A shared starting point for humanoid skills

The human problem

Building useful humanoid behaviors requires scarce demonstrations, hardware, and repeated integration work.

The prior constraint

Teams had few inspectable general-purpose starting policies that crossed perception, language, and action.

AI’s actual role

The model combines perception and instructions with learned action generation for supported humanoid embodiments.

The documented result

NVIDIA publishes the repository, inference and fine-tuning tooling, and model weights for research and developer experimentation.

Why it may matter

Shared weights can make experiments easier to compare, while safe motion around people still depends on hardware controls, testing, and human oversight.

Limitations

The weights use the NVIDIA Open Model License, which is distinct from an OSI-approved open-source license. Public artifacts do not establish site-specific reliability or physical safety.

Capabilities are described by NVIDIA in its own repository. Public code and downloadable weights are separate artifacts with separate terms.

Unresolved questions