🦾NVIDIA Cosmos 3 Powers Open Physical-AI Agent Skills
TL;DR
At CVPR, NVIDIA released open physical-AI agent skills built on its new Cosmos 3 omnimodel, automating scene reconstruction, simulation, and policy training. It also shipped Alpamayo 2 Super, an open 32B reasoning model for level-4 self-driving.
At CVPR, NVIDIA released open physical-AI agent skills built on its new Cosmos 3 omnimodel, automating scene reconstruction, simulation, and policy training. It also shipped Alpamayo 2 Super, an open 32B reasoning model for level-4 self-driving.
Key Points
Cosmos 3 unifies vision reasoning, world and action generation in one model
Alpamayo 2 Super is an open 32B vision-language-action driving model
Skills cover AV neural reconstruction, robotics sim, and visual inspection
NVIDIA Physical AI Dataset has passed 15M downloads on Hugging Face
Agent skills released open source on GitHub
Why It Matters
Bundling world models, datasets, and agent skills turns physical-AI research from hand-stitched pipelines into something closer to push-button iteration.
Quick Facts
Frequently Asked Questions
Why does this matter?
Bundling world models, datasets, and agent skills turns physical-AI research from hand-stitched pipelines into something closer to push-button iteration.
What happened?
At CVPR, NVIDIA released open physical-AI agent skills built on its new Cosmos 3 omnimodel, automating scene reconstruction, simulation, and policy training. It also shipped Alpamayo 2 Super, an open 32B reasoning model for level-4 self-driving.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 1,955 builders reading daily.