GR00T N1-2B
Open humanoid foundation model with a System-2 VLM reasoner and System-1 diffusion transformer for 120 Hz motor commands, trained on real teleoperation, simulation rollouts and neural-video-generated trajectories.
- 76%
- 59K
- 120 Hz
- 780K
Benchmarks
Numbers as reported by the authors on the named evaluation suite.
| Evaluation | Suite | Score |
|---|---|---|
Simulation benchmark average | GR00T N1 sim suite | 76% |
Real GR-1 bimanual tasks | Fourier GR-1 eval | 68% |
Data-efficiency vs BC baseline | Few-shot eval | 32 |
Training
Dual-system VLA (Eagle-2 VLM + diffusion transformer)
2.2B
NVIDIA H100 cluster (DGX)
24K accelerator-hours
6.5K h
780K
Heterogeneous mixture of real, simulated and synthetic (video-model generated) data with an embodiment-tagged action head.
Fine-tunes to a new humanoid with a few hundred demos plus DreamGen synthetic rollouts.
Linked datasets, robots & papers
Resolving related models, datasets and papers…