If It Cannot Be Seen, Can It Be Quantified? Explicit and Implicit Risks in AV–HV Mixed Traffic

Abstract

Mixed traffic of autonomous vehicles (AVs) and human-driven vehicles (HVs) poses complex safety challenges due to heterogeneous driving patterns. However, current risk assessment approaches predominantly address explicit risks, with limited attention to implicit risks hidden in driving dynamics. This study introduces a dual-perspective risk evaluation framework that jointly considers explicit and implicit risk in AV–HV interactions. Explicit risk represents the directly measurable collision threat from instantaneous motion states. It’s quantified by the reciprocal of extended time-to-collision (ETTC), integrating relative speed, distance, and motion vectors, then categorized into multi-level risk sets in the speed–distance domain. Implicit risk is reflected by underlying individual behavioral complexity and group driving heterogeneity that may precede risk conditions. It’s assessed via sample entropy to capture maneuver complexity, and group-based trajectory modeling (GBTM) to identify hidden heterogeneity in traffic flow. Applied to interweaving areas of highways and urban expressways, the framework reveals explicit and implicit risks during lane changes and car-following, and classifies three representative AV–HV trajectory clusters. The framework offers a dynamic, interpretable, and multi-scale depiction of mixed traffic risk, enabling proactive reliability enhancement and safety assurance for intelligent transportation systems.