Human-Centric Analysis of Critical Scenarios and Operational Efficiency during High-level Autonomous Driving

The development of autonomous driving systems (ADS) has significantly advanced in detecting, predicting, and reacting to driving scenarios. Yet, accidents due to system failures prompt the necessity of a "safety guard", even in high-level autonomous vehicles (AVs). As AVs evolve, the role of human riders is fundamentally shifting, with debates on the need for human supervision and intervention in high-level ADS. This research focuses on ADS safety and human-ADS interactions during high-level autonomous driving, highlighting that human interventions can identify safety gaps in ADS and are vital for ADS training. It contributes by offering a novel approach to generating rare safety-critical scenarios, a tool for analyzing complex human interventions, and methods to enhance ADS resilience and automation with a human-centric focus.