Is the first so important? There is, but for many self-driving startups in the growth stage, these first rankings are not of much value, overall, the annual report of the California Traffic Authority does not deliberately rank, it can be regarded as a California annual performance list, it is not a list, but in the inner volume of the Chinese autonomous driving industry has triggered a round of attention to the public and capital marketing war, which is not worth advocating.
Wen 丨 wisdom driving network Huang Huadan
For self-driving companies, what test data best represents its strength?
Test total miles? Takeover miles? Number of fleets?
On February 9, local time, the California Department of Motor Vehicles (DMV) released its 2021 annual departure report. The report brings together test mileage and detachment mileage data submitted by 26 companies that conducted autonomous driving tests on California's open roads in 2021, most of which are from both China and the United States, showing a trend of sino-AMERICAN competition for autonomous driving.
But in essence, this is a summary data of the California Traffic Management Bureau, but in just two days, many domestic companies and the media rushed to release relevant interpretations. For a time, there were many opinions on the Internet, and all kinds of unfamiliar terms and ranking charts were inconsistent.
Some media have adopted the test total mileage standard, Waymo deserves to rank first in the world, and Xiaoma Zhixing ranks first among domestic companies; and according to the manuscript released by the Chinese company AutoX, AutoX ranks first.
But for the reader, it is inevitable to wonder, who is the first? What is the point of the first? What report is this? Does the Traffic Authority still manage the rankings? What is MPI? What is MPD?
In this issue, We will try to interpret this report.
——01——
What exactly is a DMV breakout report?
The Disengagement Report is a statistical report compiled by the California Traffic Management Authority (DMV) based on test data submitted by 50 companies that have been licensed to test.
The content includes reporting on the mileage of all vehicles in California from December 1 of the previous year to November 30 of that year and the number of exits (including the number of times, scenarios, and reasons) of the company's vehicles on california's public roads.
An exit takeover is a situation in which a safety officer or regulatory system actively or passively takes over a vehicle when the autonomous driving mode fails or safety issues may be encountered. In the case of active takeover, it is more dependent on the subjective judgment of the security officer.
DmV only requires licensed companies to submit test data, aggregate it and publish it, but does not interpret it.
The 2021 annual report showcases data submitted by 26 companies. The remaining companies either did not conduct testing activities or did not reach the level required to submit reports, so there was no relevant data.
If Tesla is not included in the report, DMVs require vehicles that reach autopilot level L3 and above to submit data, while Tesla's FSD system only reaches L2+ level.
In addition, according to foreign media reports, a company called Imagine was cancelled for not submitting data.
DMV is currently the only government agency to disclose such test data.
However, a number of self-driving companies, including Cruise and Waymo, have expressed doubts, arguing that the departure from the report may be misleading to some extent, because the disclosed data is submitted by each company independently, and its accuracy is debatable.
Moreover, for Chinese companies, many companies have moved the test back to China, and the test data obtained in China is not included in the DMV's report.
Similarly, for U.S. companies that conduct tests outside of California, there is a large amount of test data that is not included.
But in the absence of more accurate data from other sources, dmv's report is still a powerful reference for understanding the actual road tests of the world's top autonomous driving companies.
After clarifying the basic background situation, then, what is MPI and what is MPD?
The full name of MPI is Miles Per Intervention, that is, the mileage interval for each takeover, and the full name of MPD is Miles Per Disengagement, that is, the mileage of the travel interval for each departure. The two are actually different expressions of the same concept, disengagement refers to the vehicle leaving autonomous driving mode, and takeover refers to the driver or supervision system involved in taking over the vehicle. Refers to the average mileage between two breakouts.
▲Image source: car stuff.
In this form, Aimotive and Wincher Technology, which have only 2976 miles and 39 kilometers of test miles and only 2 test cars, are in the first place, which is obviously biased.
This value is commonly used as a ranking basis on the Internet.
▲ AutoX Antu released the ranking
Although MPI can reflect the reliability and independence of autonomous driving systems to some extent, the results produced are naturally very different due to different test conditions, such as test vehicles, software platforms, test environments and purposes. Simply put, if the sample size is not at the same level, ranking autonomous driving companies according to MPI values is not completely reliable.
Let's take a look at the specifics of this year's report.
——02——
The 2021 DMV is out of the true face of the report
A total of 1175 vehicles from 26 companies in the 2021 DMV report participated in testing on California's open roads. Among them, Waymo has the most, with a total of 693 vehicles, followed by Cruise, with a total of 138 vehicles.
On test miles, all vehicles submitting data completed a total of 4,091,500 test miles.
Waymo was far ahead in test miles at 2,325,842.9 miles, followed by Cruise at 876,104.72 miles. Pony Chi Heng came in third place with 305,616.73 miles.
In terms of the total number of departures, Apple's data remains high.
However, the number of departures is closely related to the test mileage and the test environment and purpose, and the data can only be used as a reference.
The total test mileage is divided by the number of departures to give the MPI takeover mileage.
In terms of takeover miles, Chinese company AutoX topped the list with a value of 50,108 miles.
Waymo, which ranked first last year, took over only 7,965 miles in 2021, down significantly from 29,945 miles in 2020. Significant increases in fleet size and test mileage may lead to more frequent exposure to complex road conditions, resulting in more frequent disengagements.
In addition, according to foreign media reports, Waymo has also shifted the focus of its testing from Mountain View, where its headquarters is located, to San Francisco, which is obviously more complicated than Mountain View. Waymo has also switched its fleet's model from chrysler Pacifica to an all-electric Jaguar I-PACE, which may be part of the reason for the decline in the number of miles it has taken over.
The 2021 DMV report also reveals where the breakaway from the takeover occurred. Some of these detachments occur on highways, and some are on urban streets, and different road environments have different effects on autonomous driving.
AutoX, while far ahead in taking over miles, tested a total of just 50,108 miles, well below Waymo's 2,325,842.9 miles.
Therefore, we believe that in the case of the variables can not be unified, the data obtained naturally do not have a strict sense of comparability, can only be used as a reference.
——03——
Fully driverless testing
The above data are obtained in the presence of a driver.
In addition to the 50 companies that have obtained test licenses, DMV has also issued driverless test licenses to 8 autonomous driving companies, namely Baidu Apollo, AutoX Antu, Cruise, Nuro, Waymo, Zoox, Wenyuan Zhixing and Xiaoma Zhixing.
Among them, Pony Zhixing was cancelled due to an accident in the test on October 28, 2021. However, tests can still be conducted on open roads in California with drivers present.
The report released by the DMV also discloses the relevant data of several companies that conduct fully unmanned driving tests.
——04——
Chinese companies are doing well
Although the home field of the road test is basically domestic for Chinese companies, the test data in California and American companies are not much comparable. But from the data point of view, the performance of Chinese companies is still eye-catching.
Here, the data of Chinese companies is also listed separately for reference.
In summary, the detachment report released by dmv is more objectively data recording and publication of the autonomous driving test conducted in California in the previous year, providing a reference for the industry.
For the question raised at the beginning of this article, what kind of test data can represent the strength of autonomous driving companies?
In today's already crowded self-driving track, people habitually expect to simply rank the industry through a report and a data, and divide the head and waist.
But we believe that no single dimension of data is enough to draw reliable conclusions, and rankings based solely on one piece of data are not representative of a company's true strength.
Of course, many of China's new driverless companies compete on every standard, which is itself something to be encouraged, but the one-sided interpretation of the data on either side is not worth appreciating.
One of the conclusions we can draw from this data is that with the joint efforts of many top companies and policy levels, I believe that the real landing of autonomous driving is not far away.
And we expect that cities in China that are beginning to allow road testing can also release such a report once a year, so that Chinese consumers can more fully understand that driverless cars are already around them.
【Pay attention to the video number of intelligent driving, pay attention to the frontier of intelligent driving】