The Application of Multi-Agent Reinforcement Learning in Path Planning for Automated Parking

With the advancement of artificial intelligence technologies, autonomous vehicles are progressively emerging as a transformative transportation solution. However, decision-making complexities and dynamic interaction challenges in complex environments continue to impede the realization of fully autonomous driving. As an emerging paradigm, Multi-Agent Reinforcement Learning (MARL) offers an innovative perspective to address these challenges. This paper provides a systematic review of recent research advancements in MARL applications for automated parking: beginning with an exposition of the technical foundations of automated parking and MARL theoretical principles; proceeding to a detailed analysis of core algorithmic implementation mechanismsincluding Q-learning and Deep Q-Networks (DQN)and their specific applications in parking path planning; subsequently synthesizing mainstream metric systems for performance evaluation; and concluding with an examination of critical challenges such as learning efficiency and safety limitations, alongside prospective research directions. This review aims to clarify developmental trajectories within the field while highlighting MARL's significant potential for constructing intelligent, high-efficiency automated parking solutions.