Waymo Says Its Robotaxis Spared 841 People From Injury

Waymo's new safety report claims its driverless vehicles avoided 841 injury-causing crashes and were 20 times safer than human drivers over 270 million miles.

Waymo's new report, covering rides through June 2026, states that its driverless vehicles avoided 841 injury-causing crashes that human drivers would likely have caused, spanning roughly 270 million rider-only miles. The company says its vehicles were involved in 95 percent fewer serious-injury-or-worse crashes than human benchmarks — the basis for its '20 times safer' framing — along with 82 percent fewer injury-causing crashes and 82 percent fewer airbag deployments. Road & Track and Yahoo Autos both carried the report's figures, drawing on safety data from Atlanta, Austin, Los Angeles, Phoenix, and San Francisco.

According to the report quoted by Road & Track, Waymo says its driver cut injury-causing crashes involving pedestrians by 93 percent, cyclists by 86 percent, and motorcyclists by 82 percent — the road users with the least protection when something goes wrong. The report frames this as the safety impact being 'most profound where road users are most exposed,' whether someone is in a bike lane beside the vehicle, stepping into a crosswalk at dusk, or filtering through dense traffic on a motorcycle. For an enthusiast audience, those are the numbers that separate marketing from meaningful risk reduction.

The comparison is not perfectly apples-to-apples, and Waymo's own methodology explanation, quoted by Road & Track, acknowledges it. Autonomous operators must report any physical contact causing any property damage, while human crash data typically requires damage serious enough to justify a police report. NHTSA estimates that 60 percent of property-damage crashes and 32 percent of injury crashes never get reported to authorities, meaning human benchmarks may undercount crashes. Waymo compares its rider-only crash rates to human rates over the same distance, using state police crash records and Vehicle Miles Traveled data from the cities where it operates.

Independent research shows the technology still has exploitable blind spots. earth.com reports that University of Florida researchers found a simple pattern of repeating black-and-white stripes can fool the stereo-camera depth estimation used by self-driving cars, drones, and robots, making obstacles appear closer or farther than they really are. In a driving test, a checkerboard-style pattern projected onto the back of a vehicle fooled its perception system. Researchers stress no attacker is needed — chain-link fences and parking-garage grilles can trigger it naturally. The team developed fixes for both traditional algorithms and AI models, to be presented in November at the ACM Conference on Computer and Communications Safety.

The Insurance Institute for Highway Safety, while not endorsing Waymo's headline claim, praised the company's openness. 'By making detailed information about crashes and miles driven publicly accessible, Waymo's transparency will not only support independent research but foster public trust,' IIHS chief research officer David Zuby said, per Road & Track. Waymo, for its part, concedes there is always room for improvement, that formally identifying an 'absence of unreasonable risk' is an ongoing mission, and that no single technology or policy can solve what it calls the traffic safety crisis. That candor, combined with the open data, sets a benchmark rivals are now implicitly measured against.

City-level detail gives the report local texture: Waymo says its vehicles had 8 percent fewer airbag-deployment crashes than human benchmarks in Atlanta and 10 percent fewer injury crashes in San Francisco, with significantly fewer intersection crashes overall. The next phase of scrutiny will come from independent researchers using the publicly accessible crash and mileage data — the very transparency IIHS hopes other automated-driving developers will copy. Meanwhile, findings like the University of Florida's stereo-camera vulnerability show where engineering effort must go next, fixing root causes rather than simply feeding AI models more training data. Both the safety ledger and the threat model are now on the table at once.