- XPeng introduced a multimodal AI architecture targeting Level 4 autonomous mobility.
- The system is planned for robotaxis, EV models, and global deployment by 2027.
The XPeng VLA 2.0 autonomous driving model was unveiled on March 2 as a major advancement in intelligent mobility technology. Developed as a native multimodal large model that integrates vision, language, and action capabilities, the system is designed to accelerate the transition from Level 2 assisted driving toward Level 4 autonomous operation. The announcement signals XPeng's broader ambition to deploy fully autonomous systems across multiple mobility platforms.
Multimodal AI architecture for autonomous mobility
The VLA 2.0 system functions as a physical-world AI model capable of processing real-world data inputs through multiple modalities. By combining perception, contextual reasoning, and action planning, the architecture enables vehicles to interpret complex driving environments more effectively.
Beyond passenger vehicles
In addition to powering advanced driver assistance and automated driving features in passenger vehicles, the technology is designed for use in robotaxis, flying cars, and humanoid robotics platforms. This flexible design allows the model to support multiple autonomous mobility scenarios.
Road testing and deployment roadmap
Robotaxis equipped with the technology have already begun open-road testing, with trial operations planned during the year. International testing programs will follow as part of the company’s global rollout strategy, with large-scale deliveries targeted for 2027. Volkswagen Group is expected to become the first customer integrating the system.
Integration into upcoming electric vehicles
The system debuted alongside the 2026 X9 electric flagship MPV, the first production model built around the architecture. Beginning in late March, the XPeng VLA 2.0 autonomous driving model will also be deployed through software updates to selected vehicle variants, including the G7 midsize SUV and the 2026 P7+ liftback sedan.
Click above to visit the official source.