| The powered wheelchair is a common means of transportation for people who suffer from
limited mobility. This research successfully designed and implemented a two-stage control architecture utilizing Soft Actor-Critic (SAC) Reinforcement Learning (RL) to facilitate autonomous
navigation. In the first stage, a high-level path-following controller was developed to enable
wheelchairs to navigate autonomously. Simulation results indicate this controller achieves lower
trajectory deviation and faster travel compared to an optimized PID controller. However, this initial stage was not explicitly optimized to minimize energy consumption. To address this, a second
stage low-level DC motor controller was designed using a four-quadrant DC chopper to regulate
speed while specifically considering dynamic load change scenarios common in the real world.
This four-quadrant topology allows motor operation in both forward and reverse motoring modes,
providing the bidirectional drive essential for indoor autonomous systems. The reward function
for the motor controller was designed to optimize energy consumption by penalizing speed error,
high current magnitude, and current fluctuations. A comparative analysis indicates the enhanced
robustness of the SAC approach, as it consistently achieves smaller speed deviations than conventional model-based regulators under higher load disturbances.
Key Words: intelligent Control, reinforcement learning, soft actor critic, four quadrant DC chopper circuit, pulse width modulated signal, bidirectional motor drive control |