Waymo Just Dumped A Huge Amount Of Its Autonomous Driving Data

Illustration for article titled Waymo Just Dumped A Huge Amount Of Its Autonomous Driving Data
Photo: Getty Images (Getty Images)

Waymo just released a huge amount of data covering the 6.1 million miles its cars have driven since January 2019. The data includes details about the 47 collisions Waymo’s cars have been involved in during that period, which is an uncommon amount of transparency in an industry sorely lacking in it.


To be clear, the collisions were a mix of actual collisions and simulated ones, though Waymo said it wouldn’t expect serious injuries — and there were no serious injuries — in any of them. Eighteen of the collisions happened for real, and the other 27 were “simulated,” which just means that they would’ve happened, in Waymo’s estimation, if not for the intervention of the safety driver.

Let’s look at a few of them. In this first example, it was the Waymo cars being hit, not Waymo doing the hitting:

Two actual collisions involved the Waymo vehicle being struck on the rear bumper while traveling straight at a constant speed at or below the speed limit. In one collision, the Waymo Driver had slowed to a constant speed in the course of traveling over a speed bump. In the other collision (Figure 4, Event C), the Waymo vehicle, traveling straight at the speed limit, was struck by a vehicle traveling 23 mph over the posted speed limit. Both collisions were of S1 severity, with airbag deployment occurring in the striking vehicle in the latter collision within this grouping.

This second one is possible road rage, which, side note, becomes even weirder when it’s directed at an autonomous car.

The single simulated event (row 17 in Table 1) in this grouping involved a vehicle that swerved into the lane in front of the Waymo and braked hard immediately after cutting in despite lack of any obstruction ahead (consistent with antagonistic motive). The Waymo Driver was simulated to have achieved full braking in response to the other vehicle’s braking, but was simulated to contact the lead vehicle with a relative impact speed of 1 mph (S0 severity).

Finally, the worst one, which was simulated:

The simulated collision in Figure 9 (Event H) depicts a vehicle making a left turn across the Waymo vehicle’s travel path. The other vehicle did not have the right-of-way at any point leading up to the depicted sequence of events. The Waymo Driver’s simulated response to the vehicle’s action was the initiation of braking just prior to entering the intersection. Simulated full braking was achieved, resulting in a 12 mph speed reduction before simulated impact. Based on the vehicle masses and simulated vehicle speeds and geometry at impact, this event is categorized as S1 with expected airbag deployment. It is the most severe collision (simulated or actual) in the dataset and approaches the boundary between S1 and S2 classification.


The graphic for that one:

Illustration for article titled Waymo Just Dumped A Huge Amount Of Its Autonomous Driving Data
Graphic: Waymo

The level of transparency in the reports is a big part of the reason experts say Waymo is ahead in the race to go autonomous, despite what Tesla might have you think. That’s because Waymo is in fact inviting scrutiny, compared with Tesla, which sells a product called Full Self Driving that is not at all full self-driving.

You can read the full Waymo reports here, here, and here.

News Editor at Jalopnik. 2008 Honda Fit Sport.


Good on them for the transparency. It’s very interesting seeing how these incidents happen and what measures could have (or did) prevent them.

The most interesting part here is that it pretty steadily proves that the self-driving car is a better driver than the average American. The car probably reacted better to minimize damage than a real person would have, and someone ramming the car from behind or swerving erratically in front of it is just evidence of drivers who should probably have their licenses revoked.