NVIDIA unveils GR00T 2: robots learn from human demos, 40% zero-shot success
emmanuelvivier · x · 2026-11-12
NVIDIA's GEAR team has introduced GR00T 2, a new generation of foundation world-action models that jointly predict future video and robot actions, running on parallel grippers, dexterous hands, and humanoids.
Key results:
- X-Adaptability: from human-only video, a two five-fingered robot hand lines a lab funnel in 65% of trials; from a handheld UMI gripper, a YAM arm assembles a toy dump truck end to end (60%, a 4-minute long-horizon task); from the DexUMI wearable hand, robots handle dexterous tasks like making banana yogurt and striking a match.
- Zero-/few-shot: 40% success zero-shot across 50 tasks never shown on the target robot; five demos per task (<3 minutes of robot data) substantially lift performance.
- Because the model predicts the outcome it acts toward, a reasoning model can direct it.
Code opens in December.
Related event: NVIDIA releases open-source GR00T 2 humanoid action model(3 posts)→
More from Embodied
- Embodied First Exam: 1,156 tasks on 47 robots test coding agents, best agent solves only 34% in standard setting — burny_tech · 2026-11-12
- Apple affirms Vision Pro commitment; glasses rumored to be in secret development — Scobleizer · 2026-11-12
- ContactBench 1.0: A Benchmark for Contact-Rich Robot Manipulation With Printable Assets — leto__jean · 2026-11-12
- AMD runs MolmoAct2 robot policy locally on a Ryzen AI Max+ 395 mini PC — DJiafei · 2026-11-12
- Open-Source CoRL Paper Browser: 2,100+ Robot Learning Papers, a Decade of Searchable Research — promptdom · 2026-11-12
- World Labs and AMD enter robotics, Li Fei-Fei appears at CoRL — breadli428 · 2026-11-12