GR00T-N1.6-LIBERO-Long

Fine-tuned from nvidia/GR00T-N1.6-3B.

Task Information

LIBERO-10 (Long) benchmark: 95.5% success rate (191/200 episodes)

Usage

from gr00t.policy.policy import Gr00tPolicy

policy = Gr00tPolicy.from_pretrained(
    "TJ-chen/GR00T-N1.6-LIBERO-Long",
    device="cuda:0"
)

# Run inference
action, info = policy.get_action(observation)

Training Configuration

See train_config.json for full training details.

Citation

@inproceedings{gr00tn1_2025,
  title={GR00T N1: An Open Foundation Model for Generalist Humanoid Robots},
  author={NVIDIA},
  booktitle={ArXiv Preprint},
  year={2025},
}

License

Apache 2.0 - See LICENSE