Fewer companies tested autonomous vehicles on California’s public roads last year versus the one before — and yet they logged nearly twice as many miles driven, according to data released by the California Department of Motor Vehicles.
On Wednesday, the California DMV released its yearly collection of autonomous vehicle testing and disengagements data, which details the total number of driverless (including safety driver-supervised) miles driven per licensed operator (typically an autonomous driving startup or company) and the number of times each had to disengage their autonomous driving systems during the course of those drives (referred to technically as “disengagements”, as well as some context around why.
The perennial release of this data has become controversial in the industry, as some companies use it to prove advancements, while others dispute its value at validating technical progress or readiness for commercialization. Disengagements are at the center of the controversy. The problem is that companies have varying definitions of what qualifies as a disengagement, and, to further complicate things, that definition can change over time.
Still, the annual DMV report does provide some important information, including that in California from December 1, 2020 to November 30, 2021, power consolidated among a shrinking group of autonomous vehicle developers that have the capital and the desire to expand public testing and add more vehicles to their fleets.
Gone are the long lists of unknown startups licensed to actively test on public roads. In 2021, the list of active testers shrank, and was dominated by companies based in Silicon Valley and China.
Fewer companies, more miles
During the reported period, test vehicles with human safety drivers behind the wheel traveled about 4,091,500 million miles on California’s public roads, an increase of more than two million miles from the previous reporting cycle. Some 25,000 additional miles were logged by so-called “driverless” vehicles, a term that means the human safety driver is no longer sitting in the driver’s seat.
In 2020, 58 companies held drivered testing permits, which allow operators to test AVs on public roads with a safety driver behind the wheel. In 2021, that number decreased to 50 companies, and of those, only 22 actually tested on public roads during the reporting period.
Leading the charge with the highest number of miles driven was Waymo, which increased its drivered permit miles from 628,838 miles in 2020 to around 2.3 million miles in 2021. Cruise followed with 882,471 miles driven with a human safety operator, up from 770,049 miles the previous year. The company also racked up 6,365 miles without a human safety operator in the vehicle in 2021. No other companies got close to the miles driven by Waymo and Cruise. The third spot was taken by Pony.AI, with 305,617 miles, followed by Zoox with 155,125.
Waymo had a total number of 292 disengagements, which means it had one about every 7,800 miles. Cruise had a total of 21 disengagements while testing with a driver, so a total of 42,022 miles per disengagement. While Cruise improved significantly from the previous year, when it had 27 disengagements, which happened on average every 28,520 miles, Waymo seems to have gotten worse. In 2020, Waymo reported 21 disengagements, which would put them at a disengagement every 30,000 miles or so.
TechCrunch and many others in the industry no longer use disengagements to pick winners or leaders on the path to commercialization: There is no standard about what qualifies as a disengagement; as a result, companies will interpret and report their disengagements differently. Those interpretations can change over time, making it even more difficult to actually understand how a company has progressed in its technology.
While these reports do contain juicy data sets that help us learn more about the AV companies playing the field, we’re not buying claims of winning the safety or the tech battles from these reports alone. The most valuable information to come out of disengagement reports details how active companies are in public road testing in California or how much their fleets have grown or shrunk.
And even then it doesn’t tell the whole story, since numerous companies also test autonomous vehicles outside of California.
A few takeaways
Still, the report did provide a few nuggets.
For example, Waymo’s registered fleet was around 240 vehicles in 2020. The following year, it jumped to nearly 700. However, a good chunk of those registered autonomous vehicles were inactive for either all of the reported period or most of it.
It’s worth noting that the DMV doesn’t require companies to report on testing done on private tracks and closed courses. Nor do companies need to report testing done out of state or miles driven while the vehicle is collecting data in manual mode or at an autonomy level lower than Level 4.
Indeed, Waymo told TechCrunch some of its registered cars could have been at different locations during the reported time period, like outside the state of California, particularly when Waymo was updating its SF fleet with Jaguar I-Paces and relocating the Chrysler Pacificas to other states. They could have also been at closed-course testing facilities or even being driven around in manual mode for mapping purposes.
Cruise’s fleet was 137 in 2021, and pretty much all of them were actively used for testing. Interestingly, Cruise’s fleet stayed about the same between 2020 and 2021.
The data suggests that the AV scene in California is consolidating in a big way around the major players we’ve all been hearing about, like Cruise and Waymo. But, again, the numbers aren’t the only thing to tell us which company is commercializing the fastest.
Let’s take a look at Nuro. The company only tested 59,100 miles last year, but it’s also already offering a commercial delivery service rather than chasing the dream of a California Public Utilities Commission permit to operate robotaxis for a price.
Sometimes, California’s AV reports are loudest in the data that isn’t there. Imagry was notably missing from the data set because it failed to file a required report by January 1, so its permit was suspended effective February 7. In addition, Leonis Technologies’s permit expired on December 31, 2021, and the company doesn’t seem to have done anything about it.
This article has been updated to include more information from Waymo.