MotionBricks: Scalable Real-Time Motions with Modular Latent Generative Model

Authoring smart-object behaviors end-to-end inside Unreal Engine 5 — plug-and-play like stacking bricks, with no animation graph wiring and no expert animation knowledge required.

Side-by-side against six state-of-the-art in-betweening baselines.

MotionBricks is now a core organic component of NVIDIA's GR00T Whole-Body Control effort, powering the motion-generation layer of the stack. The full release is now public: the model is fully embedded in GR00T Whole-Body Control's robotics formulation.

The earlier preview release also remains available at GR00T-WholeBodyControl/motionbricks. It ships an interactive G1 demo and a self-contained synthetic training pipeline, with additional instructions for incorporating the BONES-SEED dataset, so the community can reproduce our core training loop and start training their own MotionBricks-style policies.

Finally, the official MotionBricks code is available at NVlabs/motionbricks. The release includes the MotionBricks runtime (Warp, Newton and TensorRT backends) with pretrained SOMA and Unitree G1 models (ONNX bundled, SafeTensors checkpoints on Hugging Face) and a standalone ONNX exporter. It also ships interactive demos (an Inbetweener and a Scene Preview with SOMA and G1 gym scenes), and a Smart Primitive Editor and Scene Editor as one way to author reusable interactions without retraining, along with sample motions and documentation with runnable examples. We distilled the core ideas of our Unreal Engine showcase into plain Python, with a focus on code that is easy to follow.

A sibling project focused on offline motion generation, complementary to MotionBricks' real-time runtime.

Together with MotionBricks, GEAR-SONIC anchors NVIDIA's GR00T Whole-Body Control initiative.

MotionBricks' training corpus — 350k production-grade mocap clips from real human actors and actresses.

The Newton-based solver that retargets SOMA capture onto the G1, producing MotionBricks' training data.

We thank Cyrus Hogg, John Malaska, Will Telford, Jon Shepard, Simon Ouellet, Dmitry Korobchenko, Anna Minx, Edy Lim, Eugene Jeong, Sam Wu, Ehsan Hassani, Charles Zhou, Freya Li, Ling Li, Qiao Wang, and Lina Song for their support and guidance throughout the project.

We thank Alberto Guerra, Boon Cotter, Byeong Gyu Park, Craig Christian, Gabriele Leone, Yenal Kal, Pierre Fleau, Miguel Guerrero, and Marc-Andre Carbonneau from the art team for their help in creating the Unreal Engine 5 demo and its assets.

We also thank Hung Yu Ling, Yue Zhao, Danila Krivenkov, Evgenii Tumanov, Morteza Ramezanali, Winston Chen, Yu-Shiang Wong, Jiefeng Li, Yeongho Seol, Jun Saito, Michael Buttner, Haotian Zhang, Yifeng Jiang, Chen Tessler, Sanja Fidler, Umar Iqbal, Jan Kautz, Yan Chang, and Jim Fan for their helpful discussions and feedback.

@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}, }

This webpage template is adapted from Nerfies, via GEAR-SONIC.

The website source is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.