Research On Vehicle Trajectory Prediction And Testing Method In The Coexistence Environment Of Human Driving And Autonomous Driving

As a crucial component of intelligent transportation,autonomous driving technology significantly enhances the safety,reliability,and comfort of travel,making it a focal point of research worldwide.For the foreseeable future,human-operated vehicles and autonomous vehicles will coexist within traffic environments.In realistic and complex traffic situations,autonomous driving systems must make more intelligent planning and decision-making to prevent accidents.Consequently,self-driving cars need to promptly and accurately predict the future behaviors or trajectories of traffic participants(e.g.,cars,bicycles,pedestrians),in order to enhance the system’s ability to perceive changes in its surroundings and ensure the safety and reliability of the autonomous driving system.The trajectory prediction model utilizes traffic scene data(historical trajectories of nearby traffic participants,road structures,traffic signals,etc.)as input,extracts data features,and predicts the vehicle’s future trajectory through data reasoning or model recursion.The prediction accuracy of the trajectory prediction model significantly impacts the safety of the autonomous driving system.Predicting the future trajectories of surrounding vehicles is a challenging task due to the complex interplay between various factors.While road structure,traffic signals,and traffic rules provide prior knowledge to constrain a vehicle’s trajectory,the unpredictable behavior of other traffic participants introduces uncertainty.Therefore,predicting future trajectories of surrounding vehicles requires considering both the interaction among traffic participants and the interaction with the surrounding traffic infrastructure.Ensuring the reliability of self-driving cars requires comprehensive testing before their deployment in the real world,and test scenarios form the basis of autonomous driving testing.Traditional road testing involves conducting a large number of tests to identify problems with autonomous driving systems,but this method can be expensive and inefficient.Therefore,autonomous driving tests are increasingly relying on virtual simulation scenarios.Although the cost of generating scenes using expert experience has been reduced,it is still insufficient to meet the requirements of autonomous driving testing.Developing a model that can generate diverse and realistic traffic scenarios is crucial to testing the effectiveness of autonomous driving technology1.Aiming at the problem of predicting the future trajectory of surrounding traffic participants in autonomous driving motion planning,this paper proposes a Transformer-based trajectory prediction model(Trajectory Prediction Transformer,TPT)to help autonomous vehicles predict the future trajectory of surrounding traffic participants.First,in order to effectively consider the interaction information between traffic participants and the traffic environment,traffic participants are modeled as traffic agents.And the historical trajectory of the traffic agent and the surrounding traffic environment information are encoded into a multichannel map,which is used as the input of the model.Then,the improved Transformer is used to model the traffic environment,and the noteworthy interaction information between the traffic agent and the traffic environment is captured to predict its future trajectory.Finally,experiments based on existing traffic scene data sets show that the TPT model can achieve better prediction results than other comparison models at different prediction times,and the time is shorter.2.For the problem of automatic driving traffic scene generation,this paper proposes a traffic scene generation model based on generative confrontation network(Traffic Scenario Generation Generative Adversarial Network,TSG-GAN).The TSG-GAN model uses rich traffic scene data(such as the geometry of lanes,crosswalks,traffic signals,surrounding vehicles,etc.)to quickly generate realistic and diverse traffic scenes.Given the vehicle’s driving intent,the TSG-GAN model can accurately generate unobserved traffic scenarios in reality.Experiments based on public datasets show that the proposed model can accurately generate real and diverse traffic scenarios for testing autonomous driving technology.