Once again this year, the California DMV has required a so-called Disengagement Report from all companies licensed to test and operate autonomous vehicles on all public roads in California. It is the only report known in the world that provides an objective, albeit limited, look at the state of development of this technology across several dozen manufacturers.
For the period from December 2019 to November 2020, a total of 29 companies reported test drives, four of which (Argo, Atlas Robotics, Deeprooute.AI, and QCraft) reported for the first time for the full year. Thirty-four companies reported no test drives in California, and one other company (KAIZR) did not report at all. Of these 34 companies, seven did not renew their license (Bosch, Changan, Faraday Future, SAIC, Samsung, Torc Robotics, TuSimple).
The figures include the number of kilometers driven and the so-called disengagements. A disengagement is therefore considered to be this action when a safety driver takes control of the autonomous vehicle. This calculation can then be used to determine how long each manufacturer can drive autonomously across all deployed vehicles. While the definition of disengagement gives manufacturers room for interpretation, it is the best set of numbers available to the public today to provide insight into the state of self-driving technologies from various manufacturers. This data reported to the California DMV does not include data from other regions.
This year, Waymo is again in the lead, after Baidu indicated a surprisingly (and presumably cheated) big jump ahead last year. It is striking that Baidu did not report any figures this year.
Waymo saw a big improvement from last year (see 2019 Disengagement Report) from 13,219 miles per disengagement to 29,944 miles. Cruise also jumped from 12,221 miles per disengagement to 28,520 miles. Apple also improved from 118 to 145 miles.
Here are the exact numbers in tabular form.
|Manufacturer||Miles||Km||Dis-engage-ments||Miles per Dis-engage-ment||Km per Dise-ngage-ment|
Cruise drove the most miles this year; Waymo scaled back its test drives by quite a bit.
For the first time, DMV also shared who initiated the disengagement – the safety driver or the system itself – and where the disengagement occurred – street or highway.
Some of the manufacturers also gave their assessment of the figures:
This article was also published in German.
Now Baidu’s 18,000 miles doesn’t sound like a lie when Cruise claims no disengagements in 200,000 miles. Maybe the CA DMV should verify these numbers, at least the higher numbers. The CA DMV should send people to sit in the car and count the number of disengagements. Do this for a week. Verify that the observed numbers are close to the reported numbers.