AI tools ·
NVIDIA's GR00T 2 teaches robots new skills by watching human video
NVIDIA unveiled GR00T 2 on Monday, November 9, 2026, at the Conference on Robot Learning (CoRL) in Austin, Texas — a robot foundation model that learns new manipulation skills largely by watching videos of humans doing the job. The company calls it an open frontier world action model, built on its Cosmos world-model family, and says it predicts future video and robot actions together in a single model.
The training data is the story: NVIDIA says GR00T 2 was trained on 402,730 hours of first-person human video, plus more than 10,000 hours of handheld-gripper demonstrations and 23,500 hours of robot teleoperation. The company reports the model completed 40 of 50 evaluated tasks at least once without any task-specific demonstrations, and it is releasing 100,000 hours of that human video as open data under a Creative Commons license.
The model itself is not open yet — NVIDIA says the GR00T 2 weights and code will be released in December. Alongside it, the company made Isaac Lab 3.0, its open-source robot-training framework, generally available. Foxconn, Siemens, and Nexuni are testing GR00T 2 for factory automation, laundry handling, and industrial use cases.
Why it matters
Robots have been stuck on a data problem: every new skill needed expensive robot demonstrations. If a model can learn from ordinary human video instead, the cost of teaching a robot a new job collapses — and the December open release is when builders will find out whether the claim holds up outside NVIDIA's lab.
Sources
- NVIDIA Blog — Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work
- The Robot Report — NVIDIA launches GR00T 2 and Isaac Lab 3.0 at CoRL
- Unite.AI — NVIDIA Introduces GR00T 2 for Robots Learning From Human Video
Spot an error? We correct quickly and note it. Contact us via the contact page.