Legal compliance checking of autonomous driving with formalized traffic rule exceptions

Autonomous driving (AD) systems need to obey traffic rules and sometimes execute critical maneuvers that breach existing rules to ensure safe and rule-compliant driving. To endow such legal knowledge to the AD module, we need to formalize rules considering expressiveness, decidability, scalability, and adaptability. This paper critically examines possible formalization methods and demonstrates how we can model traffic rule exceptions for compliance checking of AD models. This ensures that AD systems are safe and can identify situations requiring more complex reasoning, such as exempting ongoing rule processes. We formalize legal traffic rule exceptions hierarchically and modularly in temporal logic and ground them to sensor data for assessing model compliance. Moreover, we introduce a parsed tree structure that supports and aids neural network-based models with formal rules. We evaluate our approach by monitoring vehicle trajectories against formalized traffic rules and handling rule exceptions in various traffic scenarios. Our results show that our approach can effectively represent complex traffic rules and monitor the safety and efficiency of AD systems against legal specifications. This paper contributes to the field of legal reasoning and compliance checking by providing a methodology for formalizing traffic rules from a rule-exception perspective in a machine-readable form based on sensor data limitations.