Trajectory prediction method using deep learning for intelligent and connected vehicles

Abstract

The trajectory prediction is significant for the driving safety of intelligent and connected vehicles. To accurately predict the vehicle trajectory, a hybrid method combining physic-based and data-based methods is proposed for intelligent and connected vehicles. The proposed method applied the physic-based method to represent vehicle kinematics. Then, the error of the physic-based method, which is the unmodeled features, is modeled with the data-based deep learning method using Encoder-Decoder Long short-term memory (LSTM). The proposed method is trained and evaluated by an actual vehicle dataset. When the prediction horizon is 3s, compared with the physic-based method, the longitudinal error, lateral error, and yaw angle error decreased by 93.9%, 86.6%, and 76.0%, respectively. Results show that the proposed method improves the trajectory prediction accuracy of autonomous and connected vehicles.