Design and Preliminary Validation of an Assisted Driving System for Obstacle Avoidance Based on Reinforcement Learning Applied to Electrified Wheelchairs

Operating a motorized wheelchair poses inherent risks and demands substantial cognitive effort to achieve effective environmental awareness. Consequently, individuals with severe disabilities face heightened risk, leading to diminished social engagement which impacts their overall well-being. Therefore, we have developed a collaborative driving system for obstacle avoidance based on a trained reinforcement learning (RL) algorithm. The system interfaces with the user through a joystick, capturing the desired direction and speed, while a lidar positioned in front of the wheelchair provides information about obstacle distribution. Taking both inputs into account, the system generates a pair of forward and rotational speeds that prioritize obstacle avoidance while closely aligning with the user’s commands. Preliminary validation through simulations involved comparing the RL algorithm with the absence of an assistive system. The results are promising, showcasing that the RL algorithm reduces collisions without imposing constraints on the desired speed. Ongoing research is dedicated to expanding tests and conducting comparisons with traditional obstacle avoidance algorithms.