Investigating the contributing factors to autonomous Vehicle-Road user Conflicts: A Data-Driven approach.

Abstract

With the imminent widespread integration of Autonomous Vehicles (AVs) into our traffic ecosystem, understanding the factors that impact their safety is a vital research area. To that end, this study assessed the impact of a wide range of factors on the frequency of AV-road user conflicts. The study utilized the Woven prediction and validation dataset, which contains over 1000 h of data collected from the onboard sensors of 20 AVs in California. Two Copula-based models were developed to investigate the contributing factors to total and severe AV conflicts in road segments (model M1) and intersections (model M2). For road segments, results indicated that road characteristics (direction, number of lanes, road length, speed limit, the presence of a dividing median) and road infrastructure (presence of bus stops, presence of cycle lanes, and presence of on-street parking) have a significant impact on the hourly conflict rates. Regarding the rate of severe conflicts, road user volume, road characteristics (direction, road type, access point density, the presence of a dividing median), and the presence of cycle lanes were identified as the most influential factors. For intersections, the road user volume and the presence of a physical median were found to be positively associated with the hourly conflict rates, while road user volume, intersection characteristics (posted speed limit, lack of traffic control signals, presence of pedestrian crossing, presence of cycle lane, presence of a dividing median, and truck percentage), and the dominant land use at the intersection area were the most impactful variables on the frequency of severe conflicts.