Shao, Wenbo (Tsinghua University), Xu, Tianli (Tsinghua University), Xu, Jiahui (Beijing Institute of Technology), Li, Jun (Tsinghua University), Wang, Hong (Tsinghua University)

Dynamic Occupancy Set-Aware Planning for Autonomous Vehicles

Scheduled for presentation during the Invited Session "Learning-powered and Knowledge-driven Autonomous Driving II" (ThBT1), Thursday, September 26, 2024, 15:30−15:50, Salon 1

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 8, 2024

Abstract

The development of autonomous vehicle (AV) technologies relies heavily on advanced prediction models to navigate complex and dynamic traffic scenarios. Traditional trajectory prediction methods often face challenges due to the unpredictable and multifaceted interactions among traffic participants, leading to functional inefficiencies and safety risks. This paper introduces Dynamic Occupancy Set-Aware Planning (DOSP), building upon our previously developed Dynamic Occupancy Set (DOS) model. DOSP enhances spatial prediction by defining the minimal enveloping domain for potential future vehicle locations, thereby significantly improving the safety and efficiency of AV navigation. This methodology integrates DOS with trajectory planning to tackle these challenges, and its effectiveness is validated through rigorous comparative and ablation studies using the CommonRoad platform. The results show clear improvements in operational metrics such as success rates and collision rates, significantly outperforming traditional models and providing a robust solution to the inherent unpredictability in AV navigation environments.