README.md

---
license: cc-by-nc-sa-4.0
language:
- en
- zh
size_categories:
- n>1T
tags:
- robotics
- real-world
- dual-arm
- whole body control
- manipulation
datasets:
- OpenGalaxea/Galaxea-Open-World-Dataset
---

# 🚀 Galaxea Open-World Dataset and G0 Dual-System VLA Model
[![Project Page](https://img.shields.io/badge/Project%20Page-000000?style=for-the-badge&logo=github)](https://opengalaxea.github.io/G0/)
[![Paper](https://img.shields.io/badge/Paper-8A2BE2?style=for-the-badge&logo=arxiv)](https://arxiv.org/abs/2509.00576)
[![Videos](https://img.shields.io/badge/Videos-FF0000?style=for-the-badge&logo=youtube)](https://opengalaxea.github.io/G0/)
[![Visualizer](https://img.shields.io/badge/Visualizer-FF8C00?style=for-the-badge&logo=airplayvideo)](https://opengalaxea.github.io/G0/visualizer/index.html)
[![Modelscope](https://img.shields.io/badge/Modelscope-1890FF?style=for-the-badge&logo=alibabacloud)](https://www.modelscope.cn/organization/Galaxea)
[![Twitter](https://img.shields.io/badge/Twitter-FF6B35?style=for-the-badge&logo=x)](https://x.com/Galaxea_x)
[![Linkedin](https://img.shields.io/badge/Linkedin-1890FF?style=for-the-badge&logo=linkedin)](https://www.linkedin.com/company/galaxeadynamics/posts/?feedView=all&viewAsMember=true)


G0-VLA architecture and training pipeline: Stage 1 pre-trains a vision-language model on cross-embodiment data in an autoregressive manner. Stage 2 and post-train share the same model structure, trained on Galaxea open-world data with embodiment-specific views and high-level and subtask instructions, by supervising the Action Transformer’s action reconstruction with a flow- matching loss.


![image/png](https://cdn-uploads.huggingface.co/production/uploads/6788b21b8d4696a61ff6002c/eK9_KN0mQZArYrvGZYlIB.png)


In this repo, you can find:
- [x] G0_3B_base.pt: **Model Weights after Stage2 Pretraining**
- [x] G0_3B_base_dataset_statistics: **Statistics for Dataset Used in Pretraining** 




## 📜 Citation

All the data and code within this repo are under [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/). If you use our dataset or models, please cite:

```bibtex
@article{galaxea2025,
  title={Galaxea G0: Open-World Dataset and Dual-System VLA Model},
  author={Galaxea Team},
  journal={arXiv preprint arXiv:2509.00576},
  year={2025}
}