A healthcare robotics variant of GR00T N1.6, post-trained on the Open-H dataset for multi-embodiment surgical and healthcare robot autonomy across 16 robot platforms and 34+ institutions.

GR00T-H post-trains the GR00T N1.6 vision-language-action (VLA) foundation model on surgical robot data from multiple institutions and robot platforms simultaneously. Each institution records data differently — different robots, coordinate conventions, frame rates, camera setups, and state/action representations. GR00T-H solves this by defining per-embodiment modality configs that convert each dataset into a common representation (REL_XYZ_ROT6D for EEF poses) while sharing the core VLA backbone.

The primary additions over upstream Isaac-GR00T live in open_h/:

• Per-embodiment modality configs converting 16 healthcare robot datasets to a common action representation
• Multi-embodiment training config and dataset preparation tooling
• Extensions to the data pipeline (clutch-aware filtering, motion scaling, step filtering)

For general robotics use cases, the upstream Isaac-GR00T project is a better starting point.

• [March 2026] — Released GR00T-H with pre-trained checkpoint and the Open-H dataset

git clone --recurse-submodules git@github.com:NVIDIA-Medtech/GR00T-H.git
cd GR00T-H
uv sync --python 3.10
uv pip install -e .

If flash-attn was not built during uv sync, install it manually:

uv pip install flash-attn==2.7.4.post1 --no-build-isolation

For containerized setup, see the Docker Setup Guide.

uv run python scripts/deployment/standalone_inference_script.py \
  --model-path nvidia/GR00T-H \
  --dataset-path <INSERT_CMR_VERSIUS_DATA_PATH> \
  --embodiment-tag CMR_VERSIUS \
  --traj-ids 0 1 2 \
  --inference-mode pytorch \
  --action-horizon 8

For full inference options including TensorRT, see the inference guide.

uv run torchrun --nproc_per_node=8 --master_port=29500 \
    gr00t/experiment/launch_finetune.py \
    --base-model-path nvidia/GR00T-H \
    --dataset-path <INSERT_CMR_VERSIUS_DATA_PATH> \
    --embodiment-tag CMR_VERSIUS \
    --num-gpus 8 \
    --global-batch-size 32 \
    --max-steps 20000 \
    --output-dir /path/to/output

See open_h/README.md for a deeper dive on finetuning and multi-embodiment training and dataset preparation.

The Open-H dataset comprises 16 healthcare robot embodiments across 34+ institutions, stored in LeRobot format. See open_h/embodiments/README.md for the full embodiment comparison table.

GR00T-H builds on GR00T N1.6, a 3B-parameter vision-language-action model combining a Cosmos-Reason-2B VLM with a 32-layer diffusion transformer action head. The base model is pre-trained on 10k+ hours of robot data across bimanual, semi-humanoid, and humanoid embodiments.

This project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.

@inproceedings{gr00tn1_2025,
  archivePrefix = {arxiv},
  eprint     = {2503.14734},
  title      = {{GR00T} {N1}: An Open Foundation Model for Generalist Humanoid Robots},
  author     = {NVIDIA and Johan Bjorck and Fernando Castañeda, Nikita Cherniadev and Xingye Da and Runyu Ding and Linxi "Jim" Fan and Yu Fang and Dieter Fox and Fengyuan Hu and Spencer Huang and Joel Jang and Zhenyu Jiang and Jan Kautz and Kaushil Kundalia and Lawrence Lao and Zhiqi Li and Zongyu Lin and Kevin Lin and Guilin Liu and Edith Llontop and Loic Magne and Ajay Mandlekar and Avnish Narayan and Soroush Nasiriany and Scott Reed and You Liang Tan and Guanzhi Wang and Zu Wang and Jing Wang and Qi Wang and Jiannan Xiang and Yuqi Xie and Yinzhen Xu and Zhenjia Xu and Seonghyeon Ye and Zhiding Yu and Ao Zhang and Hao Zhang and Yizhou Zhao and Ruijie Zheng and Yuke Zhu},
  month      = {March},
  year       = {2025},
  booktitle  = {ArXiv Preprint},
}

• GR00T-H on HuggingFace — Model weights and checkpoints
• Open-H Dataset — Multi-embodiment healthcare robot benchmark
• Isaac-GR00T — Upstream base model repository
• GR00T N1.6 Blog Post — Base model details
• GR00T N1 Paper — Research paper
• NVIDIA MedTech Open Models

See CONTRIBUTING.md for guidelines.