The beginning of February is the moment when the California DMV publishes its annual Disengagement Report, or more precisely, the associated data. A disengagement report is required from the nearly 50 companies currently licensed to test autonomous cars on public roads in California. This has been in place for a number of years, I have documented 2015, 2016, 2017, 2018, 2019 and 2020.
Each year, licensees must report how many autonomous test and commercial vehicles they operated from Dec. 1 to Nov. 30, how many miles or kilometers they drove, and how many disengagements there were. A disengagement here can be described as the event when the autonomous vehicle either got stuck and handed over control to a safety driver present in the vehicle, or when a safety driver took control himself. The latter is the case if, for example, the vehicle was about to make a mistake or could no longer find its way out of a traffic situation.
One can already see from this definition that it is at the discretion of the licensees what is reported as a disengagement. Often an overzealous safety driver takes control, where it then becomes apparent in the simulation that the vehicle would have handled that situation itself after all. Also, fewer interventions are expected on freeways than in suburbs or in cities like San Francisco, where many different road users congregate. In this respect, the figures reported and published by the companies should be taken with a grain of salt. However, I interpret and present them anyway because California is currently the only region in the world where we have such data publicly available, and they offer a little insight into the progress and state of development of autonomous vehicles.
A total of 26 companies reported data for the full year ( (December 1st, 2020 to November 30th, 2021)), as did two others (Inceptio and Uber), but Uber (UATC) was sold to Aurora and thus no longer appears as separate entity. Another company (Imagry) did not report and as a result lost its license. Some other companies either did not report any activities or did not have to submit a report, therefore 26 companies are represented in the following analysis. Furthermore, 7 manufacturers currently have the license to operate the vehicles even without a safety driver, an eighth company (Pony.AI) had lost this license in the fall of 2021.
Reports with Safety Drivers
First up are the reports of rides with safety drivers in the vehicle. A total of 1,175 autonomous vehicles had been in operation in California during the period. Waymo had the most vehicles in operation with 693, followed by Cruise with 138.
All manufacturers together covered a total of 6,482,960 kilometers (4,051,850 miles). Waymo achieved the most kilometers with 3,721,349 kilometers, followed by Cruise with 1,401,768 kilometers.
The number of disengagements varies among manufacturers, but the calculation using the number of kilometers driven shows an interesting picture. For example, Waymo reported one disengagement per 12,744 kilometers for 2021, which is now four times more common than in 2020, when the company reported 47,911 kilometers per disengagement. This likely has to do with the fact that Waymo has now shifted the focus of driving around its Mountain View headquarters entirely to San Francisco. So all the vehicles in Mountain View and the surrounding area are gone (they report), but at least two depots have been opened in San Francisco. And San Francisco is a much more complex place for autonomous vehicles (and human drivers) than Mountain View.
Another reason for Waymo’s more frequent interventions is also said to be the switch to a new type of vehicle. While Chrysler Pacifica minivans were previously the workhorse of the fleet, it is now the all-electric Jaguar iPace. There seem to have been a bit more challenges than probably expected.
Cruise, a division of GM, in turn, has been developing its vehicles in San Francisco from the beginning. The disengagement rate of once per 66,751 kilometers seems to indicate this. This was down almost 50 percent from the previous year’s 45,632 kilometers. To that end, China-based startup AutoX came out on top with 80,173 kilometers per disengagement. However, the company drove only one 46th of Waymo’s distance and reported only one disengagement at all. A similar story can be seen at Argo.AI. There, too, only one disengagement was reported.
Four companies (Apollo, Cruise, Nuro, and Pony.AI) reported driverless trips. In total, these four companies reeled off 39,972 kilometers during the comparison period.
Pony.AI had an incident that caused the DMV to revoke the company’s license. However, we can see that impressive distances have already been covered by driverless rides.
Again, Tesla does not appear this year, and the reason lies in the definition of what is considered an autonomous test vehicle. Tesla’s current interpretation of the FSD Beta, which is now in use in about 60,000 customer vehicles, is that it is a driver assistance system that does not require reporting to the DMV. California is currently working to close this interpretation gap.
Apple, on the other hand, does not seem to be a competitor to be reckoned with in the near future with the current state of development; the numbers are too far away from the top companies. It is interesting to see that many Chinese companies are in the forefront and are putting significant effort into it. AutoX, for example, now has 1,000 vehicles in operation in China.
We see a densification of the field, increasing mileage per disengagement, but also challenges that rides in new cities and regions and with new vehicles offer to companies. We are particularly pleased to see driverless rides included for the first time. Even though it was “only” 40,000 miles this time, a big jump is expected with next year, as Waymo is about to launch its driverless robotaxi operation in San Francisco, and Cruise will presumably be able to expand driverless operations, previously limited to between 11 p.m. and 5 a.m.
Welche aktuellen Ängste prägen uns? Mit welchen Ängsten waren die Menschen in der Vergangenheit konfrontiert, als es die heutigen Technologien noch nicht gab? Warum mischen wir heute im Wettbewerb der Kulturen um neue Technologien nicht ganz vorne mit? Welche Maßnahmen müssen wir ergreifen, um neue Technologien nicht als etwas Beängstigendes und Feindseliges zu betrachten, sondern als ein Mittel zur Lösung der großen Probleme der Menschheit?
This article was also published in German.