Research On Map-Based Representation And Imitation Learning For Intelligent Connected Vehicle Trajectory Prediction Method

In 2022,the world’s car ownership has reached 1.446 billion,of which China’s car ownership accounts for 20%.With the increase of car ownership,how to improve traffic flow efficiency and ensure traffic safety has become an urgent problem to be solved.In order to alleviate current traffic congestion and ensure driving safety,intelligent driving has become the focus of people’s attention.Among them,intelligent connected vehicle trajectory prediction is an indispensable part of vehicle intelligent driving.It aims to estimate the vehicle’s driving,so that the vehicle can plan its path according to the driving estimate.The accuracy of the prediction results affects the accuracy of planning the vehicle’s driving route,and improving the accuracy of trajectory prediction plays a promoting role in realizing intelligent driving.To improve the accuracy of intelligent driving trajectory prediction,the main content of this article is as follows:(1)Through investigating the research status of intelligent connected vehicle trajectory prediction at home and abroad,this article analyzes the two key technologies involved in trajectory prediction: map representation method and imitation learning method,and concludes the necessity of studying them simultaneously.The construction principles of grid map and graph-based map for map representation method are explained,and analyzed three imitation learning methods,Behavior Cloning(BC),Inverse Reinforcement Learning(IRL),and Generative Adversarial Imitation Learning(GAIL),and their applications in intelligent connected vehicle trajectory prediction.(2)This paper proposes a graph-based Generative Adversarial Imitation Learning model for intelligent connected vehicle trajectory prediction to improve the accuracy of trajectory prediction.Grid-based maps can provide precise location information of the surrounding environment to assist prediction,so a maximum entropy inverse reinforcement learning intelligent connected vehicle trajectory prediction model based on grid-based maps is designed;As the grid-based map representation method is unable to capture the interaction between different objects accurately,while the graph-based representation method can more accurately and flexibly represent the interaction relationship,we propose a graph-based Generative Adversarial Imitation Learning model for intelligent connected vehicle trajectory prediction.(3)Conduct simulation experiments on the nu Scenes open dataset for autonomous driving collected by Motional.By comparing widely used evaluation metrics,the effectiveness and reliability of the map-based imitation learning intelligent connected vehicle trajectory prediction method were verified.