High-resolution urban fatal crash rate benchmarks for automated driving system assessment

Objective
This study established high-resolution, localized fatal crash rate benchmarks across the 50 most populous U.S. urban areas to facilitate future safety performance assessment of automated driving systems (ADS). This study asks the research question: what localized fatal crash trends emerge when analyzing typology, road classification, and temporal variables that can inform more precise, unbiased ADS safety performance assessments?

Methods
Benchmarks were created using several established fatal crash rate metrics to examine trends in these top 50 urban areas. Fatal crash involvement rates (FCIR), which quantified passenger vehicle fleet involvements in fatal crashes per vehicle miles traveled (VMT), served as the primary analytical metric. This measure accounts for the fatal injury status of all participants, encompassing both other vehicle occupants and vulnerable road users (VRUs). The analysis integrated 2023 Fatality Analysis Reporting System (FARS) crash data with multi-source exposure data, including Federal Highway Administration (FHWA) Highway Statistics and INRIX GPS probe data. Fatal crash rates were stratified by road type (freeways vs. surface streets), temporal groupings (day vs. night; weekday vs. weekend), and a crash typology.

Results
The median fatal crash rates across the top 50 urban areas were generally lower than the national average, indicating that lower population and rural areas tend to have a higher fatal crash rate. The analysis of trends revealed an 7.5-times difference in FCIR between the highest-risk (Memphis, TN) and lowest-risk (Boston, MA) urban areas. On average, surface streets exhibited FCIR 2.3 times higher than freeways. Temporal effects were pronounced, with weekend nighttime driving presenting a risk level 6.2 times greater than weekday daytime driving. Pedestrian, V2V intersection, and single vehicle collisions were the most prominent fatal crash types on surface streets, while single-vehicle, V2V front-to-rear, and secondary crashes dominated freeway involvements.

Conclusions
The substantial geographic, road type, and temporal variability in fatal crash rates observed across the 50 studied urban areas underscores the necessity of localized, stratified benchmarks to avoid analytical bias in ADS safety performance assessments. By proactively defining these high-resolution benchmarks, this study provides a transparent roadmap for the incremental evaluation of ADS as statistical power accumulates, ensuring conformance with safety assessment best practices such as the RAVE Checklist.