Currently reading: Turquoise revolution: autonomy the latest battleground in China

Turquoise LEDs lit up the Guangzhou motor show as car makers sought to show off their self-driving technology

If a motor show can have a signature colour, that of the recent Guangzhou event in China was turquoise. Not due to a popular body colour but LEDs illuminated at various places on the exterior of multiple cars.

Turquoise light indicates that a car is driving in a semi-autonomous or autonomous mode, and it has become the latest must-have to lure customers in the cut-throat Chinese car market.

Bright turquoise LEDs were everywhere at Guangzhou, acting as a beacon to showgoers that these models from Xpeng, MG owner SAIC, Aito, Luxeed, Maextro or Li Auto could take control to give the driver welcome respite from the city’s traffic jams.

For now these are semi-autonomous, or level-two-plus (L2+) in the current jargon. This means the driver is paying attention (or meant to be…) but the car is handling much of the lane changing, stop lights, obstacle avoidance and navigation with minimal intervention. 

As I discovered in a trial of Xpeng’s sophisticated technology, this allows hands-off driving for minutes at a time, even if that’s not strictly allowed.

“Autonomous features are one of the biggest battlegrounds for Chinese brands, particularly higher-end ones,” Tu Le, head of China-focused automotive consultancy Sino Insights, told Autocar. 

Robotaxi companies like Pony, WeRide and Baidu are locked in competition with suppliers offering sophisticated ADAS, like Momenta, Huawei and ZYT, which in turn are in a race with car makers that have in-house development teams, such as Xpeng, Nio, BYD, Li Auto and Xiaomi. “The level of competition is unique to China,” Le said.

Huawei’s stand in particular was ablaze with turquoise LEDs on cars from high-end brands such as Luxeed, Aito, Stelato (owned by Chery, Seres and BAIC respectively) that had signed up to Huawei’s Harmony Intelligent Mobility Alliance (HIMA). It included a new brand from SAIC that incorporated the technology in lower-cost models for the first time.

What used to be a differentiation however is now a must-have to appeal to the more heeled buyer. “A solid L2+ system is now an entry point,” Le said.

So car makers are now beginning to promise to take over the driving completely. Xpeng said that next year it will launch level three (autonomous highway driving) and level four (autonomous almost everywhere) self-driving cars on the market.

Even if regulations are loosened (and it’s a big ‘if,’ given the Chinese state’s recent crackdown on unrealistic autonomy claims), is this promise viable?

Back to top

Like Tesla and Britain's Wayve, Xpeng relies on the smarts of its 'end-to-end' AI model to react to dangers, rather than the more usual belt-and-braces manually programmed, rules-based system backed up with high-definition maps and lidar sensors. 

As I recently experienced in a ride in Wayve's autonomous Ford, this approach can be very impressive. It removes the rules consultation process for a faster-reacting, more human driving experience, even if regulators have yet to be convinced.

In Guangzhou I got behind the wheel of an Xpeng P7+ running Xpeng’s current autonomy system, called Navigation Guided Pilot (NGP), and found it to be largely reliable. It allowed prolonged periods of hands off driving in the city's often chaotic streets, in most cases correctly second-guessing the legendary impatience of China’s commercial drivers, whether they were operating ride-hailing cabs, food-delivery scooters or goods trucks. Relaxing, however, it was not, given the potential for things to go wrong suddenly. 

The next version of NGP will be informed by Xpeng’s 'VLA 2.0' AI large language model (LLM), arriving next year, in a swathe of new models including an updated P7+. It's a big step forward towards full autonomy, the company claims.

“The difference is huge,” Jacky Wu, Xpeng’s head of development, told Autocar. “For example, in the P7+ you sometimes you can feel the car braking [hard] or making less smart decisions. The new system will be totally different.”

Tesla still largely clings to the idea that autonomy is a mere software update away, but Xpeng has acceded to the reality that this extra ability requires serious computing power. So models will be assigned different trim levels based on their chip specs.

Max cars will pack one of the company’s new Turing chips, capable of 750 TOPS (trillion operations per second) – three times the one in the new electric Mercedes-Benz GLC EQ, itself touted as one of the industry’s smartest cars. Meanwhile, Ultra cars will come with up to 2250 TOPS, thanks to three Turing chips, and Robo cars will get the ability to properly self-drive, thanks to four Turing chips and full redundancy. Next year, Xpeng plans to launch three models with a Robo spec. 

Back to top

Xpeng is offering its VLA 2.0 LLM to other car makers, and partner company Volkswagen is going to be the first customer, fitting it to a China-only model coming next year – likely to be the Volkswagen ID Unyx 8, which is based on the Xpeng G9.

Xpeng claims that the intelligence of its new VLA 2.0 LLM means it can quickly assimilate new surroundings, for example places outside China. 

However, unlike its battery technology, China’s great leap forward in autonomous driving ability is likely to be locked within its borders for some time yet, as geopolitical concerns and lack of driving data from other countries prevent a global rollout.

To the relief of global car makers and tech suppliers, this one automotive advantage that China will struggle to export.

Join our WhatsApp community and be the first to read about the latest news and reviews wowing the car world. Our community is the best, easiest and most direct place to tap into the minds of Autocar, and if you join you’ll also be treated to unique WhatsApp content. You can leave at any time after joining - check our full privacy policy here.

Add a comment…