Autonomous vehicles must operate safely in an unpredictable world, yet hazardous driving scenarios – often rare and complex – pose significant challenges for AI-driven perception and decision-making systems. This presentation explores advanced algorithmic approaches to identify, categorize, and test these critical edge cases in large-scale driving datasets. We will discuss techniques such as reinforcement learning, scenario mining, and adversarial testing to systematically uncover and validate hazardous scenarios. By enhancing our ability to detect and address these challenges, we can improve the robustness and safety of autonomous driving systems.

In this presentation, you will learn more about:

• How AI-driven methods can efficiently search vast datasets to identify rare but critical hazardous scenarios
• Techniques for synthesizing and validating realistic high-risk situations to improve AV robustness
• Strategies to align data-driven scenario discovery with regulatory and industry safety standards

• How can AI and ML models be trained to handle unpredictable real-world conditions, including rare and adversarial scenarios?
• What are the limitations of simulation-based training, and how can real-world data be effectively incorporated to improve model robustness?
• With massive sensor data streams, how can V2X communication and edge computing optimize real-time decision-making in autonomous vehicles?
• How can AI models be designed to continuously learn, adapt, and generalize across different geographies, weather conditions, and traffic patterns?
• What are the key AI safety and validation strategies required to meet regulatory standards and ensure reliable deployment at scale?

• What are the key regulatory and certification challenges for AI-driven tools used in safety-critical automotive applications?
• How can we establish trust in these tools when traditional validation and verification methods may not be sufficient?
• What role should industry standards (e.g., ISO 26262, ISO 21448, UNECE WP.29) play in certifying AI-based development tools?
• How can we assess the reliability, robustness, and explainability of AI-driven tools supporting safety-critical development?
• What collaborative efforts between regulators, OEMs, and technology providers are needed to create a viable certification framework?

Presenting the results of our World Cafés: The moderators summarize the key takeaways of their World Café Round Tables on the main stage and share the main findings with all attendees.