This research presents a novel approach for creating realistic, scenario-driven test cases for Autonomous Driving Systems using multimodal accident data.

• Addresses critical limitations in current testing methods by preserving map data and ensuring accurate trajectories
• Automates the generation of concrete, reproducible test scenarios from real accident reports
• Provides a systematic framework for validating ADS safety in high-risk situations
• Demonstrates improved testing efficiency and coverage compared to conventional methods

This engineering breakthrough enables more thorough safety validation of autonomous vehicles before deployment, potentially reducing accidents and accelerating regulatory approval processes.

From Accidents to Insights: Leveraging Multimodal Data for Scenario-Driven ADS Testing