Comparison of Waymo Rider-Only crash rates by crash type to human benchmarks at 56.7 million miles
Level 3 - non-randomized controlled study
Retrospective observational comparison to matched population benchmarks (by design analogy, non-clinical)
OpenAlex W4410430116 · doi:10.1080/15389588.2025.2499887
What was done
Researchers conducted a retrospective safety evaluation comparing crash rates of Waymo's commercial Rider-Only (RO) autonomous driving system against human driver benchmarks across 56.7 million miles through January 2025. Human benchmarks were derived from state vehicle miles traveled (VMT) and police-reported crash databases, matched by vehicle type, road type, and operating locations. Waymo crash records were obtained from NHTSA Standing General Order reporting, and mileage was retrieved from Waymo public disclosures. Outcomes evaluated included Any-Injury-Reported, Airbag Deployment, and Suspected Serious Injury+ crashes across 11 crash typologies.
What was found
Waymo RO vehicles had statistically significant lower crash rates than human benchmarks across Any-Injury-Reported, Airbag Deployment, and Suspected Serious Injury+ categories. Vehicle-to-vehicle intersection crashes showed the largest drop, with a 96% reduction in Any-Injury-Reported crashes (87–99% CI) and a 91% reduction in Airbag Deployment events (76–98% CI). Reductions were also statistically significant for cyclist, motorcycle, pedestrian, secondary, and single-vehicle crashes for the Any-Injury-Reported metric. None of the 11 crash type groups showed a statistically significant increase in crash rate compared to human benchmarks.
Why it matters
This represents one of the largest disaggregated evaluations of commercial fully driverless vehicles against location-matched human driving data, assessing both overall rates and specific serious-injury crash mechanisms.
Limits
Waymo mileage data relied on company-reported figures. Human benchmarks depend on police-reported crash records and estimated VMT, which are subject to known human underreporting biases. The analysis is limited to the specific operational design domains, geographic regions, and weather conditions where Waymo currently operates, limiting generalizability to other operating environments.
Cited by
- partial Self-driving cars are currently about 10 times safer than human drivers.