Humanoid robots are capable of performing various actions such as greeting, dancing and even backflipping. However, these motions are often hard-coded or specifically trained, which limits their versatility. In this work, we present FRoM-W11, an open-source framework designed to achieve general humanoid whole-body motion control using natural language.
To universally understand natural language and generate corresponding motions, as well as enable various humanoid robots to stably execute these motions in the physical world under gravity, FRoM-W1 operates in two stages:
(a) H-GPT
Utilizing massive human data, a large-scale language-driven human whole-body motion generation model is trained to generate diverse natural behaviors. We further leverage the Chain-of-Thought technique to improve the model's generalization in instruction understanding.
(b) H-ACT
After retargeting generated human whole-body motions into robot-specific actions, a motion controller that is pretrained and further fine-tuned through reinforcement learning in physical simulation enables humanoid robots to accurately and stably perform corresponding actions. It is then deployed on real robots via a modular simulation-to-reality module.
We extensively evaluate FRoM-W1 on Unitree H1 and G1 robots. Results demonstrate superior performance on the HumanML3D-X benchmark for human whole-body motion generation, and our introduced reinforcement learning fine-tuning consistently improves both motion tracking accuracy and task success rates of these humanoid robots. We open-source the entire FRoM-W1 framework and hope it will advance the development of humanoid intelligence.
Release Timeline
We will gradually release the paper, data, codebase, model checkpoints, and the real-robot deployment framework for FRoM-W1.
Here is the current release progress:
[2026/01/21] 🎉🎉🎉 We have released the Technical Report of FRoM-W1!
[2025/12/16] We have released the code to train and evaluate the baseline T2M-GPT on HumanML3D-X at baselines/T2M-GPT.
[2025/12/14] We have released the CoT data of HumanML3D-X on HuggingFace Datasets.
[2025/12/13] We have uploaded the checkpoints for HGPT, Baselines (SMPL-X version of T2M, MotionDiffuse, MLD, T2M-GPT), and the SMPL-X Motion Generation eval model on HuggingFace Models.
[2025/12/10] We have uploaded the initial version of the code for two core modules, H-GPT and H-ACT!
[2025/12/10] We have released our lightweight, modular humanoid-robot deployment framework RoboJuDo!
[2025/12/10] We are thrilled to initiate the release of FRoM-W1!
Usage
The complete FRoM-W1 workflow is illustrated as below:
overview
H-GPT: Deploy H-GPT via command-line tools or a web interface to convert natural-language commands into human motion representations. We provide the complete code for training, inference, and evaluation in this module, with pretrained models available on HuggingFace.
fromw1-hgpt
H-ACT: H-ACT converts the motion representations from H-GPT into SMPL-X motion sequences and further retargets them to various humanoid robots. The resulting motions can be used both for training control policies and executing actions on real robots using our deployment pipeline.
fromw1-hact
Please refer to the preview code in the corresponding folder for now, and we will provide a quick-start example and more detailed README documents later.
Citation
If you find our work useful, please cite it in the following way:
bibtex
1@misc{li2026fromw1generalhumanoidwholebody,
2 title={FRoM-W1: Towards General Humanoid Whole-Body Control with Language Instructions},
3 author={Peng Li and Zihan Zhuang and Yangfan Gao and Yi Dong and Sixian Li and Changhao Jiang and Shihan Dou and Zhiheng Xi and Enyu Zhou and Jixuan Huang and Hui Li and Jingjing Gong and Xingjun Ma and Tao Gui and Zuxuan Wu and Qi Zhang and Xuanjing Huang and Yu-Gang Jiang and Xipeng Qiu},
4 year={2026},
5 eprint={2601.12799},
6 archivePrefix={arXiv},
7 primaryClass={cs.RO},
8 url={https://arxiv.org/abs/2601.12799},
9}
Welcome to star ⭐ our GitHub Repo, raise issues, and submit PRs!
Footnotes
Foundational Humanoid Robot Model - Whole-Body Control, Version 1↩
