Do time of day and day of week matter when comparing Waymo’s automated driving system and human crash involvement rates?

Le résumé fourni par la source

OBJECTIVE: While comparative studies have shown that human crash risk is higher at night, especially during the weekend, it has been difficult to quantify crash rates by these temporal dimensions due to a lack of human exposure data, namely vehicle miles traveled (VMT). The objective of this study was to integrate a novel data source for VMT into benchmark data to generate benchmarks by time of day and day of week for safety evaluation of the Waymo Automated Driving System (ADS) driving. METHODS: State crash and VMT data at the county level were combined with granular traffic volume data from Inrix. The study compared crash involvement rates in crashes with outcomes of Serious or Fatal Injury, Airbag Deployment (in any vehicle), and Any Injury. Crash and VMT data was queried for calendar year 2023 (the latest year of complete data available) from counties where Waymo currently operates at sufficient scales to make statistical conclusions: Maricopa, Arizona, San Francisco and Los Angeles, California, and Travis, Texas. Waymo RO crashes were identified from the NHTSA Standing General Order (SGO) database through September 2025, corresponding to 127 million autonomous miles. RESULTS: Benchmark vehicle involvement rates on surface streets were the highest on weekends and night time, peaking between midnight and 3:59 AM. Rates during this time period were between 2.00 and 4.82 times higher on weekdays and 2.67 and 6.23 times higher on weekends compared to the overall average benchmark rates. The ADS fleet had a statistically significant and lower crash involvement rate in all times/days examined for Any Injury and Airbag Deployment outcomes. Several times were also statistically significant and lower for Serious or Fatal Injury (00:00 - 03:59 on both weekdays and weekends and 20:00 - 23:59 weekdays). A benchmark population that drove the same proportion of miles by location and time/day as the ADS fleet would have between 1.41 and 1.71 times higher vehicle involvement rates than the overall population. Compared to the adjusted benchmark, the ADS fleet had a -82% difference ([-77%, -88%] 95% CI) for Any Injury, -86% difference ([-80%, -93%] 95% CI) for Airbag Deployment, and -90% difference ([-78%, -100%] 95% CI) for Serious or Fatal Injury crash involvement rate. CONCLUSIONS: This study addresses an important data alignment challenge of time/day exposure differences when comparing ADS and human crash rates. Compared to human benchmark driving, the ADS fleet drove proportionately more in areas where risk is higher (densely populated areas), and at times when risk is higher (late at night), and therefore has a larger safety impact when accounting for these dynamic factors.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.