Crash data can be transformed into realistic edge cases that reveal weaknesses in autonomous driving systems. This data-driven approach enables rigorous validation of perception, planning, and control modules, improving safety and robustness under rare, high-risk conditions.

In this presentation, you will learn:

• How crash data informs realistic edge case scenario generation
• Methods to validate AD stacks under rare, high-risk conditions
• Strategies to improve safety and robustness in autonomous driving