Skip to content
Book Open access

From Natural-Language Crash Reports to Executable Driving Scenarios via Model Transformation

Oct 2026 · Proceedings of the ACM/IEEE 29th International Conference on Model Driven Engineering Languages and Systems · 0 citations · 13 references

Abstract

Crash reports in natural language capture real traffic failures and can seed autonomous-vehicle simulation, but simulators require structured artifacts such as OpenDRIVE roads and OpenSCENARIO behaviors. Hence, report information should first be formalized as intermediate models that capture the report content and then translate to executable OpenDRIVE/OpenSCENARIO artifacts. In this paper, we propose Crash2OpenX, a model-transformation toolchain assisted by vision-language models (VLMs) and large language models (LLMs). It extracts crash reports into intermediate road and scene seed models, then transforms them into OpenDRIVE/OpenSCENARIO artifacts. From 713 NHTSA/DMV crash reports, 490 fit this fixed vocabulary and cluster into 120 road-and-maneuver crash patterns. We evaluate Crash2OpenX on the 42 most frequent, all of which compile and execute in CARLA, and provide a browser demo from PDF or text to a runnable scenario. The source code and demonstration video are publicly available 1.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.