LLM-Powered Framework for Interpretable Traffic Rule Processing in Autonomous Driving

Traffic rules are essential for ensuring road safety and efficient transportation, yet their complexity and regional variability pose significant challenges for human drivers and autonomous systems. This paper introduces the Automatic Interpretable Traffic Rules (AITR) framework, which leverages Large Language Models (LLMs) to process, interpret, and structure diverse traffic regulations into machine-readable formats suitable for integrating autonomous driving and driver assistance systems. Unlike traditional perception pipelines, AITR bridges unstructured legal texts with real-time environmental data to enable context-aware, legally compliant decision-making. The framework incorporates explainability techniques to generate human-interpretable justifications for each decision, enhancing transparency, trust, and regulatory alignment. We evaluate AITR using real-world multimodal driving datasets and simulation environments, demonstrating its ability to generalize across jurisdictions, adapt to evolving regulations, and improve decision accuracy in complex traffic scenarios. The results show that our approach enhances rule comprehension, reduces ambiguity, and contributes to safer and more interpretable autonomous navigation.