Study of reinforcement learning based shared control of walking-aid robot

Abstract

In this paper, we experimentally investigated a new reinforcement learning based robot shared control algorithm for walking-aid robot. To autonomously adapt to different user operation habits and motor ability, robot dynamically adjusted user control weight by proposed algorithm. The weight adjustment is performed online based on user control efficiency, current robot walking state and environment information by reinforcement learning algorithm. The shared control synthetizes the final robot velocity according to the control weight. The effectiveness of proposed reinforcement learning based shared control algorithm is verified by experiments in a specified environment.