
Delp SL, Anderson FC, Arnold AS, Loan P, Habib A, John CT, et al. OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEE Trans Biomed Eng. 2007;54: 1940-1950. [cited by applicant]
Young AJ, Hargrove LJ. A Classification Method for User-Independent Intent Recognition for Transfemoral Amputees Using Powered Lower Limb Prostheses. IEEE Transactions on Neural Systems and Rehabilitation Engineering. 2… [cited by applicant]
Lee UH, Bi J, Patel R, Fouhey D, Rouse, EJ. Image Transformation and CNNs: A Strategy for Encoding Human Locomotor Intent for Autonomous Wearable Robots. Robotics Automation Letters. [cited by applicant]
R. Waters, J. Perry, D. Antonelli, and H. Hislop, “Energy cost of walking of amputees: the influence of level of amputation,” J Bone Joint Surg Am, vol. 58, No. 1, pp. 42-46, 1976. [cited by applicant]
T. Lenzi, M. Cempini, L. J. Hargrove, and T. A. Kuiken, “Design, development, and validation of a lightweight nonbackdrivable robotic ankle prosthesis,” [cited by applicant]
E. M. Glanzer and P. G. Adamczyk, “Design and validation of a semi-active variable stiffness foot prosthesis,” [cited by applicant]
M. K. Shepherd and E. J. Rouse, “The vspa foot: A quasi-passive ankle-foot prosthesis with continuously variable stiffness,” [cited by applicant]
D. J. Braun, V. Chalvet, T.-H. Chong, S. S. Apte, and N. Hogan, “Variable stiffness spring actuators for low-energy-cost human augmentation,” [cited by applicant]
H. Tryggvason, F. Starker, C. Lecomte, and F. Jonsdottir, “Variable stiffness prosthetic foot based on rheology properties of shear thickening fluid,” [cited by applicant]
J. Zhang, et al., “Human-in-the-loop optimization of exoskeleton assistance during walking,” [cited by applicant]
K. A. Ingraham, C. D. Remy, and E. J. Rouse, “User preference of applied torque characteristics for bilateral powered ankle exoskeletons,” in [cited by applicant]
N. Thatte, H. Duan, and H. Geyer, “A sample-efficient black-box optimizer to train policies for human-in-the-loop systems with user preferences,” [cited by applicant]
M. Tucker, et al., “Preference-based learning for exoskeleton gait optimization,” in [cited by applicant]
K. Li, et al., “Roial: Region of interest active learning for characterizing exoskeleton gait preference landscapes,” [cited by applicant]
M. Tucker, et al., “Human preference-based learning for high-dimensional optimization of exoskeleton walking gaits,” in [cited by applicant]
T. R. Clites, M. K. Shepherd, K. A. Ingraham, L. Wontorcik, and E. J. Rouse, “Understanding patient preference in prosthetic ankle stiffness,” [cited by applicant]
M. K. Shepherd, A. F. Azocar, M. J. Major, and E. J. Rouse, “Amputee perception of prosthetic ankle stiffness during locomotion,” [cited by applicant]
B. Hu, E. Rouse, and L. Hargrove, “Fusion of bilateral lower-limb neuromechanical signals improves prediction of locomotor activities,” [cited by applicant]
F. Pedregosa, et al., “Scikit-learn: Machine learning in Python,” [cited by applicant]
A. J. Smola and B. Schölkopf, “A tutorial on support vector regression,” [cited by applicant]
D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” [cited by applicant]
M. Abadi, et al., “ [cited by applicant]
F. Chollet et al. (2015) [cited by applicant]
S. Hochreiter and J. Schmidhuber, “Long short-term memory,” [cited by applicant]
A. F. Azocar, L. M. Mooney, J.-F. Duval, A. M. Simon, L. J. Hargrove, and E. J. Rouse, “Design and clinical implementation of an open-source bionic leg,” [cited by applicant]
L. Gabert, S. Hood, M. Tran, M. Cempini, and T. Lenzi, “A compact, lightweight robotic ankle-foot prosthesis: Featuring a powered polycentric design,” [cited by applicant]
A. J. Young and D. P. Ferris, “State of the art and future directions for lower limb robotic exoskeletons,” [cited by applicant]
Boonstra et al., “Walking speed of normal subjects and amputees: aspects of validity of gait analysis,” Prosthetics and Orthotics International, vol. 17, No. 2, pp. 78-82, 1993. [cited by applicant]