DeepRoute IO 2.0 Debuts with VLA Model, Five Models Set for Mass Production

Installed on nearly 100,000 vehicles, DeepRoute AI’s urban navigation-assisted system spans SUVs, MPVs, and off-road vehicles across more than ten models.

On August 26, autonomous driving technology developer DeepRoute AI unveiled its next-generation driver-assistance platform, DeepRoute IO 2.0, in Shenzhen. The platform features the company’s self-developed Vision-Language-Action (VLA) model.

The VLA model integrates visual perception, natural language understanding, and action decision-making. Leveraging large language model reasoning capabilities for intelligent driving scenarios, it enhances recognition and response to complex road conditions.

At the launch event, DeepRoute AI founder Zhou Guang explained that the VLA model differs from traditional CNN-based end-to-end systems. Its core advantage lies in its Chain of Thought (CoT) capability—still under development—which enables sequential analysis of discrete information for long-term causal reasoning, allowing more human-like decision-making.

The model also includes an extensive knowledge base, capable of real-time interpretation of newly appearing traffic signs and road conditions, significantly improving generalization.

VLA offers four key functions: spatial semantic understanding, irregular obstacle recognition, text-based traffic guidance comprehension, and memory-driven voice control.

Spatial semantic understanding is particularly notable, enabling proactive risk assessment in visually limited scenarios such as bus occlusions, complex intersections, and underpasses, facilitating early deceleration and safe passage.

Irregular obstacle recognition identifies various unstructured objects, while text-based guidance comprehension decodes temporary signs and road markings. Memory-driven voice control supports personalized voice interactions, integrating safety protocols to execute driving decisions.

DeepRoute IO 2.0 follows a “multi-modal + multi-chip + multi-vehicle” design philosophy, supporting both LiDAR and pure vision solutions, and can be customized for a wide range of mainstream passenger vehicle platforms.

To date, the platform has secured deployment agreements with five vehicle models, with the first batch of production vehicles set to enter the market. DeepRoute AI’s existing urban navigation-assisted driving systems are installed on nearly 100,000 production vehicles, spanning SUVs, MPVs, and off-road vehicles across more than ten collaborative models.

Based on DeepRoute AI’s current customer base, brands such as Great Wall Motors and smart are expected to adopt VLA-assisted driving in the near future.

Zhou Guang noted that DeepRoute AI will continue to expand VLA’s application scope, including Robotaxi operations and the broader Road AGI ecosystem, aiming to promote multi-type intelligent mobility agents and achieve a system evolution from single-point functions to generalized intelligent agents.