Reinforcement Learning Methods for Assistive and Rehabilitation Robotic Systems: A Survey

Publication Date

1-1-2025

Document Type

Article

Publication Title

IEEE Transactions on Systems Man and Cybernetics Systems

Volume

55

Issue

7

DOI

10.1109/TSMC.2025.3555598

First Page

4534

Last Page

4551

Abstract

Advancements in robotic systems aimed at improving mobility for individuals with disabilities have required more sophisticated control and navigation methods. Traditional control approaches often lack the complexity and adaptability needed for the high-dimensional nature of human activities. Consequently, reinforcement learning (RL) has emerged as a dynamic and effective framework for managing robotic actions in complex and unpredictable human environments. This article reviews the integration of RL in robotic systems for enhancing the mobility of individuals with disabilities, addressing the limitations of traditional control methods in complex and unpredictable environments. We critically analyze various RL algorithms, discussing their advantages and challenges in assistive and rehabilitation applications. The study highlights the ongoing development of these algorithms, presenting current research directions, future prospects, and key challenges to achieving higher autonomy in assistive robots. Our findings underscore the potential of RL to improve adaptability and effectiveness in robotic control and navigation, offering insights into advancing these technologies for practical implementations.

Funding Sponsor

Mitacs

Keywords

Assistance, control design, electromyography (EMG), exoskeleton, functional electrical stimulation (FES), gait training, prosthesis, rehabilitation, reinforcement learning (RL), robotics

Department

Mechanical Engineering

Link to Full Text

Share

COinS