Driverless cars have long been promoted as a way to reduce road collisions, but measuring whether they are actually safer than people is surprisingly difficult. .

New research from the Insurance Institute for Highway Safety provides some of the strongest evidence yet: Waymo’s fully driverless vehicles were involved in substantially fewer crashes per mile than human drivers operating in the same locations and years

Waymo Recorded Fewer Crashes

Researchers analysed Waymo’s Level 4 automated vehicles operating in San Francisco, Los Angeles and Austin. Level 4 means the vehicle can perform the entire driving task without human supervision, but only within specified operating conditions.

Across approximately 50 million miles of driverless operation, Waymo vehicles had a police-reportable crash involvement rate 68% lower than that of human drivers. The human comparison covered roughly 222 billion miles driven during corresponding periods and in the same areas.

The difference was especially large for certain types of collisions. Waymo vehicles were involved in 81% fewer injury crashes and 85% fewer single-vehicle crashes per mile. Their rate of striking another vehicle from behind was 91% lower than for human drivers, while their rate of being rear-ended was 40% lower.

Results Varied By City

The overall advantage did not appear equally everywhere.

Compared with human drivers, Waymo’s police-reportable crash rate was 76% lower, 35% lower in San Francisco and 71% lower in Los Angeles.

Austin was the exception, where the rate was 4% higher. However, researchers stressed that the Austin sample was relatively small, making that result much less certain than findings from cities where Waymo had accumulated more mileage.

The results therefore should not be interpreted as proof that every driverless system is safer than every human driver. They apply specifically to the Waymo Level 4 vehicles, locations and operating periods included in the analysis.

Comparing Crashes Was Complicated

One of the biggest challenges was that automated vehicles and human drivers operate under different reporting rules.

People often report only crashes involving injury or substantial property damage. Researchers estimate that around half of all crashes and approximately one-third of injury crashes involving human drivers may never be reported to police.

Automated-driving companies, by contrast, have been required to report a much broader range of incidents to federal regulators. Until reporting rules changed in 2025, even relatively minor events could qualify.

To make the comparison fairer, researchers reviewed federal reports from 2021 to 2024 and removed duplicate records, incidents that occurred away from public roads, cases in which automation was not engaged and events that did not constitute genuine crashes.

Of 736 crashes involving automated systems on public roads, only 22% were judged likely to have been reported to police if a human driver had been involved. That included 89 Waymo crashes, of which 64 occurred during fully driverless operation.

Better Safety Data Is Needed

The study also exposed a significant problem with monitoring autonomous vehicles.

Companies must submit crash reports, but they generally are not required to disclose how many miles their automated vehicles travel or whether those miles were completed with or without human supervision. Waymo voluntarily provides mileage information, allowing researchers to calculate meaningful crash rates. Comparable figures are not consistently available for other operators.

Lead researcher Eric Teoh said the findings were encouraging but emphasised the need for a better national data system as driverless fleets expand. Without reliable mileage and crash information, determining whether safety improvements continue at larger scale will remain difficult.

For now, the results offer evidence that one mature Level 4 system can outperform human drivers on several important crash measures. They do not establish that autonomous driving is universally safer, but they show that carefully developed driverless technology may already be reducing some of the risks created by fatigue, distraction and ordinary human error.