HUMANS IN THE LOOP

What it’s like riding in an advanced self-driving car in China

WeRide’s fleet.
WeRide’s fleet.
Image: Courtesy of WeRide
By

Two self-driving car companies are currently testing their technology in Guangzhou, one of China’s most congested cities, where bikes and jaywalkers are constantly battling with cars, and at times, going against the traffic.

On a sunny afternoon in August, Quartz took a ride in an autonomous vehicle operated by one of them—WeRide, a two-year-old company based in Guangzhou, with operations in Silicon Valley. The company, previously known as Jingchi, was founded by two former executives of the search giant Baidu. WeRide’s CEO, Tony Han, recently said he’s aiming to launch autonomous taxis as soon as next year.

The vehicle, a black Lincoln MKZ (much like the one I drove in from its Guangzhou rival, Pony.ai, last year), is equipped with a host of sensing technology. It has “Level 4” autonomy, meaning it is capable of driving on its own in challenging conditions. (No company has managed to show off a fully autonomous Level 5 system, considered the top level in the Society of Automotive Engineers’ five-level classification system.) The car was mounted with two lidars—laser-radar sensor technology used by most self-driving car companies to help the vehicles see the world around them—developed by the Chinese company Hesai.

Two “safety drivers” sat in the front of the car during my roughly 30-minute ride. The one sitting in the driver’s seat was there to take over in case the car encountered a situation it couldn’t handle, which is required by road-test regulations in Guangzhou. He kept his hands close to the wheel the entire time. The other was there to monitor the situation of the road and the car. WeRide said the data during the trip was automatically recorded.

WeRide said it also had to inform the local traffic authority before it could take me on a ride outside of its home area, which is a relatively sparsely populated island south of the city. Authorities will coordinate with companies wishing to show off their technology to reporters, WeRide told me.

The ride began at WeRide’s new office building on the island. We crossed over bridges and passed through several tunnels—the longest one was around 1 km (0.6 miles). WeRide claimed the car’s ability to navigate with a weak internet connection inside of the tunnels was a sign of its advanced technology. Internet access is essential for high-precision localization, a mapping technology the car relies on to locate itself precisely within the world.

For most of the journey, I couldn’t really tell that the car wasn’t being driven by a human. We drove at around 40 kmph (24 mph) for most of the trip. The car even managed to cross four lanes of traffic at one point with relative ease.

There was a moment where the car had to make an unprotected left turn in a crossroad—a complex decision in which a car needs to watch out for cars traveling in three directions, including cars coming on its left, on the right, and from behind or next to it, without any traffic lights. (Cars in China drive on the right, like the US.) Our car waited for three cars coming from both left and right before deciding to start turning left. It took around twenty seconds—probably around the same time I would have waited if I had been driving.

There weren’t many cars next to us after that. The WeRide car decided what speed to drive by looking at the speed limits on road signs and how fast car in front of it were moving, according to an engineer in the car. It drove within the speed limit and made a timely brake when a bike suddenly appeared in front of it.

The car also encountered a maneuver that would’ve been a bit tricky even most human drivers. A truck in front of us stopped at an intersection and then suddenly backed up in order to make a right turn into a parking area. A human driver would have two options—back up as well to avoid a crash or switch lanes. Our car chose neither way, instead asking a human to take over. The engineer sitting next to me said the idea of backing up on an open road “is quite intimidating,” so such action is not programmed into the system. (The car has a backing feature for parking though.) The driver could either brake or take control of the wheel to take over the car, the engineer said. In our case, the driver braked before steering the car.

Approaching the end of our ride, we came upon a double-parked car that had stopped by the sidewalk—not unlike how a taxi might stop to drop off a passenger. The WeRide car stopped for around 10 seconds before swerving to the left—and made a steady lane adjustment at the same time when a different car passed us from the left. It wasn’t entirely clear why the car chose to do the maneuver, though it could be because it thought the parked car had stopped for too long. Quartz ran into a similar situation during the ride in Pony.ai’s car. The company said its cars had the capability to change lanes when they came upon a stopped car, but at that time hadn’t made the feature available to for public tests yet because passing can be dangerous.

At WeRide’s Guangzhou office, a real-time dashboard showed the company’s autonomous cars had driven 600,000 km (373,000 miles) since starting road-testing in China around a year and a half ago. But accumulating lots of miles doesn’t necessarily make autonomous cars ready for transporting passengers in the real world. Zhong Hua, senior vice president of engineering of WeRide, declined to reveal where it plans to start its autonomous taxi routes, or how much of the city they will cover.

WeRide’s car handled pretty well in relatively complicated scenarios, and the overall driving experience was smooth. But the algorithm is not perfect—for instance, the company still needs to figure out how to communicate with a pedestrian trying to cross the road. A human driver might just wave his hand telling the pedestrian to pass first.

Given the ride I took, it seems like it’s going to be a while before fully autonomous WeRide taxis hit the road. Most likely, a real-world autonomous taxi ride could be even more complicated and require more improvising than the one I experienced. For instance, if self-driving taxis are programmed to pick up and drop off passengers at fixed destinations, they would need to figure out where to park if the destined location is occupied by another car. If WeRide does eventually launch robo-taxis, they probably won’t be traveling on the most densely crowded streets of Guangzhou, at least at first.

Correction, Aug. 22: The persons sitting in the front are “safety drivers” and they are not involved in monitoring or recording data during the trip. The story earlier said that they were engineers.