Volkswagen starts Level 3 road testing in China

Volkswagen Group China has received approval to test Level 3 automated-driving vehicles on public roads in Hefei, Anhui province. The permit also allows large-scale driving-data collection. According to Volkswagen, the approval was granted under a strict regulatory framework for highly automated driving.

Testing has already begun. Volkswagen plans to make its first Level 3 functions available to customers in China from 2027. “At Volkswagen, safety and reliability in autonomous driving have the highest priority. Technological progress must never come at the expense of safety,” says Ralf Brandstätter, CEO of Volkswagen Group China. He adds that the permit enables testing in real traffic and large-scale data collection as the group moves towards Level 3 readiness.

AI, chiplets and virtualisation reshape vehicle computing

AI, chiplets, virtualisation and new interconnects are reshaping automotive computing. At ACC 2026 in Munich, experts will discuss how these technologies are changing future vehicle architectures, development processes and the path towards production-ready systems.

Dr. Martin Large

Where is Volkswagen building its automated-driving stack?

The road-test approval is part of a broader localisation strategy. Volkswagen has expanded its automated-driving activities in China and is using Carizon to build an in-house technology stack spanning software, AI models, data platforms and semiconductors.

In summer 2026, Volkswagen expanded its cooperation with Horizon Robotics and gained white-box access to an AI foundation model. The aim is to accelerate Level 3 development while preparing the same technical basis for later Level 4 applications. Volkswagen plans to develop its own AI driving model capable of interpreting traffic situations, deriving driving decisions and recognising rare scenarios more reliably.

Carizon launched its first L2+ system at the end of 2025. The more advanced L2++ version has since entered series production and is intended to add automated functions for urban traffic. Volkswagen continues to target the second half of 2027 for the first Level 3 functions.

The Chinese roadmap therefore goes further than the current strategies of several European premium OEMs. BMW and Mercedes-Benz are presently concentrating more heavily on Level 2++ development, arguing that higher automation levels still face questions around customer value and system cost.

How is AI expanding virtual validation?

Volkswagen is also increasing the role of AI in development. Its in-house GAIA 2.0 World Model runs ADAS safety simulations equivalent to around five million virtual kilometres every day, using a base of up to 30 million recorded traffic scenarios.

For the next stage, Volkswagen plans to increase the training base to the equivalent of ten million kilometres per day. The objective is to expose the system to rare and unexpected traffic situations during development before customers encounter them on public roads. This reinforces the growing role of virtual testing and validation alongside real-world driving data.

What role do C7H and the China architecture play?

In parallel, Volkswagen is developing the C7H system-on-chip. Together with the AI models and data platform, it is intended to form the technological basis for future Level 4 functions.

The group also plans closer integration with the China Electronic Architecture, the locally developed software and hardware platform for upcoming vehicle generations. Linking the automated-driving stack with the wider E/E architecture should make the technology easier to scale across platforms and model ranges.

When does the vehicle become AI-defined?

The Level 3 programme forms part of Volkswagen’s broader China strategy, which increasingly positions the country as a development centre for AI-based mobility technologies. Locally developed vehicles are intended to use agent-based AI systems that can control functions contextually and connect different software domains.

The longer-term vision is a multi-agent system coordinating driving functions, cockpit applications and digital services. That moves the architecture towards an AI-defined vehicle in which intelligence is not confined to a single assistant or driving function, but becomes a coordinating layer across several vehicle domains.

“AI-defined vehicles separate intent from implementation”

Cockpits are becoming more intelligent, contextual and personalised as assistants, displays and driver interfaces converge. Mercedes-Benz AI specialist Shivam Bahedia explains what it takes to bring generative AI into production vehicles.

Benjamin Müller