AI BriefingNVIDIA Unveils Cosmos 3 Edge for Robots
Key point
NVIDIA has released Cosmos 3 Edge, a 4B-parameter world model capable of real-time inference on edge devices.
Details
NVIDIA has released Cosmos 3 Edge on Hugging Face, designed to help robots and vision AI agents understand their surroundings and generate actions in real time. This model is an open model with 4 billion (4B) parameters, built to bring data-center-level performance to memory-constrained edge environments.
Cosmos 3 Edge combines two transformer towers:
- Autoregressive tower: processes vision and text tokens to perform situational understanding and reasoning
- Diffusion tower: processes vision, audio, and action tokens to perform prediction and generation
The model achieves a real-time control rate of 15Hz on NVIDIA Jetson Thor, and ranked #1 among models of the same size on the VANTAGE-Bench vision analysis benchmark. It also maps the movements of different physical systems (vehicles, cameras, robot arms, etc.) into a Common Action Representation, enabling consistent control.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Odyssey Releases General-Purpose Physical Intelligence Model 'Odyssey-3'
Apple Releases LensVLM-9B for Processing Compressed Text Images