Reinforcement Learning DDPG Algorithm Based Wheeled Mobility Aid Robot Control Methods

Abstract

Junkai Li, Mohd Rizon Mohamad Juhari and Tiang Sew Sun

When using wheeled walker robots to help individuals with limited walking ability improve their mobility, the stability of the robot's motion control and trajectory tracking accuracy are critical. In this paper, a new trajectory tracking method for wheeled walking robots is proposed by combining the Deep Deterministic Policy Gradient (DDPG) algorithm in reinforcement learning with a Proportional Integral Differential (PID) controller. The article first analyzes the kinematic model of the chassis of the wheeled walking robot, and then introduces the design principle and structure of the adaptive PID controller that combines the DDPG algorithm and the PID controller. Finally, the effectiveness of the research scheme and control strategy is verified by joint simulation experiments, and the results show that this DDPG-based PID controller can automatically adjust the parameters when tracking the trajectory to ensure the accuracy of the trajectory, and it has a strong anti-interference capability.

PDF