Automated vehicle collisions in California: Applying Bayesian latent class model. (December 2020)

• Record Type: Journal Article
• Authors: Das, Subasish; Dutta, Anandi; Tsapakis, Ioannis
• Is Part Of: IATSS research. Volume 44:Number 4(2020)
• Page Start: 300
• Page End: 308
• Publication Date: 2020-12
• Subjects: Automated vehicle -- Bayesian model -- Traffic collision; Traffic safety -- Periodicals; Transportation and state -- Periodicals; Verkeersveiligheid; Internationale organisaties; Traffic safety; Transportation and state; Periodicals; 363.1256
• DOI: 10.1016/j.iatssr.2020.03.001
• Languages: English
• ISSNs: 0386-1112

Abstract

Abstract: The emerging technology of automated vehicles (AV) has been rapidly advancing and is accompanied by various positive and negative potentials. The new technology is expected to affect costs mainly by reducing the number of collisions and travel time, as well as improving fuel efficiency and parking benefits. On the other hand, safety outcomes from AV deployment is a critical issue. Ensuring the safety of AVs requires a multi-disciplinary approach that monitors every aspect of these vehicles. The California Department of Motor Vehicles has mandated that AV collision reports be made public in recent years. This study collected the scanned collision reports filed by different manufacturers that are assessing AVs in California (September 2014 to May 2019). The collected data offers critical information on AV collision frequencies and associated contributing factors. This study provides an in-depth exploratory analysis of the critical variables. We demonstrated a variational inference algorithm for Bayesian latent class models. The Bayesian latent class model identified six classes of collision patterns. Classes associated with turning, multi-vehicle collisions, dark lighting conditions with streetlights, and sideswipe and rear-end collisions were also associated with a higher proportion of injury severity levels. The authors anticipate that these results will provide a significant contribution to the area of AV and safety outcomes. Highlights: Used the most comprehensive automated vehicle collision data (2014–2019) in the analysis. Provided in-depth exploratory analysis of the critical variables. Applied Bayesian latent class models to identify collision patterns. Key crash contributors include turning movement, rear-end collision, multi-vehicle collision, and roadways with no lighting.