Earlier this month, Shenzhen published a new set of rules which will allow autonomous commercial driving services to be offered in the tech hub from next month. That includes fully autonomous cars if such technology ever comes to fruition and meets safety requirements.

“(The regulation) will be a milestone in the history of autonomous driving in China,” He Xiaopeng, founder of electric vehicle startup Xiaopeng, said in a social media post. Carmakers and analysts expect more cities to follow suit – several have already allowed extensive testing.

Just this week of July 18, Beijing allowed a commercial trial of a small number of taxis in a designated suburban area. Company insiders say the test in the capital, which has been more cautious, could win over cities that are as yet unconvinced. Search engine giant Baidu and self-driving startup Pony.ai were the first cabs off the rank.

China’s autonomous vehicles market is projected to reach US$98.9 billion by 2030, expanding with a compound annual growth rate of 58.9% from 2021, according to a report from Renub Research, a think tank.

Over the past few years, Chinese carmakers and tech companies have been gearing up to compete for their slice of this growth with global giants like Tesla and Google. Some companies like Xiaopeng, Nio and Li Auto have already started selling partially automated cars.

They have also been raising funds to finance self-driving development. In June, Nasdaq-listed Li Auto announced it would issue US$2 billion of new shares to fund new technologies such as driving automation and smart cockpits.

The drive has accelerated as tech companies, roiled by a government crackdown on their businesses, have looked to invest in comparatively safe sectors. Along with cloud computing and data storage, the government priority area of vehicle automation has emerged as one.

From L2 to L3

Shenzhen’s autonomous vehicle standards are split into “levels” of ascending sophistication, similar to those in the US.

A vehicle equipped with level 3 (L3) automation can drive itself in some environments, such as highways, while a level 4 (L4) vehicle can drive itself in most road environments. Level 5 (L5) vehicles, which do not currently exist, would be able to drive themselves in all situations without human input.

Shenzhen requires L3 and L4 vehicles to always have a driver on hand to take over from the system.

Chinese carmakers have already commercialised level 2 (L2) technology, which offers limited steering and speed assistance to drivers. In the first quarter, 23.2% of new cars made in China had L2 automation, up from 7.5% in the same period last year, according to data from market intelligence provider International Data.

Among electric vehicles sold in China, 35% of new cars support L2 automation, in contrast with 19.9% of conventionally powered cars.

The automation technology of EV-makers like Tesla, Xiaopeng and Nio has surpassed that of many traditional carmakers, an engineer at a Chinese self-driving software company told Caixin.

Some EV-makers plan to roll out models that can support L3 and even L4 automation, though most Chinese cities have not legalised these levels of self-driving on their streets.

Xiaopeng’s new G3 model could come with the software and hardware needed for L3 automation next year, analysts at brokerage Kaiyuan Securities wrote in a note. Li Auto plans to incorporate hardware capable of supporting L4 automation in all the new cars starting this year. FAW Jiefang Group, a traditional carmaker, has plans to put L4 autonomous vehicles into commercial use on highways by 2025.

Hardware and software

As with many other smart devices, artificial intelligence chips are at the core of carmakers’ hardware competition, as an autonomous vehicle must process huge quantities of data per second. This increases exponentially as the automation gets more complex – an L4 vehicle must process orders of magnitude more data than its L3 counterpart.

Last year, Nio unveiled its ET7 model EV with four Nvidia Orin chips capable of enough data processing to allow L4 automation. Xiaopeng plans to install two Nvidia Orin chips in its new G9 model.

BYD, the biggest-selling Chinese EV brand, plans to equip its new vehicles with chips made by Chinese chipmaker Horizon Robotics, which said its Journey-5 chip has enough calculation capacity for L3 automation.

Besides chips, which act as the vehicles’ brain, the hardware that acts as the cars’ eyes is also a key area of competition.

Major Chinese carmakers use dozens of sensors including lidar and cameras for their new cars and plan to install more in the future. Automotive lidar uses spinning lasers to scan obstacles around a vehicle.

Chinese carmakers mainly source such technology from foreign companies, such as Velodyne Lidar and Quanergy Systems of Silicon Valley, and Germany’s Ibeo Automotive Systems. But domestic suppliers like RoboSense (Suteng Innovation Technology) and Hesai Technology are catching up in terms of technology and prices, analysts at brokerage Huaxi Securities wrote in a note.

Tesla uses a different approach, relying mainly on cameras to gauge the environment around a vehicle.

“One reason why domestic manufacturers choose this different approach is that their software, especially algorithms, have not caught up with Tesla’s, so their vehicles are not able to drive autonomously using only data collected by cameras,” the software services company engineer said.

A major challenge posed by Chinese companies’ approach is writing algorithms that can combine data collected by cameras, lidar and high-definition maps to inform vehicles’ decision-making, he said.

Also, some market insiders said China’s urban traffic is a particular challenge for autonomous driving testing as local driving habits and frequent construction complicates the picture.

But one advantage that domestic carmakers have over foreign companies like Tesla comes in the form of a hand on the scales. China’s data management rules mean transferring data from vehicles in China overseas is strictly controlled and foreign companies’ access is limited, which means Tesla may have to set up a China-specific software team in the future.

Vehicles and infrastructure

Another pillar of automated driving is infrastructure. The Chinese government, eager to find a productive avenue for economy-boosting infrastructure spending, aims to build smart roads to facilitate autonomous driving – an approach described as “vehicle-to-everything” (V2X).

China is likely to be a leader in smart infrastructure because the country has made, and continues to make, major investments in advanced telecommunication technologies like 5G, accountant company Deloitte wrote in a report.

In the US, by contrast, carmakers tend to rely on vehicles themselves rather than infrastructure to achieve automation because of weaker government infrastructure investment and a more cautious approach to privacy.

Major Chinese carmakers have been developing V2X in selected areas of some cities, such as Wuxi in Jiangsu Province, and in the provincial capital of Changsha. In testing areas, local governments connect infrastructure like traffic lights and cameras to self-driving vehicles to help them drive safely.

Company insiders told Caixin that it is likely that higher-level autonomous vehicles will only be allowed to operate on certain routes after they are legalised in order to guard against safety risks.

In August last year, a young businessman named Lin Wenqin died while driving his Nio ES8 SUV down a highway in Fujian Province, apparently while using his car’s driving assistance function. The incident raised fresh concerns about the safety of such systems. No final report on the incident has been released.

Some in the industry have raised concerns that the marketing of driver assistance systems does not match their current limited capabilities, which can be a risk to both individual drivers and the public.

Autonomous driving has to be developed step-by-step to ensure safety, which means they will first only be allowed in less complex environments, a representative of WeRide, a self-driving technology supplier, told Caixin.

But that could in turn blow out costs, company insiders said. Self-driving tech is expensive, and restrictions on where such vehicles can drive can make it difficult to slash costs by achieving economies of scale, they said.

For now, a feasible revenue source for self-driving technology companies is providing taxi services – known as robotaxis. But rules about where they can drive – and crucially where they are allowed to pick up and drop off passengers – are an obvious drawback.

On Thursday, Baidu unveiled a new, cheaper robotaxi prototype which it plans to mass produce for commercial use by 2024. The Apollo RT6 has been designed for “fully autonomous driving” the company said, and includes a detachable steering wheel. Despite this, the car is billed as an L4 instead of an L5 autonomous vehicle, meaning it is unable to drive itself in all situations without human input.

One industry insider said: “For the development of autonomous driving, creating sustainable business models is as important as researching technologies.”