ropedia-ai/Ropedia-W0

Ropedia W0

Ropedia W0 is an egocentric world foundation model that learns loco-manipulation from 10k hours of human experience and transfers it across embodiments, from robot arms to humanoids and from grippers to dexterous hands, through one shared kinematic interface of camera poses and projected skeletons.

• Project page: https://ropedia.github.io/ropedia-w0/
• Code: https://github.com/Ropedia/ropedia-w0
• Paper: https://arxiv.org/abs/2610.26855
• Video: https://youtu.be/5I3a1G-mDko
• Xp-Teleop dataset: https://huggingface.co/datasets/ropedia-ai/Xp-Teleop

Checkpoints

File	Size	Model	Training data
w0-5b-pretrain.safetensors	10.0 GB	5B, pre-trained	human egocentric video
w0-5b-midtrain.safetensors	10.0 GB	5B, mid-trained	continued from the pre-trained checkpoint on the human + robot mixture

Both are full DiT state dicts for the Wan2.2-TI2V-5B backbone with the camera (96-dim Plücker) and skeleton (48-dim) condition branches included. The full base-model repository is downloaded separately from Wan-AI/Wan2.2-TI2V-5B. SHA256SUMS lists the checksums. The A14B checkpoints and the benchmark fine-tunes will follow.

Usage

Installation and the input format are described in the repository README. The model generates 81 frames at 640×480 from a single image, a camera trajectory and a skeleton video; the repository ships 18 example clips.

git clone https://github.com/Ropedia/ropedia-w0.git && cd ropedia-w0
hf download Wan-AI/Wan2.2-TI2V-5B --local-dir models/Wan2.2-TI2V-5B   # base model
hf download ropedia-ai/Ropedia-W0 --local-dir weights
CFG=3.0 FPS=30 bash scripts/run_single_case_5b.sh examples/xp_teleop_playdoh_tub weights/w0-5b-midtrain.safetensors outputs/playdoh_tub

Each checkpoint is loaded with --ckpt <file> --ckpt_type full --camera_in_dim 96 --skeleton_in_dim 48 (the defaults of inference_5b.py); --base_model_dir points at the Wan2.2-TI2V-5B download (default models/Wan2.2-TI2V-5B).

License

The weights are released under CC BY-NC 4.0. The code is Apache-2.0; the Wan2.2 base model keeps its own license.

Citation

@article{liu2026ropedia,
  title={Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience},
  author={Liu, Tianqi and Zhao, Hanfei and Liang, Qifeng and Pang, Hui En and Xu, Jiarui and Mu, Sayang and Qiu, Penghao and Zhang, YingSheng and Wang, Jun and Cao, Ziang and Sun, Mingze and Cao, Yukang and Hong, Fangzhou and Chen, Zhaoxi and Liu, Ziwei},
  journal={arXiv preprint arXiv:2610.26855},
  year={2026}
}