Nvidia and Hugging Face open-source Cosmos 3 Edge, a 4B world model for on-device robot control
Nvidia and Hugging Face released Cosmos 3 Edge, an open 4-billion-parameter world model that reasons about a scene and generates robot actions on-device — 32 actions per inference at 15 Hz control on Jetson Thor, and #1 on VANTAGE-Bench among 4B models. Running that on memory-constrained edge hardware instead of a data center moves real-time physical AI onto the robots and cameras themselves, and extends Nvidia's stack from Jetson silicon up to the model.
Source: huggingface.co ↗
A world model helps the robot reason about these relationships. It can predict the visual result of an action, infer the action that caused a change, or generate an action to produce a desired outcome.
Nvidia / Hugging Face
Why this matters
- → Moves real-time robot control from data centers onto edge devices—faster inference, lower latency
- → 4B model achieves state-of-the-art on benchmark vs. peers; open-source enables custom fine-tuning
- → Unifies vision, reasoning, and action in single model; developers can adapt for domain-specific tasks
Physical AI at the edge