Companies running self-driving cars must report their crashes to federal regulators — from low-speed bumps to fatal collisions. Here is every one Waymo has reported for its driverless cars, mapped and put in context.
Living dataset, updates weekly · How it works
Filled circles are sized by crash count. Hollow circles mark cities where Waymo has announced plans but doesn't yet carry public riders.
One of Waymo's two biggest markets by miles, and its densest. More miles in harder traffic means more reports — that's exposure, not necessarily risk. Per mile, the Bay Area records about -- reported crashes per million miles, against -- in Phoenix.
The clusters trace busy corridors and intersections. Click one and keep zooming — most dots are low-speed events: a tap at a stoplight, a parking-lot bump.
Each red marker is one of them. Click to read the full narrative Waymo filed with federal regulators. An injury on this list doesn't mean the Waymo caused the crash.
of crashes happened while the Waymo was stopped or parked. Most are minor: -- of -- involved a moderate or worse injury.
What happens in these crashesinjury crashes per mile than human drivers, by Waymo's own analysis. An independent study found a smaller but still large gap.
Compared with human driversreported crashes per million miles, before and after June 2025. Much of that drop is paperwork: new federal rules stopped requiring reports of many minor bumps. The injury-crash rate barely moved.
Crash rates over timecrashes involved a death. In each, another road user hit the Waymo or was thrown into it; no Waymo has been found at fault.
The fatal crashesCrash counts and per-mile rates across Waymo's cities
Raw counts mostly track how much Waymo drives in each city. Where Waymo publishes its mileage — -- of its -- cities so far — the cards also show crashes per million miles and how often its cars are in an injury crash compared with human drivers in the same area.
Rates count crashes through --, the period Waymo's published mileage covers. Human comparisons are Waymo's own analysis (see below); the range is a 95% confidence interval.
From the "Roadway Type" field the company fills in on each federal report.
The circumstances behind -- reported crashes
The word "crash" conjures images of high-speed collisions. Most of this dataset looks very different. Here is every crash — one square each, colored by the worst injury reported to anyone involved:
Injury levels are what the reporting company was told or alleged ("Highest Injury Severity Alleged"); they cover everyone involved, not only Waymo passengers. Crashes from supervised testing with a safety driver — -- on record — are excluded from all figures here; see the methodology.
-- crashes on record involved a death. In each of the crashes summarized below, according to the reports Waymo filed and to police accounts, the Waymo was stopped or slowing and was hit as a result of another road user's actions; no Waymo has been found at fault.
Sources: Waymo's reports to NHTSA (report IDs 30270-9724, -11713, -16196); San Francisco, Tempe and Dallas police via CBS SF, azfamily, NBC DFW and Fox 4.
The other serious crashes follow a similar pattern, with exceptions. Of the 28 crashes with a moderate or worse injury on record when we reviewed them in October 2026, the Waymo was stopped or parked in 14. But Waymo's own reports describe a few where the car or its riders played a part: passengers opening doors into a cyclist and a scooter rider, a Waymo that stopped across two lanes in front of a bus that then hit it, and one that rolled onto a rider's foot as the rider stepped out while it was still moving. In ten, someone riding in the Waymo was hurt — in several cases, not wearing a seat belt.
In --% of reported crashes, the Waymo was stopped or parked when the crash happened — and in --% it was going 5 mph or slower.
Based on -- crashes with speed data. Median speed: -- mph. Mean: -- mph.
The most common crash is a rear-end collision — and in --% of those, it was the Waymo that was hit from behind. Crash types describe the physical collision, not who was at fault.
Crash types come from Waymo's quarterly data release and cover -- of -- crashes; the most recent crashes are awaiting classification.
Crashes involving pedestrians, cyclists and motorcyclists — the people most likely to be hurt — account for -- of -- crashes (--%).
Crash counts can't answer that. Crashes per mile can.
Cumulative rider-only miles. Filled points are Waymo's quarterly data releases, added automatically; hollow points are milestones from Waymo announcements and research papers. Hover for each source.
Waymo now drives about -- million driverless miles a quarter, so crash counts rise even if each mile gets safer. Dividing crashes by miles gives a rate — and the rate for all reported crashes nearly halves in mid-2025.
Much of that drop is a paperwork change. On June 16, 2025, new federal rules stopped requiring reports for many minor crashes in which the self-driving car was hit and damage stayed under $1,000. The rate of crashes in which someone was injured — which must still be reported — barely moved.
Crashes from Waymo's quarterly release, divided by the miles Waymo drove that quarter (the difference between consecutive cumulative mileage releases). The dashed line marks the reporting-rule change. Each quarter has only a few dozen injury crashes, so quarter-to-quarter swings in that chart are mostly noise.
Waymo compares its crash rates with human drivers on the same kinds of roads in the same cities, using a method published in a peer-reviewed journal. Across -- driverless miles through --, its cars were in far fewer crashes of every severity:
Bars show the 95% confidence interval; the dot is Waymo's estimate. "Expected for human drivers" is how many crashes human drivers would have had over the same miles. Source: Waymo Safety Impact Data Hub (CSV3, "dynamic" benchmark), read automatically each quarter.
Read with care. This is Waymo's analysis of Waymo. The method has been peer-reviewed (Kusano et al., 2025), but human benchmarks can't perfectly match where, when and how Waymo drives. The rarest outcomes rest on very few crashes: the serious-injury and fatal comparisons are based on -- and -- Waymo crashes, hence the wide ranges.
An independent check points the same way, less dramatically. The Insurance Institute for Highway Safety (July 2026) analyzed about 50 million Waymo miles from 2021–2024 and found 68% fewer police-reportable crashes and 81% fewer injury crashes per mile than human drivers in the same cities — but only 35% fewer in San Francisco, and slightly more in Austin, where its sample was small.
Crashes cluster in the afternoon and evening and on Fridays and Saturdays. Part of that is busier roads; part is simply when Waymo's cars are out carrying riders. Waymo doesn't publish miles by hour, so these patterns show when crashes happen — not when driving is riskier.
Timing of -- crashes with a recorded time
7–10 a.m. and 5–8 p.m.
11 p.m.–5 a.m.
6 a.m.–8 p.m.
Crashes with an exact date only (from Waymo's quarterly release).
Counts rise with Waymo's expansion — more cities and more miles mean more reports; see the per-mile rates above. The dip after June 2025 reflects the reporting-rule change, not fewer crashes. Months after -- are incomplete: companies have until the 15th of the following month to file, and NHTSA publishes with a further lag.
Every reported crash is mapped below. Use the filters to find patterns — or look up what's happened near you. Click a marker to read the report Waymo filed.
Federal reports redact addresses, so locations come from Waymo's quarterly release. -- crashes are placed at the intersection or address, -- on the right street near it, and the rest — mostly the -- newest crashes — near the center of their ZIP code or city. Each marker's popup says which. Download all crashes (CSV)
Searches the narrative Waymo filed for each crash, plus city and other party.
ZIP codes come from Waymo's data release, which stopped including them for crashes after June 2025; newer crashes are matched by map position where it is precise enough.
Crashes during testing with a human safety driver behind the wheel. Excluded from the headline statistics.