Reinforcement Learning in Autonomous Racing

Abstract-Reinforcement Learning (RL) is rapidly advancing as a key approach to training autonomous agents, particularly in complex, real-time environments such as autonomous racing. This review discusses the latest developments in RL applied to endurance and competitive racing, including telemetry data integration and the application of advanced deep reinforcement learning models. The paper explores the architecture and strategies behind “Formula RL,” a system designed to optimize vehicle performance on the racetrack through RL. We delve into how RL algorithms such as Deep Deterministic Policy Gradient (DDPG) and Proximal Policy Optimization (PPO) are employed to enhance racing strategies, vehicle control, and decision-making, ultimately setting a course for the future of autonomous racing.