All terms

Glossary · Platforms

NVIDIA Cosmos

In brief

NVIDIA Cosmos is a foundation-model platform for world models — neural networks trained to simulate physical scenes for robotics. Robotics teams use Cosmos to generate synthetic training data, predict environment dynamics, and pretrain perception components.

Cosmos is the world-model component of NVIDIA's broader robotics stack. It complements GR00T by handling the "what does the world look like" side: synthetic data generation, photoreal scene rendering, and physics-aware video prediction. Robotics teams can train perception and control on Cosmos-generated data and skip a meaningful chunk of real-world data collection.

For humanoid programs, Cosmos is most useful in the sim-to-real loop: realistic synthetic environments at scale, less domain gap than legacy simulators.

Related terms

See it in the wild

Browse robots and brands using these techniques

Glossary entries are upstream. The catalog is where the implementations live.