Abstract:

Takeover failures in Level 3 (L3) automated driving constitute a critical bottleneck for roadway safety, and the challenge is particularly acute in pedestrian-conflict scenarios. This study examined two key factors, i.e., scenario criticality and non-driving-related tasks (NDRTs) by designing three pedestrian-conflict takeover scenarios using a high-fidelity driving simulator. (Scenario 1 involved a pedestrian crossing at an unsignalized intersection, Scenario 2 involved a pedestrian crossing on a straight road segment, and Scenario 3 involved a pedestrian suddenly appearing from a blind spot on a straight road segment.) A total of 42 participants were recruited to complete the experiment, during which drivers’ driving behavior and physiological responses were synchronously recorded. An evaluation indicator system was established across five dimensions: physiological sensitivity, response timeliness, control stability, ride comfort, and evasive safety. Statistical analyses were conducted to characterize takeover behavior under varying levels of scenario criticality and NDRT engagement, and an integrated takeover-performance assessment method combining physiological and behavioral measures was proposed. Results show that both higher scenario criticality and NDRT engagement significantly degrade takeover performance, with a significant interaction effect between the two factors. The Takeover Performance Index (TPI) exhibited a gradient decline across the three scenarios, decreasing from 0.601 in Scenario 1 to 0.523 in Scenario 3, with 39.2% of participants classified as "poor" performance in Scenario 3. NDRT engagement reduced overall TPI by approximately 11.4%, with the largest decrement observed in the blind-spot scenario. Among TPI components, the evasive safety indicator (minimum TTC) carried the highest weight (0.227), whereas physiological sensitivity indicators had relatively lower weights. The largest performance decrement occurred when participants performed an NDRT in the occluded (blind-spot) pedestrian sudden-appearance scenario. For the TPI, behavioral indicators dominated the composite evaluation. Although physiological measures sensitively captured changes in stress-related arousal and cognitive workload, their discriminative power in the final performance grading was comparatively limited. Thus, they are better positioned as complementary information for interpreting driver state fluctuations and underlying risk mechanisms. Collectively, these findings provide methodological and theoretical support for takeover strategy design and safety-risk assessment in conditionally automated driving.

Key words: non-driving-related tasks (NDRTs), driving behavior, physiological characteristics, takeover performance, improved CRITIC method, comprehensive evaluation