Vehicle Trajectory Prediction for Autonomous Driving Applications: State-of-the-Art Review, Research Challenges, and Future Directions

Abstract Trajectory prediction technology is considered to be the core part of the decision-making control system of autonomous vehicles. The accuracy of trajectory prediction directly determines the height of autonomous driving performance. Accurate trajectory prediction can reduce the risk of collision between autonomous vehicles and other vehicles, and can provide accurate location-based services. Traffic conditions can also be monitored and predicted in advance by it, and then the best route is recommended to users. In order to fully understand the frontier problems and research progress of vehicle trajectory prediction, the key points of trajectory prediction technology are systematically sorted out and summarized. Firstly, based on the existing research, the vehicle trajectory prediction methods are divided into three categories: traditional physical representation and probability distribution prediction, deep learning prediction, and reinforcement learning prediction. Secondly, the main challenges faced by the three types of prediction methods, such as agent interaction modeling, motion behavior intention prediction, trajectory diversity prediction, multi-modal data fusion, are described, and the key issues and main research progress are summarized and analyzed. Finally, the significant influence of trajectory prediction technology on improving the intelligent level of autonomous vehicles is summarized, and its future development trend and technical challenges are prospected.