82%
Injury-causing crashes vs. human benchmark. 841 fewer injury-causing crashes than the human benchmark over the same distance, per Waymo’s June 2026 update.
Safety evidence
Crash rates, not crash headlines
Autonomous vehicles crash. The useful question is how often, how badly, and compared with what. This page collects published rate comparisons, the federal crash-reporting rule, and the limits of applying that evidence to Rhode Island.
Limits, up front
- Most detailed rate comparisons come from one operator (Waymo) in a handful of U.S. cities. Those cities are not Rhode Island, and they are not a winter-climate sample.
- Waymo’s public safety dashboard and several papers are operator-produced or operator-affiliated. Swiss Re supplied insurance-claims data for one comparison; Waymo is still a party to that analysis.
- NHTSA’s Standing General Order requires AV operators to report crashes, including many low-speed contacts that never appear in police data. Human crash databases miss a large share of minor crashes. Comparing raw SGO crash counts to police-reported human crashes is misleading.
- SGO files do not include a complete public miles-driven denominator for every operator. Counts without miles are not rates.
- There is not yet a comparable, peer-reviewed, miles-normalized safety study for Tesla robotaxi or Zoox at the same scale as the Waymo papers cited here. We do not invent those numbers.
1. Federal crash reports (official)
NHTSA’s Standing General Order on Crash Reporting requires ADS operators to file crashes, including many low-speed contacts. The public CSVs name the reporting entity, vehicle, month, roadway, and what was struck. They are not a rate: NHTSA does not publish a complete miles-driven denominator for every operator in that file.
Waymo and others have argued, and NHTSA’s own human-crash research supports the general point, that police data miss a large share of property-damage and some injury crashes. SGO counts will look “high” next to police counts even when the rate is lower. We therefore do not chart raw SGO crash totals as a safety ranking.
2. Peer-reviewed comparison (Waymo-affiliated authors)
Kusano, Scanlon, Chen, McMurry, Gode, and Victor (2025), Traffic Injury Prevention, 26(sup1). Open preprint: arXiv:2505.01515.
- Statistically significant lower crashed-vehicle rates than matched human benchmarks for any-injury-reported, airbag-deployment, and suspected-serious-injury-or-worse outcomes at the aggregate level.
- Largest crash-type reduction: vehicle-to-vehicle intersection crashes, 96% lower any-injury-reported (87–99% CI) and 91% lower airbag deployment (76–98% CI).
- Cyclist, motorcycle, pedestrian, secondary, and single-vehicle any-injury-reported crashes were also statistically reduced. No crash type showed a statistically significant disbenefit.
- Authors are Waymo-affiliated. Crash events were taken from NHTSA SGO reports; human benchmarks were built from state VMT and police-reported crash data, aligned to vehicle type, road type, and location.
Mileage: 56.7 million rider-only miles through January 2025, in Phoenix, San Francisco, Los Angeles, Austin.
3. Insurance claims using Swiss Re data (operator-involved)
Waymo published a comparison that uses Swiss Re auto-liability claims over 25.3 million fully driverless miles. Swiss Re’s book of human driving is the benchmark (the write-up cites more than 500,000 claims and 200 billion miles of exposure). Over the same distance, the human benchmark implied 78 property-damage and 26 bodily-injury claims; Waymo recorded 9 and 2.
Source: Waymo research page describing the Swiss Re comparison, December 2024. Label: operator-involved analysis.
Versus all human drivers: 88% fewer property-damage claims and 92% fewer bodily-injury claims. Versus 2018–2021 vehicles with driver-assistance systems: 86% and 90%.
4. Operator dashboard (not independent)
Waymo’s public Safety Impact page reports 271.3 million rider-only miles through June 2026 and the percentage reductions below. These are the company’s own matched-human comparisons. They cover more miles than the 2025 paper, and they are not a third-party audit.
95%
Serious injury or worse crashes. Waymo reports 95% fewer serious-injury-or-worse crashes (55 fewer) than the human benchmark.
93%
Pedestrian injury crashes
86%
Cyclist injury crashes
82%
Motorcyclist injury crashes
Source: Waymo Safety Impact dashboard. Phoenix, the Bay Area, and Los Angeles dominate the miles. Atlanta is still small.
What we are not claiming
- That Tesla, Zoox, or any other operator matches Waymo’s published rates. We did not find a peer-reviewed, miles-normalized study at this scale for those fleets.
- That these rates would hold on unplowed Rhode Island roads, rural Route 1, or I-95 in a blizzard.
- That AVs never kill anyone. SGO and news accounts include serious and fatal events. Rare events need huge mileage to measure well.
- That NHTSA has certified robotaxis as “safer than humans” as a class. It has not.