Interpretable Decision-Making for Autonomous Vehicles with Retrieval-Augmented Reasoning via LLM

View a PDF file from the paper entitled Driving with Organization: Making Interpretated Decisions of Self -Government Vehicles with logical thinking in the return via LLM, by Tianhui Cai and 5 other authors

PDF HTML (experimental) view

a summary:This work provides an explanatory framework for decision -making for independent vehicles that integrate traffic regulations, rules and safety guidelines in a comprehensive manner and enable smooth adaptation to different areas. While the traditional methods based on the rules are struggling to integrate the full range of traffic rules, we develop a TRR registration agent based on the generation of retrieval activation (RAG) to automatically recover the bases and traffic guidelines of the wide regulation documents and relevant records based on the case of the ego vehicle. Given the semantic complexity of the rules that have been recovered, we also design a unit of thinking supported by a large language model (LLM) to interpret these rules, distinguish compulsory guidelines and safety guidelines, and evaluate procedures for legal compliance and safety. In addition, logic is designed to be interpreted, which enhances both transparency and reliability. The frame shows a strong performance on both supposed and realistic cases through various scenarios, as well as the ability to adapt to different areas easily.