Toward Safe Motion Planning for Autonomous Driving in Highway

Ensuring safety in autonomous driving remains a significant challenge. The safety of autonomous vehicles (AVs) is profoundly influenced by the driving behavior of different drivers, especially in complex and dynamic scenarios. To address these safety risks arising from various driving behavior, this study introduces a low-risk, high-efficiency motion planning approach based on a trajectory prediction model. Firstly, an unsupervised driving style recognition framework using real-world data information is developed to categorize vehicles into three distinct driving style: cautions, normal, and aggressive. Next, a trajectory prediction model incorporating different driving styles is established using the Transformer prediction network to predict the future trajectories of surrounding vehicles (SVs). Subsequently, a motion planning framework that incorporates driving intentions is constructed to generate candidate planning trajectory cluster. Moreover, based on trajectory prediction and driving style information of SVs, a personalized risk assessment model is established, while also comprehensively considering the impact of safety, efficiency and comfort. Finally, the proposed approach is verified through real datasets and comparative experiments. The results show that the proposed framework achieves high-precision trajectory prediction and attains a 99% success rate in closed-loop testing with real datasets, thereby providing safe motion planning for AVs in dynamic environments.