Waymo’s driverless cars had a 68% lower crash-involvement rate than human drivers

## 80 million kilometres adds real weight to the comparison

A study published by the US-based Insurance Institute for Highway Safety (IIHS) compared Waymo’s fully driverless vehicles with human-driven cars in Phoenix, San Francisco, Los Angeles and Austin.

Between 2021 and 2024, Waymo vehicles covered approximately 80.5 million kilometres in driverless operation. Human-driven vehicles travelled around 357 billion kilometres in the same areas over the same period.

The researchers limited the comparison to incidents that a reasonable driver would probably report to the police. Per kilometre travelled, Waymo’s driverless cars were involved in 68% fewer such crashes than human-driven vehicles.

The wording matters. The study measured crash involvement, not fault. It would therefore be inaccurate to say that Waymo caused 68% fewer crashes. What the data does show is that the company’s Level 4 system was involved in substantially fewer crashes over a comparable distance.

## The gap widened further for more serious crashes

The most significant results concerned injury crashes and single-vehicle crashes. Waymo vehicles were involved in 81% fewer injury crashes and 85% fewer single-vehicle crashes than human-driven vehicles.

The rate at which Waymo vehicles rear-ended other vehicles was 91% lower, while their rate of being rear-ended was 40% lower. All rates were adjusted for distance travelled.

The findings are consistent with the expected strengths of automated driving, including continuous monitoring of the surroundings, consistent following distances and stable speed control. The study itself does not establish which technical feature was responsible for the difference.

The crash-rate metric also does not capture every instance of sudden braking, traffic obstruction or a situation in which another road user had to intervene to avoid a collision. A lower crash rate does not therefore mean flawless behaviour in traffic.

## Austin stood apart from the other cities

The results varied by city. Waymo’s crash-involvement rate was 76% lower than that of human drivers in Phoenix, 71% lower in Los Angeles and 35% lower in San Francisco.

Austin was the only city studied where Waymo’s rate was higher than that of human drivers, by 4%. IIHS stressed, however, that the Austin sample was relatively small. The result therefore does not support a firm conclusion that Waymo is less safe than a human driver in the city.

The city-by-city variation also highlights a defining limitation of Level 4 automation. A Level 4 system is designed to operate within a defined operational design domain — particular areas, roads and conditions — rather than everywhere. Results from one city cannot automatically be transferred to an area with different infrastructure, driving culture or weather.

## A fair comparison required manual review of hundreds of cases

Crash data for automated vehicles and human drivers cannot be compared directly. Under earlier versions of NHTSA’s Standing General Order, automated-driving developers had to report even very minor incidents, while human drivers often failed to report crashes to the police.

According to IIHS, roughly half of all crashes and one-third of injury crashes in the US go unreported. Until the third amendment to the Standing General Order took effect on 16 June 2025, automated-driving companies also had to report incidents that an ordinary driver might not necessarily regard as a crash, such as a vehicle scraping its undercarriage as it entered a car park.

The researchers reviewed reports submitted to NHTSA’s federal database between 2021 and 2024. They removed duplicate entries, incidents that occurred away from public roads, cases in which automated driving was not engaged and events where no actual collision took place.

After the data had been cleaned, 736 incidents on public roads remained in the analysis, all involving engaged driving automation. The researchers estimated that only 22% of them met, or probably met, the threshold for reporting a crash to the police.

Waymo vehicles accounted for 89 such crash involvements. Of these, 64 occurred in fully driverless operation, while the remainder took place during supervised automated driving.

Manual review made the comparison more reliable, but it also exposed a weakness in the current data system. Manually coding hundreds or thousands of incident descriptions is not scalable as automated fleets and incident counts expand.

## Rivals could not be assessed using the same method

The researchers also examined crash reports involving Cruise, Zoox and other developers. However, they could calculate a comparable crash rate only for Waymo because the other companies did not publish sufficiently detailed mileage data.

That limitation matters when comparing competitors. Safety cannot be assessed from the number of crashes alone. It is also necessary to know how far the vehicles travelled, which driving mode was active and in which cities and conditions the system operated.

Waymo’s advantage therefore lies not only in its lower crash rate. The company has also published enough mileage data for independent researchers to calculate that rate. No equally robust conclusion can currently be drawn about its competitors.

## Level 4 is not the same as driver assistance

Waymo’s vehicles operate at SAE Level 4. Such a system drives the vehicle itself within a defined operational design domain and does not require continuous supervision by a human driver.

That is fundamentally different from the Level 2 systems commonly fitted to production cars. A driver using Level 2 assistance must monitor the road at all times and be ready to take control immediately.

The Waymo study therefore does not establish the overall safety of lane-keeping assistance, adaptive cruise control or other driver-assistance systems. Nor does it show that every Level 4 system drives more safely than a human. Its conclusions apply only to the Waymo vehicles, cities, years and driverless operating mode covered by the research.

## Europe must demonstrate the result under local conditions

The European Union introduced type-approval requirements for the automated driving systems of fully automated vehicles through Implementing Regulation (EU) 2022/1426. In March 2026, Commission Implementing Regulation (EU) 2026/481 amended the framework, including new provisions for automated valet parking and changes to safety-management requirements.

Estonia and 17 other EU member states also signed a Joint Declaration of Intent on 8 June 2026 to establish large-scale cross-border testbeds for autonomous vehicles. The initiative focuses on common approval principles and coordinated permitting procedures, together with practical applications in public transport, freight and logistics.

Waymo’s result gives Europe a valuable benchmark, but not a ready-made answer. Performance in the US cities covered by the study does not prove that the system will work equally well in European winters, cities with different traffic arrangements or cross-border operation.

Any robust European monitoring regime will need to link every crash to precise mileage, the active driving mode, the operational design domain and the severity of the incident. Otherwise, measurable safety risks being replaced by isolated videos, corporate marketing and crash totals stripped of context.

Waymo has not proved that a machine drives better than a human everywhere. It has, however, shown over a substantial distance that a Level 4 driverless vehicle can be involved in markedly fewer crashes than a human driver within a defined operating area.

In the global race, that gives Waymo a stronger advantage than another demonstration or promise about the future. The company now has an independently analysed safety metric based on tens of millions of kilometres driven in real-world traffic.