Valeo Group World Foundation Model developed with NATIX Network

Valeo Group has partnered with NATIX Network, a decentralized physical infrastructure (DePIN) camera network, to develop an open-source multi-camera World Foundation Model designed to learn, predict, and reason about real-world motion and interaction. The initiative focuses on advancing scalable Physical AI capabilities.
By combining Valeo’s deep expertise in world modeling with NATIX’s rapidly growing multi-camera data ecosystem, the collaboration targets faster and safer deployment of Physical AI solutions across autonomous driving and robotics applications. The approach emphasizes real-world understanding rather than simulation-only learning.
The World Foundation Model builds on Valeo’s existing open-source frameworks, including VaViM, a video autoregressive model, and VaVAM, a video-action model. These frameworks have so far relied mainly on front-camera video streams and large-scale online datasets for training and validation.
NATIX Network’s decentralized 360-degree camera infrastructure significantly broadens this scope. The network has already collected more than 100,000 hours of multi-camera driving data within seven months, spanning the US, Europe, and Asia. This dataset enhances spatial awareness and long-term temporal reasoning.
By incorporating diverse viewpoints and regional driving patterns, the Valeo Group World Foundation Model is expected to improve prediction accuracy for complex interactions involving vehicles, pedestrians, and infrastructure under varying traffic and environmental conditions.

• All resulting models, datasets, and training tools will be released openly, enabling developers and researchers to fine-tune world models, benchmark Physical AI systems, and adapt them to region-specific driving behaviors without restrictions.