yjguo/Ctrl-World

<div align="center"> <h2><center>👉 Ctrl-World: A Controllable Generative World Model for Robot Manipulation </h2>

Yanjiang Guo*, Lucy Xiaoyang Shi*, Jianyu Chen, Chelsea Finn

\*Equal contribution; Stanford University, Tsinghua University

<a href='https://arxiv.org/abs/2510.10125'><img src='https://img.shields.io/badge/ArXiv-2510.10125-red'></a> <a href='https://ctrl-world.github.io/'><img src='https://img.shields.io/badge/Project-Page-Blue'></a>

</div>

TL; DR:

**Ctrl-World** is an action-conditioned world model compatible with modern VLA policies and enables policy-in-the-loop rollouts entirely in imagination, which can be used to evaluate and improve the instruction following ability of VLA.

<p> <img src="ctrl_world.jpg" alt="wild-data" width="100%" /> </p>

Model Details:

This repo include the Ctrl-World model checkpoint trained on opensourced **DROID dataset** (~95k trajectories, 564 scenes). The DROID platform consists of a Franka Panda robotic arm equipped with a Robotiq gripper and three cameras: two randomly placed third-person cameras and one wrist-mounted camera.

Usage

See the official **Ctrl-World github repo** for detailed usage.

Acknowledgement

Ctrl-World is developed from the opensourced video foundation model Stable-Video-Diffusion. The VLA model used in this repo is from openpi. We thank the authors for their efforts!

Bibtex

If you find our work helpful, please leave us a star and cite our paper. Thank you!

@article{guo2025ctrl,
  title={Ctrl-World: A Controllable Generative World Model for Robot Manipulation},
  author={Guo, Yanjiang and Shi, Lucy Xiaoyang and Chen, Jianyu and Finn, Chelsea},
  journal={arXiv preprint arXiv:2510.10125},
  year={2025}
}