Yu, Ruilin (Jilin University), Wang, Cheng (Heriot-Watt University), Lv, Zhouhang (Jilin University), Zhang, Yuxin (Jilin University)

Requirements Decomposition for Perception Systems of Autonomous Vehicles: A Case Study of Multi-Object Tracking

Scheduled for presentation during the Regular Session "Advanced Vehicle Safety Systems III" (FrBT8), Friday, September 27, 2024, 13:30−13:50, Salon 16

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 December 26, 2024

Abstract

Safety requirements decomposition is critical to ensure a safe autonomous vehicle (AV) by design despite the importance of safety verification and validation. This study proposes a method called QUASARS (QUAntifying SAfety Requirements using Shapley) for efficiently decomposing AV perception safety requirements into component-level and effectively quantifying them. QUASARS models the quantification of the impact of component-level faults on system-level faults as a feature importance calculation problem. We demonstrated QUASARS using a multi-object tracking system as an example and validated component-level safety requirements 100 times on the test set. After meeting the generated component-level safety requirements, the testing system was able to meet the system-level safety requirements, indicating the effectiveness of this method in decomposing system-level safety requirements into component-level.