Journal of South China University of Technology(Natural Science Edition) ›› 2026, Vol. 54 ›› Issue (3): 52-64.doi: 10.12141/j.issn.1000-565X.250056

• Intelligent Transportation System • Previous Articles     Next Articles

Research on Predictive Motion Planning for Intelligent Connected Vehicles based on Transformer

LI Anran1,2 PAN Yuyan3 XU Zhenlin4 GAO Bolin2 LI Yongxing1 YU Hongsheng5 CHEN Yanyan1   


  1. 1. College of Metropolitan Transportation, Beijing University of Technology, Beijing 100124, China;

    2. School of Vehicle and Mobility, Tsinghua University, Beijing 100084, China;

    3. Department of Civil and Environmental Engineering, Pennsylvania State University, PA 16802, USA;

    4. Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft 2826 CN, Netherlands;

    5. Institute of Electronic Computing Technology, China Railway Science Research Institute Group Co., Ltd., Beijing 100081, China

  • Online:2026-03-25 Published:2025-09-26

Abstract:

This paper proposes a Transformer-based traffic prediction model named ST-Trans and employs ST-Trans to develop a predictive motion planner for intelligent connected vehicles. ST-Trans utilizes Transformer to extract spatial-temporal evolution patterns from real-time vehicle data and lane segment structural information provided by dynamic high-definition maps, predicting the traffic state of lane segments. Meanwhile, ST-Trans further enhances prediction accuracy by incorporating connectivity among lane segments and phase information at intersections. The prediction result demonstrates that ST-Trans outperforms the best baseline model by 12.2%, 12.1%, and 3.55% in mean absolute error, root mean square error, and accuracy, respectively. Based on the prediction results of ST-Trans, the predictive motion planner significantly reduces the computational complexity of motion planning tasks so that it swiftly computes the path plan and speed curve among road sections ahead. This study combines SUMO and CARLA to validate the predictive motion planner and the simulation results demonstrate that the ST-Trans-based predictive motion planner is capable of achieving predictive path and speed planning, and it surpasses traditional motion planners in terms of safety, efficiency, comfort, and computational speed.

Key words: dynamic high-definition map, motion planner, traffic prediction model, deep learning, Transformer