MotionBricks Checkpoints (Unitree G1)
Rehosted pretrained checkpoints for MotionBricks: Scalable Real-Time Motions with Modular Latent Generative Model and Smart Primitives — a real-time latent generative model for interactive motion control of the Unitree G1 humanoid.
- Project page: https://nvlabs.github.io/motionbricks/
- Paper: https://arxiv.org/abs/2604.24833
- Original code (Apache-2.0): https://github.com/NVlabs/GR00T-WholeBodyControl (see
motionbricks/)
These weights are a faithful re-upload of the checkpoints released by NVIDIA in the
GR00T-WholeBodyControl repository under motionbricks/out/. They are hosted here to
support a Hugging Face Space demo (loaded at build time via preload_from_hub). No
weights were modified.
Contents
The out/ layout mirrors the paths the MotionBricks code expects:
out/
├── G1-clip.ckpt # motion clip cache (~7.4 MB)
├── motionbricks_vqvae/version_1/checkpoints/model-step=2000000.ckpt # VQ-VAE tokenizer (~272 MB)
├── motionbricks_pose/version_1/checkpoints/model-step=2000000.ckpt # pose model (~1.5 GB)
└── motionbricks_root/version_1/checkpoints/model-step=2000000.ckpt # root model (~391 MB)
The accompanying config/skeleton/stats files (hparams.yaml, skeleton/, stats/)
are small and vendored directly in the demo Space; only the large .ckpt weights are
hosted here.
Usage
from huggingface_hub import hf_hub_download
pose = hf_hub_download(
"suvadityamuk/motionbricks-ckpts",
"out/motionbricks_pose/version_1/checkpoints/model-step=2000000.ckpt",
)
License
This repository redistributes model weights that are licensed under the
NVIDIA Open Model License (the source code of the original project is separately
licensed under Apache-2.0). The full license text is included in this repo as
LICENSE.
Licensed by NVIDIA Corporation under the NVIDIA Open Model License.
The NVIDIA Open Model License permits commercial use, redistribution, and the creation of derivative models with attribution, subject to NVIDIA's Trustworthy AI terms (https://www.nvidia.com/en-us/agreements/trustworthy-ai/terms/).
Citation
@misc{wang2026motionbricksscalablerealtimemotions,
title={MotionBricks: Scalable Real-Time Motions with Modular Latent Generative Model and Smart Primitives},
author={Tingwu Wang and Olivier Dionne and Michael De Ruyter and David Minor and Davis Rempe and Kaifeng Zhao and Mathis Petrovich and Ye Yuan and Chenran Li and Zhengyi Luo and Brian Robison and Xavier Blackwell and Bernardo Antoniazzi and Xue Bin Peng and Yuke Zhu and Simon Yuen},
year={2026},
eprint={2604.24833},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2604.24833},
}