End-to-End Deep Reinforcement Learning for Autonomous Racing

Autonomous racing provides a fast-paced and demanding platform for testing and advancing autonomous vehicle technologies. Unlike conventional road driv- ing, the competitive environment of motorsport demands rapid adaptability, high-speed manoeuvring, and operation at the physical limits of vehicle dynam-

ics. End-to-end (E2E) neural networks (NNs) trained with reinforcement learn- ing (RL) have emerged as a promising approach for addressing these challenges, enabling autonomous systems to learn optimal racing strategies through trial and error in simulated or real-world environments [1]. However, these systems

still have significant issues in three critical areas: transferability between simula- tion and real-world environments, computational efficiency, and interpretability of decision-making processes. This project aims to address these fundamen- tal limitations.