Over the past two years, China has been setting the stage for a revolution in autonomous driving. Just don’t expect that revolution to come quickly.
In the United States, individual states are taking the lead in deciding how self-driving cars will be rolled out in their communities, and the companies making those cars are determining what “safe” self-driving technology looks like. Meanwhile, China’s regulatory bodies only started allowing local governments to issue road-test licenses to autonomous-car companies last year, and China has only recently begun the arduous process of building compatible highway infrastructures.
China’s approach isn’t the fastest when it comes to putting AVs on road, but that doesn’t necessarily mean the country is falling behind. When Beijing does give the green light for self-driving vehicles on roads everywhere, China is betting the adoption of AV technologies will go more smoothly and scale more rapidly than it would have otherwise.
Safety first
In January 2018, China’s top economic policy planner proposed that half of the new cars on the road be partially or fully autonomous by the end of 2019. Four months later, the proposed policy was finalized (these and many other links in Chinese). While that deadline is drawing near, and the goal nowhere close to being realized, the statement itself has motivated local governments. Since May 2018, when the national government began allowing provincial and city-level authorities to issue licenses for road tests, 18 local governments have allowed such tests to take place.
“[The government is] trying to create an environment that is as conducive as possible to speedy development,” says Mark Natkin, founder of Beijing-based Marbridge Consulting, which tracks China’s technology developments. “But at the same time, they want to be cautious not to let things move too quickly. China’s traffic conditions are more chaotic than those in, say, California, so from a safety perspective, it will take longer for the technology to mature. The last thing the authorities want is accidents involving self-driving cars that result in serious injury or fatality.”
To avoid that kind of mishap, Chinese companies set strict standards for road tests, and crack down on offenders. In a sparsely populated area of the southern city Guangzhou, officials halted a robo-taxi service from self-driving-car company WeRide in November, saying it lacked proper approvals. A little over a year after AV company Pony.ai put AVs on the road in Guangzhou’s Nansha district, they are still only allowed to pick up and drop off people within that district on an invitation-based trial ride. In Beijing, self-driving companies have to complete more than 5,000 kilometers (3,100 miles) of driving on approved roads to even obtain an open road-test certificate.
China’s self-driving car companies also have plenty of direct communication with authorities, as local governments often set up designated organizations to supervise road tests.
By contrast, some 36 US states (equal to provinces in China) have already rolled out pilots for AVs, and new programs are emerging every month. Silicon Valley’s self-driving car companies are proposing their own safety standards, and the timetable for national safety standards remains unclear: Last year, the Trump administration dissolved its AV advisory committee after just one meeting. That laissez-faire attitude may be helping American AV companies now, but could also set them up to clash with national standards when such standards do emerge.
In China, while additional open-road tests are up to regulators, local firms are bullish on the timeline. Search giant Baidu said it’s looking to roll out 100 robo-taxis next year in the central Chinese city of Changsha, and WeRide is also looking to deploy a robo-taxi fleet as soon as 2020.
“It’s not about how fast you run in the first kilometer of a marathon,” Pony.ai co-founder Lou Tiancheng told Nikkei last month. “The real game-changer is how to improve the scalability of the technology. It’s not about how good the one or two cars perform during the tests; we are talking about hundreds of thousands of vehicles.”
The long road ahead
When it comes to raw numbers, China’s abundance of self-driving-car caution has put it squarely behind the US. As of last year, Alphabet’s self-driving car company, Waymo, led the industry in both test miles driven and disengagement rate (how often a human driver has to intervene).
Although some argue that miles driven and disengagement rate are limited metrics, since not all of the data is from real-world open roads, they have thus far been essential benchmarks for the industry. In 2018, Baidu’s cars traveled 29,118 km (18,000 miles) on California roads and saw a disengagement once every 331 km on average. Pony.ai clocked in at 26,322 km (16,355 miles) in California, and registered one disengagement every 1,645 km. Neither came close to Waymo, whose cars traveled more than 2 million km (1.3 million miles) and recorded disengagements only once every 17,731 km.
But these numbers only reflect how self-driving cars engage with, well, California. Years of investment have left China with strong road infrastructure, fast internet, and more electric cars than any other country in the world. All of that could help Chinese AVs communicate with each other and their surroundings more effectively. Zhong Hua, senior vice president of engineering at WeRide, posits that in addition to recognizing road conditions, future AVs could interact directly with things like buildings and traffic lights. Says Zhong, “You can think of coordination between a car and the road infrastructure as adding another sensor to the car.”
Already, at least three provinces and cities—including Zhejiang, Shandong, and Beijing—have announced plans to build “smart” freeways. The Beijing government is setting up two routes for self-driving cars on a 97-kilometer (60-mile) freeway, and in southeastern Zhejiang province, the local government says it’s building a testing route that by 2022 could enable wireless charging and cars-to-infrastructure communication, and serve self-driving cars.
Of course, waiting for pressure-tested technology is also a good way to save money on human employees. Pony.ai’s Lou says AV costs will plunge once the technology is good enough to do away with drivers entirely. In Guangzhou city, for example, a driver’s salary accounts for a third of a taxi’s monthly revenue; but a single lidar sensor, which helps a car map its surroundings, can cost $75,000.
“In China, it makes sense to move more slowly and really refine the algorithm, such that it can better anticipate all of the more unexpected traffic scenarios,” says Marbridge Consulting’s Natkin. “Ten years down the line, China is more likely to have developed an algorithm that can deal with far more complex, chaotic traffic scenarios, which it can then successfully deploy to almost any market in the world.”