Waymo has published two pieces of road-safety research, and the interesting part is the benchmark rather than the conclusion. National crash databases record blood alcohol levels but not how many miles sober and impaired drivers each cover, so Waymo built an exposure reconstruction model, cross-referencing US fatality records against telematics to estimate the split across the 50 most populous urban areas.
Taking alcohol-involved drivers out lowers the average urban fatal crash involvement rate by 23%, from 1.42 to 1.10 per 100 million miles, ranging from 31% in Houston to 13% in Minneapolis. Fully sober drivers still face 3.1 to 3.9 times the fatal crash risk at night as in daytime. A second study, on 37,654 fatal cases from the 2023 US fatality database, found more than 77% involved impairment, speeding or an unworn seat belt or helmet, 9.2% involved all three, and only 8.2% came down to behaviour alone.
Why this matters for anyone reading autonomous driving numbers: the sober baseline is a harder target than the human average robotaxi comparisons usually use, so Waymo is proposing to be judged against it. The first study is accepted for publication rather than published, the mileage split is reconstructed rather than observed, and the claim that autonomous systems could address up to 83% of contributing failures is Waymo’s own estimate.
Source: Waymo, 8 October 2026.