The autonomous driving market in China is expected to reach 10 trillion yuan (US$1.4 trillion) by 2030. Half will be freight, and two-thirds of that freight will be on intercity highways. But it’s going to a long time before tucks really do drive themselves.
Photo by Tang Jun
By WU Yangyu
In late 2021, the J7 convoy sailed onto a designated freight lane. The trucks looked nothing special but in a three-month trial, the fleet covered over 91,000 kilometers, with the trucks driving themselves 96 percent of the time.
The autonomous driving market in China is expected to reach 10 trillion yuan (US$1.4 trillion) by 2030. Half will be freight, and two-thirds of that freight will be on intercity highways.
Autonomous driving companies are making good headway. Since last year, a handful of them are in mass production. PLUS, which developed the J7 fleet, delivered its first order this summer. Two months later, Pony.ai delivered its first batch. Meanwhile, CORAGE announced a plan to build the world’s first autonomous road.
Zooming in, the picture is more nuanced.
“Anyone claiming mass production is a Level 3,” said a manager at an autonomous truck company. “Level 3 means trucks drive themselves most of the time but with a human at the wheel. Level 4, to which all aspire but none have been, human drivers will be a relic of the past.”
The pitch by Level 3 companies is that autonomous trucks save labor. Typically, two drivers take turns on long-haul trips, each paid between 100,000 to 150,000 yuan a year. But CHEN Mo, founder of autonomous truck company TuSimple reckons the claim is pure nonsense.
“You take away one driver, so only lone driver never sleeps?” he said. “A driver in a wonderful Level 3 truck will just sit there for 16 hours straight? Is that even legal?”
The cost-saving argument holds for short trips. XIONG Xingming, CEO of PLUS’s first client Rongqing Logistics, claims autonomous trucks not only allow him to cut one driver per vehicle but are safer and more fuel efficient.
TuSimple does not only sell trucks but, has plans to overload buyers with “services” – a ragbag of anything that can be packaged for sale rather than anything crucial to operations, and a whole new realm of costs and mutual dependency - something most trucking companies have managed to live without thus far. Shippers buy trucks made by a manufacturing partner and then are enslaved forever to TuSimple’s software and services, paid per mile, and lower than the cost of human drivers.
Ambitious, but ineffective, TuSimple has no manufacturing partner and the consensus is that Level 4 mass production is a dream, not a goal. The reality is that there are many unsolved issues regardless of autonomous levels.
“The algorithm has to be more than ready, so that everything, including brake, steering and sensors, is fail-safe,” said a manager at an autonomous driving company. “Then, your manufacturing partner has to be able to build a vehicle for this algorithm, which can easily cost hundreds of millions.”
Convincing the public that vehicles are absolutely safe will not be easy, though in China that may not be necessary: convincing those in authority is probably enough. Even on designated highways algorithms cannot possibly cover all.
Suppose a pig jumps off a truck on the highway, how would the algorithm that has never seen a flying pig in its training data know how to respond? If the LiDAR detects a ten-inch thick object on the road, how can it tell whether it’s a paper bag, safe enough to hit, or a tire that must be avoided?
Then there is the vehicle itself.
“Truck manufacturing has never reached the same level of sophistication as cars,” said engineer ZHANG Xi. “Put the same software on different trucks, their behavior varies wildly.”
New problems, specific to trucks, constantly come up. What if the LiDAR gets dislodged on a bumpy ride? How do you keep lenses clean on dusty roads? Even the most thoroughly tested vehicle will encounter unforeseen issues during its ten-year lifespan. Not to mention the idiosyncratic treatment by both drivers and maintenance workers.
However, autonomous truck companies have no choice but to think beyond safety and performance.
“We try to predict what users want and how to improve their experience,” said Zhang.
This may come down to details as small as notifications. Is a flashing light enough, or should there be voice instruction? Engineers talk to many drivers to understand their preferences; features iterate based on driving data.
“If 90 percent of the drivers don’t use a feature, then it is worthless,” Zhang said.
The pressure to upgrade is intense, both to Level 4 and trips off the highway. PLUS will roll out features that enable trucks to navigate autonomously through toll stations as early as next year. The hope is that by 2025 the trucks will be able to drive themselves on city streets.
It takes persuasion and creativity to bring ideas to commercial reality. Most companies partner with traditional truck manufacturers for mass production – Pony.ai with Anyi, PLUS with Jiefang, for example – often through joint ventures rather than contract manufacturing.
Large manufacturers don’t jump on board easily, considering the hundreds of millions in costs and enormous risk. Sometimes shipping companies are pulled in to secure funding and demand so that the deal is more appealing to everyone. Still, to get a share of the 10 trillion markets, most autonomous truck companies need to convince one small buyer at a time.