Safety Effectiveness of Autonomous Vehicles and Connected Autonomous Vehicles in Reducing Pedestrian Crashes

This research aims to study the safety effectiveness of autonomous vehicles (AVs) and connected autonomous vehicles (CAVs) in reducing pedestrian crashes in various scenarios. The proposed methodology involves (1) identifying factors that contribute to pedestrian crashes, (2) developing crash-frequency models to predict the pedestrian crash and identifying the model that performs the best, (3) identifying the AV and CAV technologies that can minimize and remove those identified factors, and (4) assessing the effectiveness of AV and CAV technologies in reducing pedestrian crashes for various road classifications. Using crash data obtained from San Francisco Transportation Injury Mapping System (TIMS) for 2016 to 2020, a two-level Bayesian Poisson lognormal (TLBPL) model is developed to assess the effectiveness of AVs and CAVs in reducing pedestrian crashes. The outcomes of the TLBPL model suggest that weather, lighting, and road classifications tend to influence more vehicle–pedestrian crashes in all road classifications. The results of TLBPL indicate that driver faults related to prediction ability contribute more to pedestrian crashes for all road classifications, while driver fault related to sensing (perception) on urban arterials is the factor contributing most to pedestrian crashes. This paper provides a framework for researchers and engineers to evaluate AVs’ and CAVs’ safety effectiveness by considering crash contributing factors and road classifications.