CAREER: Deep Learning Based Control Approaches to Increase the Availability and Affordability of Personalized and Home-Based Rehabilitation

This Faculty Early Career Development (CAREER) award supports research that increases the availability of rehabilitation for people with movement disorders, thereby advancing the national health, promoting the progress of science, and advancing prosperity and welfare. Specifically, this project will increase access to rehabilitation by developing a deep learning-based control framework that reduces the cost of home-based hybrid exoskeletons, which combine functional electrical stimulation with actuated robots.

Traditional telerobotic frameworks consist of a leader system remotely generating and sending a desired trajectory to a follower system and the follower generating its own control commands. In this project, the costs of each follower hybrid exoskeleton will be reduced by having the leader computer (located at a medical facility) generate the control commands for the hybrid exoskeletons (located at each individual's home) based on local state information shared by the exoskeletons, which moves the computational demand from each follower to the single leader.

However, the communication between the computer and exoskeletons will be delayed due to communication limitations, which could destabilize the control system. Another challenge is that the dynamics of a hybrid exoskeleton are inherently uncertain and nonlinear. This project will solve these challenges by enabling the remote control of hybrid exoskeletons based on deep neural networks (DNNs) despite the existence of communication delays and uncertainty in the robot dynamics.

Through education and outreach activities focused on controls and rehabilitation engineering, this project will also increase the interest of K-12 and undergraduate students in science and engineering, particularly those from underrepresented groups.

This research aims to make fundamental contributions to Lyapunov-based delay-compensating control frameworks that guarantee system performance for uncertain telerehabilitation and telerobotic systems, despite the dynamic models being nonlinear, uncertain, and delayed. DNNs can potentially compensate for system uncertainty by adaptively approximating the uncertain system dynamics.

Through the Lyapunov-based stability analysis, adaptive update laws for the DNNs will be developed to improve the DNN learning performance in real-time. Beyond compensating for model uncertainty, the DNN-based control system has an added bonus of personalizing the control system for each individual. Successful completion of this project could transform the rehabilitation industry by significantly increasing the availability and affordability of personalized rehabilitation for millions throughout the nation.

Novel DNN-based control frameworks will be developed for uncertain general telerobotic systems with known and unknown input delays, and for uncertain home-based hybrid exoskeletons with unknown input delays. Transformative classes of DNN-based observers and controllers will be developed to enable uncertain general telerobotic systems with known and unknown input and output delays, and uncertain home-based hybrid exoskeletons with unknown input and output delays.

This project is jointly funded by the Dynamics, Control and Systems Diagnostics (DCSD) program, and the Established Program to Stimulate Competitive Research (EPSCoR).

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.