Enhancing Autonomous Vehicle Safety Based on Operational Design Domain Definition, Monitoring, and Functional Degradation: A Case Study on Lane Keeping System

Zhengmin Jiang, Wenbo Pan, Jia Liu, Yunjie Han, Zhongming Pan, Huiyun Li, Yi Pan

Abstract: Autonomous vehicles encounter safety challenges in dynamic and unpredictable environments at present. To address this issue, this paper introduces a series of safety mechanisms related to the Operational Design Domain (ODD) for defining, monitoring, and implementing functional degradation strategies of lane-keeping systems. Initially, causal inference and vehicle dynamics stability theory are employed to establish the ODD space. Subsequently, a counterfactual-based lane detection monitor is developed, utilizing structural equations to instantiate causal relationships between lane detection accuracy and perception-related ODD elements. Simultaneously, a lateral control monitoring method is introduced through the fitting of stable boundary parameters. Functional degradation maneuvers are triggered upon any warning from ODD monitoring. To strengthen the quantitative analysis of the safety benefits of the presented mechanisms, a Kriging-based Subset Simulation (KSS) algorithm is proposed. This algorithm requires only 11.96% to 13.02% of the computing resources compared to standard subset simulation technology. Experimental results demonstrate the potential of our ODD definition, monitoring, and functional degradation strategy approach in significantly reducing lane departure rates from ${1}{\rm{.01 \times 1}}{{{0}}^{{\rm{ - 3}}}}$ to ${6}{\rm{.82 \times 1}}{{{0}}^{{\rm{ - 6}}}}$. Overall, our research introduces an innovative, interpretable, and scalable analytical framework for the safety enhancement of autonomous vehicles.

External IDs:dblp:journals/tiv/JiangPLHPLP24