360-Degree Sensor Integration to Optimize Reinforcement Learning for Autonomous Vehicles

The prevalent use of front-view and top-view cam-eras in reinforcement learning (RL)-based autonomous vehicle agents introduces limitations in observing the surroundings. To address these challenges, we propose the integration of 360-degree sensors on the agent, ensuring comprehensive observation cover-age. Our approach involves strategically placing four cameras on the vehicle's sides, supplemented by Light Detection and Ranging (LiDAR) technology to eliminate blind spots. Deep reinforcement learning (deep RL) is employed to develop a driving policy controlling the agent's velocity and steering based on the sensor information. Two sensor configurations are explored, namely the common top-view camera and our 360-degree sensor integration. Simulation results demonstrate excellent performance for the agent equipped with our proposed sensor integration, particu-larly in complex traffic scenarios. The evaluation matrix reveals improved route completion rates, reduced collisions, and lower average collision impulse, underscoring the effectiveness of our proposed sensor configuration.