Readiness Evaluation on Intersections for Automated Vehicle Based on Left-Turn Safety Sight Zones: A Field Test Study

Intersections are high-risk traffic locations, with left turns being especially hazardous due to their complexity. The detection of required left-turn safety sight zones (SSZs) at intersections is critical for automated vehicles (AVs) to safely execute driving tasks. However, current intersections designed for humans may not adequately accommodate AVs’ safe driving, and the performance of AVs in detecting left-turn SSZs remains insufficiently researched. This study proposes a method for evaluating intersection readiness based on driving tasks. Field test Data were collected at six smart signalized intersections in Shanghai, China. Considering traffic safety, the SSZs and critical detection points were identified for AVs to complete the two-phase task. Detection performance of SSZs was modeled using the CatBoost machine learning algorithm, with interpretability enabled by SHapley Additive exPlanations (SHAP) and partial dependence plots (PDPs). Results showed that three traffic environment factors and six intersection design factors strongly correlated with detection performance. Failure probability was higher in traffic environments with lower vehicle speed (<5 m/s), more vehicles, and occlusion of lead vehicles. Conversely, lower failure probability was associated with intersection design factors including smaller left-turn radius (<28 m), narrower widths of major roads (<29 m) and minor roads (<28 m), and the presence of a three-phase signal. Two interaction effects were also discovered. These findings offer guidance for designing intersections that better support AVs, while also deepening the understanding of AV perception, contributing to logical scenario generation, and operational design domain refinement.