Multimodal vehicle trajectory prediction at urban uncontrolled intersections considering multiple types of traffic participants and driving risks

The trajectory prediction of the subject vehicle (SV) contributes to the autonomous vehicle’s recognition of potentially risky situations, serving as an essential basis for decision-making and planning. Complex interactions between multiple types of traffic participants at non-signalized intersections increase vehicles’ driving risks, which poses challenges to vehicle trajectory prediction. In this work, a multimodal vehicle trajectory prediction method considering multiple types of traffic participants and driving risks is proposed, which adds the selection of the optimal trajectory and the avoidance of driving risks to multimodal trajectory prediction. Specifically, a Transformer model based on spatio-temporal feature fusion is constructed to extract spatio-temporal interaction features of multiple types of traffic participants and generate candidate multimodal trajectories. An inverse reinforcement learning reward function is designed to evaluate candidate multimodal trajectories and select the optimal trajectory. A risk avoidance module based on the driving risk field is proposed to ensure safe interaction. Experimental results indicate that the model achieves higher trajectory prediction accuracy while fully considering interaction with multiple types of traffic participants. The driving risk measurement results highlight the model’s excellent risk avoidance performance. This work provides an effective new idea for improving the driving safety and efficiency of autonomous vehicles.