Crash reports offer a more detailed way to assess self-driving cars than isolated, alarming incidents. Records from Waymo and Cruise show many low-speed collisions over millions of driverless miles, with other road users often appearing to cause the impact. The data point toward promising safety performance, particularly for Waymo, but they do not settle the comparison with human drivers.
What the reported crashes show
Waymo and Cruise had driven a combined 8 million driverless miles, including more than 4 million miles in San Francisco since the start of 2023. The source’s review of company crash reports counted 102 crashes involving driverless vehicles over roughly 6 million miles. That works out to one reported crash for every 60,000 miles.
Most of the collisions were at low speed and did not pose a serious safety risk. In many cases, another vehicle appeared to be responsible: cars rear-ended a stopped robotaxi, backed into one, or scraped it while squeezing past. The reports therefore provide context that a raw crash count cannot: who or what was involved, and how serious the incident appeared to be.
California requires self-driving companies to report every significant crash. That requirement gives the public a view into incidents that might otherwise be difficult to evaluate. Still, crash records are one part of the safety picture, and the source cautions that definitive comparison requires much more driving data.
Waymo’s record includes strengths and limits
In a report about its first million miles of fully driverless operation, mostly in the suburbs of Phoenix, Waymo described 20 crashes. Many involved another vehicle making contact with the Waymo car, including cars backing out of parking spaces or driveways. There were only two cases in which a Waymo struck another vehicle: one involved a riderless motorcycle after its driver fell, and the other involved a car cutting in front of the Waymo.
Waymo also hit objects such as a construction pylon, a parking lot barrier arm, and a shopping cart at speeds between 8 and 13 miles per hour. The company’s San Francisco driving showed similar challenges with objects and road conditions. Reports included collisions with cardboard debris, a chain connected to a temporary sign pole, and a pothole that punctured a tire, as well as scrapes involving parked vehicles.
The source counted five San Francisco crashes where Waymo was clearly at fault. Other reported incidents included vehicles hitting or scraping stopped Waymos, and a vehicle sideswiping one while changing lanes. Waymo also had two more serious crashes there: a car struck the side of a Waymo during a left turn, and an SUV rear-ended one hard enough for its passenger to report injuries.
A separate crash in May killed a dog. That Waymo had a safety driver behind the wheel, so it was outside the review of driverless vehicles. Waymo said it reviewed the event and concluded that neither its software nor a human driver could have avoided hitting the dog. The source notes that the public did not receive enough information to verify that conclusion.
Promising evidence is not proof
For Waymo’s San Francisco driverless fleet, the source’s tally found other vehicles ran into Waymos 28 times, compared with four incidents in which a Waymo ran into another vehicle. Two of those four involved another vehicle cutting in front of the Waymo. The source also identified three or four serious crashes involving Waymo, none of which appeared to be its fault.
Those figures cover more than 2 million miles of driving. They suggest that Waymo’s vehicles have often avoided causing collisions with other vehicles, while also showing recurring difficulties with stationary or unusual objects. The source describes Cruise’s record as less impressive than Waymo’s, while still seeing reason to think its technology may be on par with, or better than, a human driver.
There is a basic challenge in judging safety: serious crashes are relatively uncommon, so a small sample can leave substantial uncertainty. The source says human beings drive close to 100 million miles between fatal crashes. It argues that hundreds of millions of driverless miles would be needed for 100 percent certainty about whether autonomous vehicles are safer.
Why continued measurement matters
The reports are useful because they let readers look beyond broad claims and examine the circumstances of individual crashes. They show both areas of progress and problems that still need attention, including interactions with irregular objects and the need to scrutinize serious incidents. The picture is mixed, but the evidence for better-than-human performance is beginning to accumulate, especially for Waymo.
Whether to let driverless taxi services continue operating is ultimately a decision for policymakers. The source’s case for allowing the experiment to continue rests on the possibility that, at scale, safer-than-human driving could save lives. More miles and careful review of crash reports would help determine whether that promise is borne out.