
Nvidia unveils GROOT N2 for humanoid robot training
The AMW Read
The release updates Nvidia's robotics model offering, with reported training gains that could matter across the segment if they hold on different hardware.
Nvidia unveils GROOT N2 for humanoid robot training
Nvidia released GROOT N2, a general-purpose robot foundation model pretrained on roughly 400,000 hours of first-person video, according to ChosunBiz. The footage shows people performing physical tasks. Nvidia said a robot using the model completed a little over 20 of 50 tested motions without additional training, about 40 after three training runs, and about 48 after five. The report does not specify the test conditions or how success was measured.
The release adds to competition in robot foundation models, a horizontal layer intended to serve many physical tasks. GROOT N2 uses a world action model approach that predicts actions alongside changes in the environment, the report says. Nvidia also supplies Isaac platforms and training tools, giving robot makers several connected parts of a development stack. The reported results suggest that human video may reduce task-specific training, but they do not establish performance across different robots or production settings.
For builders and investors, the key test is whether the reported gains transfer to the hardware and tasks customers use. The article notes that performance can vary with robot hardware and components. Teams evaluating GROOT N2 should compare task success, training effort, and repeatability on their own systems. Nvidia's existing relationships with humanoid makers may aid distribution, but partnerships alone do not show how widely the new model will be deployed.



