🌌 DreamX-Phi

DreamX-Phi 1.0: Action-Conditioned Video World Model for Robotic Manipulation

DreamX-Phi is a geometry-aware video world model built on Wan2.2-TI2V-5B. Given an initial observation and a prescribed robot action sequence, it predicts the corresponding future video while preserving robot motion, scene geometry, and the state of the manipulated object.

✨ Highlights

• Geometry-aware action conditioning for end-effector poses and gripper states.
• Interaction-aware learning for coherent scene and object evolution.
• DMD-based few-step generation for efficient inference.
• First place on WorldArena 2.0 Track 1 and tied for second on Track 2.

🚀 Release

Model weights and inference code will be made publicly available after the WorldArena 2.0 IROS Challenge concludes.

📖 Citation

Read the technical report on arXiv. If you find this work useful, please cite:

@misc{dreamxteam2026dreamxphi10actionconditionedvideo,
  title         = {DreamX-Phi 1.0: Action-Conditioned Video World Model for Robotic Manipulation},
  author        = {DreamX Team and Rui Chen and Xiangxiang Chu and Geng Li and Jifan Li and Qingfeng Shi and Datao Tang and Jing Tang and Jun Wang and Pengfei Zhang},
  year          = {2026},
  eprint        = {2608.13489},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2608.13489}
}

📄 License

This repository is released under the MIT License.