Geng, Jiaheng (Tongji University), Du, Jiatong (Tongji University), Li, Ye (Tongji University), Li, Shangwen (Tongji University), Huang, Yanjun (Tongji University)

Towards Safer End-To-End Urban Driving: A World Model-Based Approach with Enhanced Safety

Scheduled for presentation during the Regular Session "S11b-Motion Planning, Trajectory Optimization, and Control for Autonomous Vehicles" (WE-EA-T11), Wednesday, November 19, 2025, 13:50−14:10, Broadbeach 1&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 19, 2025

Abstract

End-to-end autonomous driving has gained significant attention due to its ability to jointly optimize perception, decision-making, planning, and control. However, its black-box nature and lack of safety mechanisms pose challenges, especially in complex urban environments. To address this, we propose a monocular vision-based end-to-end autonomous driving framework enhanced by the world model. The world model allows the system to predict future states by modeling time-series data, thereby enhancing its forecasting accuracy and minimizing potential risks. Moreover, enhanced safety is integrated by utilizing predicted trajectories and intermediate features, which contribute to more reliable control. Our approach is validated in CARLA under diverse urban scenarios, demonstrating superior performance in perception, planning, and enhanced safety, significantly improving the reliability of end-to-end autonomous driving in urban environments. The results indicate that the proposed safety enhancement strategy leads to substantial gains in driving performance and collision reduction, further verifying the practicality and robustness of the framework in complex urban scenarios.