Analysis of Simulated Autonomous Wheelchair Driving using GA-PID and RL based Controllers

The powered wheelchair is a common means of transportation for people who suffer from limited mobility. For many decades researchers have been trying to facilitate powered wheelchairs with the latest technology. In recent years, autonomous wheelchairs have become of interest, in particular, reinforcement learning has emerged as a promising control algorithm, capable of enabling wheelchairs to navigate complex environments with minimal human intervention. With a view to exploring the compatibility of an RL based controller in an intelligent wheelchair, this paper presents the design and simulation-based implementation of a soft-actor-critic reinforcement learning controller. The objective of this work is to design and demonstrate how an actor-critic based RL controller performs compared to an optimized PID controller. Simulation results indicate that the actor-critic based RL controller achieves lower trajectory deviation and faster travel, while the optimized PID controller demonstrates superior energy efficiency. In addition, both controllers are evaluated with multiple tracks of varying complexity, demonstrate the reliability of the designed controllers.