Waymo Crash Rate 68% Lower Than Humans: IIHS

A new IIHS study found Waymo's driverless vehicles crash 68% less often than human drivers across four U.S. cities, though gaps in national crash-reporting data limit broader safety comparisons.

Waymo’s driverless vehicles were involved in 68% fewer crashes per vehicle mile traveled than human drivers across four U.S. cities, according to a new study from the Insurance Institute for Highway Safety (IIHS). The research, titled “Rise of the Machines: Crash Experiences of Highly Automated Vehicles and Human Drivers,” compared Waymo’s Level 4 driverless operations in San Francisco, Phoenix, Los Angeles and Austin against human-driven vehicles in the same locations. IIHS restricted its analysis to crashes serious enough that a person would typically report them to police, a filter researchers say is necessary for a fair comparison. IIHS President David Harkey said the results are encouraging but that the country’s current crash-data infrastructure isn’t built to monitor a large-scale robotaxi expansion.

Highlights

  • Waymo’s driverless crash involvement rate was 68% lower than human drivers overall, ranging from 76% lower in Phoenix to 4% higher in Austin, where the sample size was small.
  • Researchers reviewed 736 public-road crashes involving engaged automation and determined only 22% — 89 involving Waymo, 50 Cruise, 10 Zoox and 10 others — were likely serious enough to be police-reportable.
  • Waymo’s driverless fleet logged about 50 million miles during the study period, compared with roughly 222 billion human-driven miles in the same cities.
  • Single-vehicle crashes were 85% lower and injury crashes 81% lower for Waymo’s driverless vehicles than for human drivers, on a per-mile basis.

Crash Rates Vary Widely by City

The 68% overall reduction masked significant city-by-city variation. Per vehicle mile traveled, Waymo’s crash involvement rate was:

  • Phoenix: 76% lower than human drivers
  • Los Angeles: 71% lower
  • San Francisco: 35% lower
  • Austin: 4% higher, a result researchers attributed to a relatively small sample size

Of Waymo’s 89 police-reportable crash involvements identified in the study, 64 occurred during driverless operation; the rest involved supervised use of the automation.

Why Comparing Self-Driving and Human Crash Data Is Hard

Human drivers are generally only required to report crashes involving injury or more than $1,000 in property damage, and the threshold varies by state. IIHS researchers noted that roughly half of all crashes and a third of injury crashes go unreported, often because drivers want to avoid an insurance premium increase.

Self-driving companies operate under different rules. Since 2014, California has required permit holders testing autonomous vehicles on public roads to report any crash involving a fatality, injury or property damage of any amount. NHTSA layered on a second requirement in 2021, mandating that automated-driving-system operators report qualifying crashes under a Standing General Order (SGO). Until a 2025 amendment aligned the thresholds more closely, that SGO required companies to log minor incidents — like a vehicle scraping its undercarriage in a parking lot — as reportable crashes, making AV operators more likely than the average driver to report every qualifying incident.

How Researchers Filtered the Crash Reports

To build a realistic comparison, IIHS researchers cleaned up crash reports submitted to NHTSA by Cruise, Waymo, Zoox and other operators from 2021 to 2024. The first pass culled hundreds of duplicate records and eliminated incidents where the automation wasn’t engaged, the vehicle wasn’t on a public road, or no real crash occurred — removing about a quarter of all reported crashes.

Researchers then applied what they called a “common-sense standard” to judge whether a remaining incident was severe enough that a reasonable person would have called police. An airbag deployment was automatically coded as police-reportable, for example, regardless of other damage or injury, given the higher impact speed and replacement cost involved.

Because usable mileage data was only available for Waymo, the company’s crashes formed the basis for the self-driving crash-rate calculation.

A Patchwork System for Tracking Autonomous Vehicle Safety

The SGO was primarily designed to help NHTSA track defects and enforce recalls, but it remains the only national data-collection tool for crashes involving self-driving vehicles — making it a key resource for safety researchers despite that narrower original purpose. Neither the SGO nor California’s database makes it easy to compare Level 4 crash rates against human benchmarks.

Part of the problem is exposure data. Companies must report crashes but aren’t required to disclose how many miles they log or whether a human was driving. Waymo voluntarily discloses its driverless mileage; other operators do not, which IIHS said makes it impossible to calculate how often their vehicles crash relative to distance traveled.

What Automation Level Was Involved?

Waymo’s vehicles are classified as Level 4 under SAE International’s automation scale, meaning the system can handle all driving tasks with no human supervision under defined conditions. That distinguishes it from the technology in more common consumer vehicles:

  • Level 2 (partial automation): Systems such as General Motors’ Super Cruise or Ford’s BlueCruise require a human driver to stay alert and ready to take over at all times. IIHS noted that the monotony of supervising these systems can itself lead to attention fatigue.
  • Level 3 (conditional automation): The vehicle can operate without constant supervision in specified conditions, but a human must remain in the driver’s seat, ready to intervene when requested. Vehicles with Level 3 systems are still rare.
  • Level 4 (high automation): No human supervision is required within the system’s defined operating conditions — the classification that applies to Waymo’s driverless fleet in this study.
  • Level 5 (full automation): Capable of driving without human input under all conditions. No Level 5 vehicles exist yet.

“Those are encouraging signs for the future of driverless vehicles,” said Eric Teoh, IIHS director of statistical services and the study’s lead author. “Now we need to get the data collection system right, so that we can ensure that level of safety continues as these technologies become more prevalent.”

Self Drive News
Self Drive News Staff

Self Drive News is the trade publication of record for vehicle autonomy. Published by Hagman Media and edited by founder Brian Hagman, it covers autonomous vehicles, robotaxis, ADAS, self-driving software and hardware, and L4 commercial deployments for an audience of AV engineers, software safety professionals, and mobility investors.