Ego Vehicle Trajectory Prediction Based on Time-Feature Encoding and Physics-Intention Decoding

In the stage of man-machine cooperative driving, accurately predicting the trajectory of the ego vehicle can help intelligent system understand future risk and adjust the control authority of the man-machine, thereby improving the performance of the man-machine system and eliminating man-machine conflicts. However, existing high-performance trajectory prediction methods are more focused on fully autonomous vehicles, and it is difficult to deal with the problem of driving trajectory prediction with different risks when the driver is in the loop. So, an ego vehicle trajectory prediction method based on time-feature encoding and physics-intention decoding (TFE-PID) is proposed. Through the bidirectional enhancement of the encoding and decoding process, it can accurately predict the trajectory of the ego vehicle by using only the state data of the vehicle and the driver. In the encoding stage, time and feature information are used for dual encoding, which makes the amount of information carried in the context vector used for decoding more abundant. In the decoding stage, context vector, physical prediction data, and driver’s intention are used to control the flow of information in the network, which enables the model to converge in a direction that is more consistent with the physical characteristics of the vehicle and driver’s intention. The experimental results show that TFE-PID can accurately predict the trajectory of the ego vehicle under different risky driving behaviors of drivers, and has good prediction stability and generalization ability.

I. Introduction

In the stage of man-machine cooperative driving, the driving task is still dominated by humans, and the intelligent system only obtains certain driving authority to correct the vehicle’s status when there is a risk, thereby improving vehicle safety [1], [2]. At this stage, by predicting the trajectory of the ego vehicle, the intelligent system can understand in advance the risk situations that the driver may face under operation. When the risk of the vehicle is high, the system can actively assist some driving operations based on the driver’s expected action space, such as minor braking and steering correction, to improve driving safety and reduce unnecessary man-machine conflicts [3]. Therefore, accurately predicting the trajectory of the ego vehicle is of great significance for improving driving safety and reducing man-machine conflicts.