LLM Pipeline for Deriving AV Driving Requirements
Fig. 1: Overview of the pipeline to derive driving requirements. Our pipeline takes traffic scenarios as input and produces actionable driving requirements for AVs. During the training phase, a taxonomy with node-wise semantic indexes is learned to empower the LLM with better grounding in the associations between traffic scenarios and traffic laws. During inference, AV Perceptron is mapped to the learned taxonomy to retrieve applicable legal provisions. After that, a reasoning process derives actionable driving requirements from applicable provisions, supporting route choice, tactical planning
Paper context
Paper title: Towards Lawful Autonomous Driving: Deriving Scenario-Aware Driving Requirements from Traffic Laws and Regulations Abstract: Driving in compliance with traffic laws and regulations is a basic requirement for human drivers, yet autonomous vehicles (AVs) can violate these requirements in diverse real-world scenarios. To encode law compliance into AV systems, conventional approaches use formal logic languages to explicitly specify behavioral constraints, but this process is labor-intensive, hard to scale, and costly to maintain. With recent advances in artificial intelligence, it is promising to leverage large language models (LLMs) to derive legal requirements from traffic laws and regulations. However, without explicitly grounding and reasoning in structured traffic scenarios, LLMs often retrieve irrelevant provisions or miss applicable ones, yielding imprecise requirements. To address this, we propose a novel pipeline that grounds LLM reasoning in a traffic scenario taxonomy through node-wise anchors that encode hierarchical semantics. On Chinese traffic laws and OnSite dataset (5,897 scenarios), our method improves law-scenario matching by 29.1\% and increases the accuracy of derived mandatory and prohibitive requirements by 36.9\% and 38.2\%, respectively. We further demonstrate real-world applicability by constructing a law-compliance layer for AV navigation and developing an onboard, real-time compliance monitor for in-field testing, providing a solid foundation for future AV development, deployment, and regulatory oversight. Passages referencing this figure: al-logic specifications [ 28 , 29 , 30 , 31 , 32 , 33 , 34 ] to specify when a rule applies and what constraints the vehicle must satisfy. Although effective in well-scoped settings, such logic-based rule bases depend heavily on expert-defined specifications, making them labor-intensive to construct, costly to maintain as regulations change, and difficult to scale to complex real-world scenarios. Fig. 1: Overview of the pipeline to derive driving requirements. Our pipeline takes traffic scenarios as input and produces actionable driving requirements for AVs. During the training phase, a taxonomy with node-wise semantic indexes is learned to empower the LLM with better grounding in the associations between traffic scenarios and traffic laws. During inference, AV Perceptron is mapped to the leading to mismatched requirements. In this work, we propose a novel pipeline to derive driving requirements, namely behavioral constraints, from traffic legal provisions and regulations. Because driving requirements are not universally applicable but depend on scenario conditions, deriving them from legal text first requires identifying the conditions under which a provision applies. As shown in Fig. 1 , our pipeline consists of two components. (1) To reduce ambiguity and hallucination, we implement explicit law–scenario matching by grounding both laws and scenarios in a traffic scenario taxonomy. Taxonomies are a well-established mechanism for semantic indexing that facilitates grounding and retrieval. Building on OpenDRIVE [ 53 ] and OpenSCENARIO [ 54 ] , we unify their representations
