AI BriefingNVIDIA Unveils Physical AI Models and Datasets
Key point
NVIDIA has released the Cosmos Transfer model, a large-scale dataset, and Isaac GR00T N1 to advance robotics and autonomous driving.
Details
NVIDIA announced three key open source resources at GTC 2025 to accelerate Physical AI development.
Cosmos Transfer: A New World Foundation Model (WFM) With 7 billion parameters, Cosmos Transfer supports high-precision virtual world scene generation. It can generate photorealistic videos with precisely controlled physical structure and motion by leveraging various multi-control inputs such as 3D bounding boxes, trajectories, depth, and segmentation maps.
Physical AI Dataset: A Large-Scale 15TB Dataset The Physical AI Dataset, designed for robotics training, has been released on Hugging Face. It contains 15TB of data, including over 320,000 trajectories and over 1,000 OpenUSD assets, and can be used for post-training of models such as Cosmos Predict.
NVIDIA Isaac GR00T N1: A Model Dedicated to Humanoids NVIDIA Isaac GR00T N1, the world's first open foundation model for reasoning and skill acquisition in general-purpose humanoid robots, has been launched. It is a cross-embodiment model that processes multimodal inputs including language and visual information.
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.