Autonomous vehicles must safely navigate a complex and unpredictable world. Rare but hazardous driving scenarios—often referred to as “edge cases”—pose significant challenges for AI-based perception and decision-making systems. This presentation explores advanced algorithmic techniques for detecting, categorizing, and evaluating high-risk scenarios within large-scale driving datasets. We will explore methods such as reinforcement learning, scenario mining, and adversarial testing to systematically uncover and evaluate critical situations. Enhancing our ability to find and address these scenarios is key to improving the overall safety and robustness of autonomous driving systems.

• How AI-based techniques can efficiently mine large datasets to identify rare but high-impact driving scenarios
• Methods for generating and validating realistic hazardous scenarios to strengthen autonomous vehicle performance
• Approaches to ensure that scenario discovery aligns with evolving regulatory and industry safety standards