Exploring Factors Influencing Crash Occurrence Involving Autonomous Vehicle: Random Parameters Model With Heterogeneity in Means and Variances Approach

Autonomous vehicle (AV) technologies are expected to play a crucial role in reducing traffic crashes occurred by human factors; however, foreseeable coexistence of AVs with human-driven vehicles (HDVs) for an extended transition period raises safety concerns. Understanding factors influencing AV-involved crashes is crucial, especially as human drivers may struggle to comprehend the behavior of AVs during interactions. This study addresses this gap by employing a random parameter probit model with heterogeneity in means and variances. Dataset comprises AV crash records obtained from the California Department of Motor Vehicles from 2018 to the first quarter of 2024. Crashes on roadway segments and intersection are modeled separately. Modeling results reveal that factors such as poor lighting conditions, braking maneuver of AVs, proceeding straight movement of HDVs, involvement of bikes/scooters, residential land-use significantly contribute to AV-involved crash occurrence on segments and at intersections. On segments, first quarter of the year, the retail/entertainment land use, sideswipe collision, dangerous maneuver of HDVs and proceeding straight moment of AVs affect the likelihood of AV-involved crashes. Meanwhile, at intersection, rear-end collision, raining/snowing, unusual road conditions, four-leg intersection, lack of pedestrian island/intersection control significantly increases the probability of AV-involved crashes while angle collision and large skew angle decreases it. The findings highlight the need for more targeted goals to improve AV’s safety, such as enhancing AV sensor perception capabilities, incorporating scenario-based tests by categorization of crash location, and developing mass education initiatives to facilitate the broader acceptance and understanding of AV technologies.