Driving Smarter: Teaching LLMs Traffic Laws to Assist Autonomous Vehicles

This project explores the capability of Large Language Models (LLMs) to understand and apply traffic regulations effectively. By using driving license theoretical questions as test sets, we assess the extent to which LLMs can successfully pass a driving theory exam. Beyond answering the questions, the project integrates official traffic regulation legal texts into the LLMs. This enhancement enables the models not only to provide correct answers but also to cite relevant legal references, explaining the reasoning behind their responses. The ultimate goal is to evaluate how well LLMs can interpret, apply, and justify traffic laws, contributing to their potential use in legal reasoning and autonomous driving systems.

Legal Text Parser

Presented recently at ESSV 2025, a parser designed to extract key information from legal text has demonstrated high effeciency on StVO legal documentation.
This parser relies upon the inherent structure of the legal text, where information is clearly segmented into several levels, such as chapters, paragraphs, sentences, etc, and is highly inter-referenced. Using this information parser creates a knowledge graph, that is highly reliable for reconstructing the original content and can serve as a solid foundation for further research.

Parser workflow:

  1. Pattern (Ψ) matching : Detects relevant text.
  2. Entity Extraction (EX): Extracts legal entities.
  3. Tree Building (TB): Creates a hierarchical tree.
  4. Entity Tagging (TG): Tags tree nodes with their entity types.
  5. Translation (TR) (Optional): Translation from German to English.
  6. KG Construction (KGC): Defines edges based on citations and hierarchies.

Project Team

Prof. Dr. Munir Georges
Prof. Dr. Munir Georges
Phone: +49 841 9348-2331
Room: K305
E-Mail:
Ibrahim Siddig
Phone: +49 841 9348-6475
E-Mail:
Sviatoslav Tugeev
E-Mail: