Prediction of Cut-In Trajectory for Autonomous Vehicles Based on Intention Query Attention Network with Physical Constraint

Lane change of vehicle is considered to be one of the main factors causing traffic accidents. In the context of autonomous driving, the early prediction of cut-in movement of surrounding vehicles for autonomous vehicles is a key element to ensure a high-level of road safety. The accuracy of long-term trajectory prediction has always been a focus. Since driving intention is an important clue for vehicle trajectory prediction, a method of vehicle trajectory prediction based on intention query attention network with physical constraint is proposed this paper by using deep learning. The historical trajectory features of the target vehicle and surrounding vehicles are effectively extracted by the model based on the intention query mechanism, and the interaction between vehicles is captured by a multi-headed attention mechanism. In the end, the predicted trajectory of the target vehicle is generated by the decoder. Moreover, physical constraint losses are designed to constrain the longitudinal acceleration, lateral acceleration, longitudinal velocity, and the change in heading angle difference of the trajectory to generate trajectories that match the real world. The proposed model is trained on the NGSIM and highD dataset and compared with the state-of-the-art algorithms. In addition, Natural driving data matching Chinese road conditions are processed and ablation experiments are conducted based on this dataset to evaluate the effectiveness and robustness of our proposed approach. The experimental results show that the model proposed in this paper greatly improves the prediction effect of cut-in vehicle trajectory, which has important research significance for enhancing the safety of autonomous vehicles.