This paper provides a comprehensive review of the research on reinforcement learning(RL)based autonomous driving in structured road environments.Firstly,the basic principles of reinforcement learning are introduced.Next,the role of visual and non-visual perception information representation methods in RL tasks is discussed.Then,from the perspective of the role of RL in autonomous driving,the review is divided into two levels:decision-making planning and decision-making control.For each level,specific RL application methods are discussed based on the main research scenarios.Finally,the current research status and the new problems brought by this new technology are summarized.The review shows that the application of RL technology to autonomous driving in structured road environments still requires ongoing research.Applying RL technology to multiple scenarios needs further exploration,and using existing methods to transfer to multi-vehicle and human-vehicle interaction scenarios requires further validation and research.
维普
万方数据