ITSC 2025 Paper Abstract

Paper FR-LM-T35.1

Lian, Zhexi (Tongji University), Yan, Xuerun (Tongji University), Bi, Ruiang (Tongji University), Wang, Haoran (Tongji University), Hu, Jia (Tongji University)

PSP: Physical-Informed Sparse Learning for Interaction-Aware Vehicle Trajectory Prediction

Scheduled for presentation during the Regular Session "S35a-Optimization, Control, and Learning for Efficient and Resilient ITS" (FR-LM-T35), Friday, November 21, 2025, 10:30−10:50, Surfers Paradise 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Keywords Integration of Autonomous Vehicles with Public and Private Transport Networks, AI, Machine Learning for Real-time Traffic Flow Prediction and Management, Autonomous Vehicle Safety and Performance Testing

Abstract

Predicting future trajectories of surrounding vehicles has long been regarded as a crucial task in autonomous driving. However, trajectory prediction remains a significant challenge due to i) the black-box nature of pure neural networks, which overlooks physical dynamics; ii) the difficulty of explicitly modeling complex inter-vehicle interactions; and iii) the trade-off between learning efficiency and model performance. To this end, we propose PSP, a Physical-informed Sparse learning approach for interaction-aware trajectory Prediction. The main contributions of our work are as follows: i) a novel physical-informed modeling paradigm to enhance physical plausibility of predictions; ii) an explicit modeling of inter-vehicle interactions to enhance prediction explainability; and iii) a sparse learning framework to improve learning efficiency. The evaluation on NGSIM dataset demonstrate that PSP achieves the superior performance across all RMSE metrics from 0.26m @1s to 1.81m @5s. In addition, PSP enables interaction-aware predictions while providing explicit representations of inter-vehicle influences over time. Ablation studies further validate the effectiveness of the proposed sparse learning design. The code repository is available at: https://github.com/zhexilian/PSP-Physical-informed-Sparse-learning-for-interaction-aware-vehicle-trajectory-Prediction