WristWorld Checkpoints (HuggingFace Model Card)

Public model weights and usage instructions for the 4D world model WristWorld for generating wrist-view videos. This repository contains multiple subdirectories for different components/scales:

• BaseModel/: Base representations/priors (e.g., VAE, encoder, etc.)
• VideoModel/: Spatiotemporal generation backbone (video diffusion/flow matching)
• VGGT/: Auxiliary vision modules for conditioning/estimation (e.g., VGGT)

Citation

If you find WristWorld useful, please cite our paper:

@article{qian2025wristworld,
  title   = {WristWorld: Generating Wrist-Views via 4D World Models for Robotic Manipulation},
  author  = {Qian, Zezhong and Chi, Xiaowei and Li, Yuming and Wang, Shizun and Qin, Zhiyuan and Ju, Xiaozhu and Han, Sirui and Zhang, Shanghang},
  journal = {arXiv preprint arXiv:2510.07313},
  year    = {2025}
}