Evaluating Gaze Behavior of Drivers in a Mixed Autonomous Driving Traffic Flow Based on Spatial-Temporal Analysis and Causal Reasoning

Autonomous vehicles (AVs) will coexist with human-driven vehicles (HVs) for a long transition period. During this period, mixed traffic flow will pose a potential threat to traffic safety by affecting human drivers’ hazard perception and decisions when interacting with different styles of AVs. This study operationalized a human-in-the-loop driving simulation experiment to investigate the effect of different styles of AVs on drivers’ perception in an overtaking situation. A trajectory planning algorithm using the risk field as the cost function was established. The results extracted from gaze behavior indicated that the drivers’ attention at AV is significantly higher for an aggressive AV than others. Furthermore, through a causal reasoning assessment based on ensemble learning and SHapley Additive exPlanations (SHAP), overtaking duration during the second lane change process was found to be the most important factor affecting driver’s gaze behavior when overtaken by AVs, which provides a direction to improve the overtaking algorithm.