To date, China’s self-driving road test efforts have lagged behind other regions. The California Vehicle Administration (DMV) says 64 companies have been granted licenses for road tests with a human in the passenger seat, with former Google self-driving project Waymo the sole company in the state permitted to test without a human in the vehicle. Waymo has completed 2.34 million km of California road tests, followed by GM Cruise’s 1.33 million km and others such as Pony.ai, Baidu, Nuro and Zoox.
It’s not surprising that California is a world leader in self-driving road testing, considering the large number of AI companies located in the state. But a recent report suggests China has picked up speed, with Beijing emerging as a new self-driving vehicle hot spot.
According to the Beijing Autonomous Vehicle Road Test Report 2019, as of December 31, 2019, Beijing roads were hosting a total of 13 companies engaged in self-driving testing: six tech companies, six OEMs, and one map-maker. With 285 road test licenses granted, testing companies dispatched 77 vehicles. In 2019 Beijing’s cumulative self-driving testing reached 1.04 million kilometres – an equivalent to circling earth 20 times.
Search-engine giant Baidu ranked first in autonomous vehicle testing in 2019. Leveraging the company’s flagship self-driving platform Apollo, Baidu’s 52 vehicles drove 754,000 km on Beijing roads. Pony.ai, a startup that recently raised US$400 million from Toyota, had 7 cars that test drove 120,000 km. Traditional automaker Toyota’s four vehicles logged 11,100 km.
The report observes that most of the actively testing companies opt for a combination of camera, millimeter-wave radar, and LiDAR solutions for their self-driving vehicles. LiDAR is a key component for many autonomous vehicle systems’ sensing and positioning, and the development and deployment of domestic LiDARs brands is on the rise: 40 percent of LiDARs used in the road testing were manufactured in China, an increase of 20 percent from 2018.
There has also been a significant shift in the choice of LiDAR types. Most vehicles are no longer equipped with the industry’s highest channel (laser beam) counts of 128 or above. Instead, the use of 40 channel models has increased, from 50 percent of the vehicles in 2018 to 73 percent in 2019.
Crucial lessons can be learned through studying even minor operational failures during testing, and for vehicles with less than 5,000 km experience this is done in closed track settings. These sites have seen a total of 133,600 km in tests and a total of 16 collisions, or on average one for every 8,350 km driven per vehicle. The accidents include driving off the road, rubbing against roadside rocks, collisions with sidewalks or trees, scraping traffic facilities, failing to avoid a collision with simulated pedestrians, and collisions with other vehicles.
Behind the scenes, 34 software and hardware failures also occurred during the testing (quick repairable failures are not counted), for an average of one failure per 3,929 kilometres per vehicle.
Beijing ranked first among all Chinese cities in terms of the number of applications, licenses issued, and road mileage. To help further fuel the industry and narrow the R&D gap with California, the Beijing government is opening up additional self-driving testing blocks, open test roads and fully open testing areas. A dedicated development zone has been set up to explore next-generation business models such as Robotaxis. The city is also devising new testing grounds for unmanned delivery, cleaning, retail, and security patrols.
Localization: Meghan Han | Editor: Michael Sarazen