Researchers developed a framework, ADV3D, to test the safety of self-driving vehicles (SDVs). The industry typically uses closed-loop simulation to test how SDVs interact in synthetic and real scenarios. However, these tests usually only evaluate the motion planning module and consider only behavior variations. The proposed ADV3D framework takes real-world scenarios and performs closed-loop sensor simulation to evaluate the complete autonomy system. The framework modifies the vehicle shape in a realistic manner to degrade autonomy performance, thereby identifying potential failures and uncomfortable SDV maneuvers. The shape variations found with ADV3D are more effective than those in open-loop, highlighting the importance of finding scene appearance variations that affect autonomy in the interactive setting.
Publication date: 2 Nov 2023
Project Page: https://waabi.ai/adv3d/
Paper: https://arxiv.org/pdf/2311.01446
