Exploratory analysis of automated vehicle crashes in California: A text analytics & hierarchical Bayesian heterogeneity-based approach

Abstract

Automated vehicles (AVs) represent an opportunity to reduce crash frequency by eliminating driver error, as safety studies reveal human error contributes to the majority of crashes. To provide insights into the contributing factors of AV crashes, this study created a unique database from the California Department of Motor Vehicles 124 manufacturer-reported Traffic Collision Reports and was linked with detailed data on roadway and built-environment attributes. A novel text analysis was first conducted to extract useful information from crash report narratives. Of the crashes that could be geocoded (N = 113), results indicate the most frequent AV crash type was rear-end collisions (61.1%; N = 69) and 13.3% (N = 15) were injury crashes. These noteworthy outcomes and a small sample size motivated us to rigorously analyze rear-end and injury crashes in a Full Bayesian empirical setup. Owing to the potential issue of unobserved heterogeneity, hierarchical-Bayes fixed and random parameter logit models are estimated. Results reveal that when the automated driving system is engaged and remains engaged, the likelihood of an AV-involved rear-end crash is substantially higher compared to a conventionally-driven AV or when the driver disengages the automated driving system prior to a crash. Given the AV-involved crashes, the likelihood of an AV-involved rear-end crash was significantly higher in mixed land-use settings compared to other land-use types, and was significantly lower near public/private schools. Correlations of other roadway attributes and environmental factors with AV-involved rear-end and injury crash propensities are discussed. This study aids in understanding the interactions of AVs and human-driven conventional vehicles in complex urban environments.

Keywords: Automated vehicles; Bayesian binary logit; Crashes; Prior elicitation; Random parameter.