CaltechAUTHORS
  A Caltech Library Service

Recurrent Neural Network Control of a Hybrid Dynamical Transfemoral Prosthesis with EdgeDRNN Accelerator

Gao, Chang and Gehlhar, Rachel and Ames, Aaron D. and Liu, Shih-Chii and Delbrück, Tobi (2020) Recurrent Neural Network Control of a Hybrid Dynamical Transfemoral Prosthesis with EdgeDRNN Accelerator. In: 2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE , pp. 5460-5466. ISBN 9781728173955. https://resolver.caltech.edu/CaltechAUTHORS:20200928-130620642

[img] PDF - Accepted Version
See Usage Policy.

2902Kb

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20200928-130620642

Abstract

Lower leg prostheses could improve the life quality of amputees by increasing comfort and reducing energy to locomote, but currently control methods are limited in modulating behaviors based upon the human’s experience. This paper describes the first steps toward learning complex controllers for dynamical robotic assistive devices. We provide the first example of behavioral cloning to control a powered transfemoral prostheses using a Gated Recurrent Unit (GRU) based recurrent neural network (RNN) running on a custom hardware accelerator that exploits temporal sparsity. The RNN is trained on data collected from the original prosthesis controller. The RNN inference is realized by a novel EdgeDRNN accelerator in real-time. Experimental results show that the RNN can replace the nominal PD controller to realize endto-end control of the AMPRO3 prosthetic leg walking on flat ground and unforeseen slopes with comparable tracking accuracy. EdgeDRNN computes the RNN about 240 times faster than real time, opening the possibility of running larger networks for more complex tasks in the future. Implementing an RNN on this real-time dynamical system with impacts sets the ground work to incorporate other learned elements of the human-prosthesis system into prosthesis control.


Item Type:Book Section
Related URLs:
URLURL TypeDescription
https://doi.org/10.1109/icra40945.2020.9196984DOIArticle
https://arxiv.org/abs/2002.03197arXivDiscussion Paper
ORCID:
AuthorORCID
Gehlhar, Rachel0000-0002-4838-8839
Ames, Aaron D.0000-0003-0848-3177
Delbrück, Tobi0000-0001-5479-1141
Additional Information:© 2020 IEEE. This work was supported by the Samsung Global Research Neuromorphic Processor Project, the National Science Foundation Graduate Research Fellowship under Grant No. DGE1745301 and NSF NRI Grant No. 1724464 and Swiss National Center of Competence in Research Robotics (NCCR Robotics). This research was approved by California Institute of Technology Institutional Review Board with protocol no. 16-0693 for human subject testing. The authors also gratefully acknowledge discussions with Andrew Taylor and Prof. Yisong Yue, and the opportunity to build the first prototype at the July 2019 Telluride Neuromorphic Engineering Workshop.
Funders:
Funding AgencyGrant Number
Samsung Global ResearchUNSPECIFIED
NSF Graduate Research FellowshipDGE-1745301
NSFIIS-1724464
Swiss National Center of Competence in Research (NCCR)UNSPECIFIED
Record Number:CaltechAUTHORS:20200928-130620642
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20200928-130620642
Official Citation:C. Gao, R. Gehlhar, A. D. Ames, S. -C. Liu and T. Delbruck, "Recurrent Neural Network Control of a Hybrid Dynamical Transfemoral Prosthesis with EdgeDRNN Accelerator," 2020 IEEE International Conference on Robotics and Automation (ICRA), Paris, France, 2020, pp. 5460-5466, doi: 10.1109/ICRA40945.2020.9196984
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:105586
Collection:CaltechAUTHORS
Deposited By: Tony Diaz
Deposited On:28 Sep 2020 21:12
Last Modified:28 Sep 2020 21:12

Repository Staff Only: item control page