EmbodieDreamer: Advancing Real2Sim2Real Transfer for Policy Training via Embodied World Modeling

EmbodieDreamer is a unified Real2Sim2Real framework that incorporates PhysAligner for rapid physical parameter optimization from real observations and VisAligner for generating visually realistic scenes. Furthermore, EmbodieDreamer supports RL training of policy models through preference learning based on trajectory evaluation, and facilitates IL training by generating diverse, unseen observations. Models trained within EmbodieDreamer show significantly better performance compared to policies fine-tuned on real-world data.

Project Status

• [2025/7]： ✅Repository Initialization.

---

BibTeX

@article{wang2025embodiedreamer,
    title={EmbodieDreamer: Advancing Real2Sim2Real Transfer for Policy Training via Embodied World Modeling}, 
    author={Boyuan Wang and Xinpan Meng and Xiaofeng Wang and Zheng Zhu and Angen Ye and Yang Wang and Zhiqin Yang and Chaojun Ni and Guan Huang and Xingang Wang},
    journal={arXiv preprint arXiv:2507.05198},
    year={2025},
}