Today, the California Department of Vehicles (DMV)’s 2019 Autonomous Vehicle Disengagement Reports was released. The annual report is often used to rank self-driving companies, but it has sparked as much controversy in recent years as it has received.
Since 2015, companies conducting autonomous driving tests on California public roads have been required to submit road test reports to the DMV every year. The Autonomous Vehicle Disengagement Reports include the company’s road-tested mileage and the number of manual takeovers over the past year. “Disengagement” refers to a situation where a human safety officer has to take over the automatic driving system due to a malfunction of the automatic driving system or for driving safety reasons.
In its latest 2019 annual report, the DMV said there are currently 64 companies that can test self-driving cars on California public roads “with safety officers” and that Waymo can test “full” vehicles “without safety officers.” Unmanned” autonomous driving. From December 1, 2018 to November 30, 2019, 36 companies completed 2.88 million miles of self-driving tests on public roads in California, an increase of 800,000 miles from the previous year.
Among them, Waymo tested 1.45 million miles on the road in California, with 0.076 manual takeovers per 1,000 miles, down from 0.09 in 2018; Cruise tested 831,000 miles, with 0.082 manual takeovers per 1,000 miles, down from 0.19 in 2018 Significant decline; Nuro road test 68,760 miles, 0.49 takeovers per 1,000 miles; Aurora road test about 13,400 miles, 10.6 manual takeovers per 1,000 miles, similar to the 2018 data.
In the Chinese company, Baidu has a road test of 108,300 miles, with 0.055 manual takeovers per 1,000 miles, which is lower than Waymo; AutoX has a road test of 32,000 miles with 0.094 takeovers per 1,000 miles; Pony.ai has a road test of 174,800 miles. , 0.154 times every 1000 miles; Didi road test 11,700 miles, 0.682 times every 1000 miles; 5917 miles Weride road test, 6.591 times every 1000 miles; .
It is worth noting that this report is completed by each company’s own reporting of data and statistics. Even DMV is difficult to verify whether each company’s data is “adding oil and vinegar” or “shortcomings”. After all, each company decides whether to artificially It is difficult to unify the standards of takeovers. Some companies may take overs more frequently due to safety considerations and conservative driving, while some companies may have aggressive styles and may take overs less frequently.
Moreover, the road sections tested by each company are very different. Running on a wide freeway or a sparsely populated suburb is likely to be much smoother than running in the busy city center of San Francisco or Los Angeles, and the number of manual takeovers may be less. In addition, it is not excluded that some companies may choose to test complex road conditions in other states and simple road conditions in California in order to make the report look decent.
In fact, self-driving companies are eager to take over by humans to some extent, because it means that there is a small probability of corner cases that machines cannot solve. These small-probability events are once-in-a-lifetime for some leading companies, and they are a good opportunity for them to iterate their algorithms and further enhance the stability and safety of the autonomous driving system.
At present, many companies such as Waymo and Cruise have said that the DMV report is not enough to reflect the maturity of autonomous driving technology, and it is not enough to compare the level of each company. But it is undeniable that even with a large number of flaws, this report is still one of the most objective and public quantitative evaluation standards for the autonomous driving industry in the United States.
Waymo said on its official Twitter today that the DMV’s report is not enough to judge the competitiveness of self-driving companies (Source: Waymo’s Twitter screenshot)
While questioning the value of this report, self-driving companies do need some way to present their technological progress to the public, such as inviting the media and ordinary citizens to take part in test rides. 36氪 have tried Robo-Taxi and self-driving trucks of many domestic self-driving companies in the past year, each time ranging from 10-20 minutes. I have never experienced a driving accident, but I did experience braking too fast. , start too slow, hesitant to turn, unreasonable path planning, long waiting time and other problems.
At the end of the day, reports and test rides are just the tip of the iceberg and part of the truth, and the part hidden under water is the whole picture of the industry.
(Cover image via pexels.com)
I am Wang Yijin, a reporter from 36氪. You can add WeChat catherineyijin for business communication. Please note your name + company + position + purpose.