Driving with Regulation: Trustworthy and Interpretable Decision-Making for Autonomous Driving with Retrieval-Augmented Reasoning

TL;DR

DriveReg tackles the challenge of region-aware, trustworthy autonomous driving by grounding action decisions in retrieved traffic regulations via a two-agent system (TRR and TRA). It combines a retrieval-augmented generation mechanism with LLM-based reasoning to check action compliance and safety against legal rules, offering interpretability through rule-referenced explanations. The DriveReg Scenarios Dataset enables evaluation across multiple cities with hypothesized and real-world scenarios. Experimental results show high compliance and safety, strong regional generalization, and practical potential for real-world deployment with interpretable reasoning.

Abstract

Understanding and adhering to traffic regulations is essential for autonomous vehicles to ensure safety and trustworthiness. However, traffic regulations are complex, context-dependent, and differ between regions, posing a major challenge to conventional rule-based decision-making approaches. We present an interpretable, regulation-aware decision-making framework, DriveReg, which enables autonomous vehicles to understand and adhere to region-specific traffic laws and safety guidelines. The framework integrates a Retrieval-Augmented Generation (RAG)-based Traffic Regulation Retrieval Agent, which retrieves relevant rules from regulatory documents based on the current situation, and a Large Language Model (LLM)-powered Reasoning Agent that evaluates actions for legal compliance and safety. Our design emphasizes interpretability to enhance transparency and trustworthiness. To support systematic evaluation, we introduce the DriveReg Scenarios Dataset, a comprehensive dataset of driving scenarios across Boston, Singapore, and Los Angeles, with both hypothesized text-based cases and real-world driving data, constructed and annotated to evaluate models' capacity for regulation understanding and reasoning. We validate our framework on the DriveReg Scenarios Dataset and real-world deployment, demonstrating strong performance and robustness across diverse environments.

Figures (10)

• Figure 1: Overview of Driving with Regulation (DriveReg) Framework. The framework consists of two main components: the Traffic Rules Retrieval Agent and the Reasoning Agent. The Traffic Rules Retrieval Agent retrieves relevant rules from traffic regulation documents based on the generated traffic rule retrieval query. The Reasoning Agent then identifies the applicable rules from the retrieved set and performs compliance and safety checks based on those applicable rules.
• Figure 2: Illustration of the proposed Traffic Regulation Retrieval (TRR) Agent. The retrieval results are obtained through the similarity score between scene description and well-curated regulation documents with a pre-defined relevance metric.
• Figure 3: Pipeline of processing the selected scenario. The correct action is labeled in green background.
• Figure 4: Example from the DriveReg Scenarios Dataset showing scenario descriptions, relevant traffic rules, and action-level compliance/safety annotations.
• Figure 5: Inference results from the DriveReg (Real-World) Scenarios Dataset. Actions in green boxes represent the final decision-making outputs, which are both compliant and safe. Results demonstrate that our framework successfully retrieves relevant traffic rules and interprets and assesses actions based on those rules. The correct action is labeled in a green background. (Due to space constraints, some actions and reasoning details are omitted.)