Investigating Human's Takeover Capability for L3 Autonomous Driving: Driver Behavior and Physiological Sequences and Their Impact on Risk Levels

The deployment of Level 3 autonomous driving necessitates confirmation of the driver's takeover capability. Current NDRT-based analysis is limited and may yield inaccurate predictions due to comprehensive situation. We propose a research framework based on takeover performance and risk-level grading, implemented through data collection from 36 drivers on simulator including eye movements, ecg, gsr, and vehicle data. By employing dimensionality reduction and cluster analysis, we examined the behavioral and physiological characteristics before (non-driving behavior and condition, NDBC) and during (takeover behavior and condition, To) the takeover across different level clusters revealing significant differences. Positive and negative eye-gsr-ecg-risk sequences indicate potential for predicting takeover behavior and risk. This study may advance precise takeover capability assessment and risk prediction models.