Inside Chinese Automakers′ Second Globalization: From Validation Assets and Robotaxi Operations to Physical AI and User Experience AAG 2026

Guangzhou officially launched an industry convention for autonomous mobility services and an initiative for shared autonomous public transport at the 18th China (Guangzhou) Automotive Parts Technology Conference, reflecting China’s push to move autonomous-driving technology into real-world operations and public mobility services.  

Last issue, AEM followed Chinese automakers past the moment a car lands overseas — supplying parts, repairing vehicles, and building diagnostic and customer-support networks into new markets. This time we went further back, to where development and operations actually start. Roads around the world are becoming development data. Autonomous vehicles are turning into services that run every day. AI decisions are beginning to take physical form, in seats that adapt to occupants and chassis systems that anticipate the road. Seven presentations in Guangzhou, in the same week, made the case for where Chinese automakers are building the capabilities behind the finished vehicle.

By Sang Min Han _  han@autoelectronics.co.kr

한글로보기

1. Driving 1 Million km, Keeping Just 40,000 km: How Leapmotor Turns Global Driving Data into Reusable Validation Assets
2. How Hello Is Turning the Mobility Experience of 340 Million Users into an Operating System





AEM again attended AAG 2026 (Auto Aftermarket Guangzhou) at the invitation of Messe Frankfurt Hong Kong. Last issue argued that Chinese automakers' global competition no longer comes down to export volume alone. Once a car reaches another country, it has to be maintained and repaired there, supplied with parts and diagnostic data, and backed by local customer support — a local service infrastructure we called the "second globalization."
But an overseas service network comes into play only after the vehicle reaches the market. If that vehicle is to perform reliably across different roads, climates, and regulatory environments — and if problems found locally are to inform the next product — global requirements have to be addressed much earlier in development.
The 18th China (Guangzhou) Auto Parts Technology Conference, held the same week in the same city, showed that shift already underway. The topics ranged from global validation to robotaxi, from AI agents and intelligent seats to chassis, lighting, and generative AI.
Data no longer just accumulates — it gets refined into assets that can be reused. Autonomous driving is expanding into dispatch, charging, maintenance, and safety monitoring as an operating business. AI decisions increasingly shape how seats, lights, brakes, steering systems, and suspensions respond in the physical world. The competition isn't about how many new features a car carries. It's about how fast experience gathered on the road can be fed back into the next decision and the next product.



The 40,000 km that mattered more than 1 million

Leapmotor's Shi Wei presented 1 million km of driving data gathered across 195 cities in 32 countries. But the number he actually cared about wasn't 1 million. Once raw data is filtered for meaningful scenes and labeled, what survives as scenario knowledge that can be reused for development and validation is about 4 percent — 40,000 km.
The competitive value lies in reuse: applying the data to another market or vehicle, converting a customer complaint into a test condition, and replaying a risk scenario in simulation. According to Leapmotor, 59 percent of the scenarios collected overseas were considered transferable to other markets.
That figure points to a shift underway in how Chinese automakers validate cars globally — from deploying more test vehicles in more places to reusing what one region learns in another, and in the next project. Instead of duplicating development organizations and test environments country by country, they're turning knowledge gathered around the world into assets that can travel.
Building a validated car doesn't finish the job either. Hello Autonomous' Liu Xiaoxi located the next problem in operations.
Hello isn't a traditional automaker or an autonomous-driving startup. It's a mobility platform that grew out of bike-sharing, battery swapping, carpooling, and car rental. What it offered as its case for entering the robotaxi market wasn't algorithm performance alone — it was years of running dispatch, charging, maintenance, billing, and customer service every single day. A vehicle may be technically capable of driving itself, but it cannot become a viable business if it cannot be charged on time or if maintenance and safety monitoring are not integrated into daily operations. If the end-to-end model and the redundant, purpose-built vehicle handle safe driving, the operating platform is what gets thousands of cars out on the road every day and keeps utilization and profitability under control.
The question in robotaxi competition is shifting too — from whether a car can drive itself to whether an operator can send it out safely, every day, at scale, as a service that lasts.



Once AI decides, how does the car move

Desay SV, XPENG, and Dongfeng Motor each showed a different stage of turning AI decisions into a physical experience.
The AI agent Desay SV's Zhou Jian described isn't a voice assistant that answers commands. It understands a user's situation, plans what needs to happen, and calls on vehicle control, mobility, and lifestyle services to carry it out — an integration layer that coordinates driving, cockpit, vehicle control, and the cloud around what the user actually wants.
For that to work, the car has to know more than what a user says. It has to know the state of their body. XPENG Group's Ma Jia didn't present the zero-gravity seat, which reclines to 128 degrees, as a comfort feature. A seat, he argued, remains in contact with the occupant's body for most of the journey and across a large surface area, making it a natural interface for sensing body shape, posture, pressure distribution, heart rate, and breathing. Where seats used to move only after a button was pressed, future seats will read weather, travel time, past habits, and current fatigue to adjust posture, support, temperature, and massage in advance. Save those settings to a cloud account, and they can follow the user from a private car into a rental, a shared vehicle, or a robotaxi.
Dongfeng Motor's Wang Dengfeng took the same problem and scaled it up to the whole vehicle's motion. If intelligent driving is the brain, he said, the chassis is the cerebellum — it has to understand the state of the body and coordinate drive, brake, steering, and suspension. Even a well-planned route is unsafe to execute if the system doesn't know how much grip the tires have on the current surface, or how hard the car can actually brake or turn. The intelligent-driving system has to share its perception of the road with the chassis, while the chassis has to report back what the vehicle can physically execute. Connect that loop, and the car can prepare its suspension before it hits a speed bump, or redistribute drive and brake force before it slips on a wet road. This is physical AI — an AI decision that doesn't stop at a screen, but becomes the car's physical behavior.



Who proves a good experience, and with what

Where the companies laid out a future in which the car and its components judge and react first, Fudan University's Professor Lin Yandan asked whether those reactions are actually safe and comfortable for people.
An autonomous vehicle's blue indicator light drew people's eyes faster than a white one, and made the car easier to identify. But in a real-vehicle nighttime experiment, once brightness passed roughly 102.5 cd/m², identification stopped improving in any meaningful way — while cognitive load and glare discomfort kept climbing. Being more visible and being safer aren't the same thing. Even the same color produced different reactions depending on brightness, mounting position, viewing distance, and surroundings.
Terms like a "healthy cockpit," a "comfortable seat," or a "natural AI agent" are hard to prove as product value without measuring a person's gaze, posture, biosignals, and cognitive load. Competing on experience means turning a user's reaction into an engineering standard that can actually be measured.



Before three months and RMB 300,000 go into a prototype

Once you've measured how people respond, that result has to feed back into product development. Zheng Zhilei, head of LYNWAY's Guangzhou automotive lighting R&D center, showed how the company is putting generative AI and digital twins to work in lamp design, validation, and report writing.
Designing a new interactive lamp and confirming it with an actual prototype has typically taken about three to four months and could cost more than RMB 300,000. Get the combination of screen size, pixel count, optical module, and projection distance wrong, and you build the physical part before finding out — and start over. LYNWAY uses APT Coder and APT Node, along with an internal knowledge base and an enterprise WeChat bot, to search company regulations and past development records, write code, process documents, and generate test reports. Its ISD interactive screens and projection headlamps are tested in digital twins first — adjusting pixels, brightness, color, and projection effects — to catch problems before a physical prototype exists.
AI doesn't replace vehicle-level testing. What it does is bring evaluation forward — before the three-to-four-month physical-prototype cycle is complete — and filter out weak designs before they are built. Whether a result chosen in a virtual environment holds up on an actual road still has to clear optical measurement, human-factors evaluation, and on-road testing.



Clockwise from top left:Wang Dengfeng of Dongfeng Motor; Shi Wei, Head of Global Validation and R&D Digitalization at Leapmotor; Ma Jia, Chief Engineer of Interior and Cockpit at XPENG; Lin Yandan, Professor at Fudan University; Zheng Zhilei of the LYNWAY Guangzhou Lamp R&D Center; and Liu Xiaoxi, Vice President of Hello Autonomous.  



Behind the export network, development and operations connect

It is too early to say that the overseas aftermarket covered last issue is already fully connected with the development and operational capabilities examined here. Much of what these companies presented is still in development or on a long-term roadmap, and the performance and efficiency claims made on stage still need to hold up at production scale, in the field.
But the direction was clear. Scenarios and validation records built during development should be usable in diagnosis and repair overseas. Failures, environmental data, and user data collected abroad should feed back into product development. A vehicle's decisions get built into the physical response of its chassis, seat, and lights — and that response has to be validated by real people on real roads. The results of that validation have to make their way back into code, design, and test conditions.
China's second globalization isn't just about building service centers and parts warehouses overseas. It includes the ability to reuse experience gained in one market in another, to turn technology into an operable service, and to translate a car's decisions into motion a person can actually feel.
What the seven presentations showed wasn't one finished model. It was the foundation — of development and validation, of operations and experience — that different companies and researchers are each building from where they stand. China's next round of automotive globalization does not begin when a vehicle leaves port, nor does it end when the vehicle is sold overseas. The real test is whether experience gathered on the road can return as data, code, service improvements, and better physical responses in the next vehicle.


 

---

AAG 2026 Concludes Its Largest-Ever Edition
54,219 visitors from 93 countries and regions highlight its role as a gateway for China’s aftermarket industry

The 11th Auto Aftermarket Guangzhou (AAG 2026) concluded in Guangzhou, China, on 21 August. This year’s event was the largest in its history, occupying seven halls and 85,000 sqm at the Poly World Trade Expo Center.
The show brought together 1,818 exhibitors and welcomed 54,219 visitors from 93 countries and regions, a 7.5% increase from the previous edition. The top overseas visitor markets were Russia, Myanmar, Thailand, Yemen, Pakistan, Malaysia, Saudi Arabia, Iran, Uzbekistan and Egypt. A total of 48 professional buyer groups from China and abroad also attended.
The exhibition covered parts and components, accessories, customisation, automotive services and remanufacturing. Products ranged from chassis, braking systems, engines and hybrid powertrains to off-road and camping equipment, NEV care products, and remanufactured engines, transmissions and ECUs. Its 36 fringe events addressed overseas expansion, business matching, vehicle customisation and key industry trends.

AEM(오토모티브일렉트로닉스매거진)

---