Motus_Wan2_2_5B_pretrain

WAN 2.2 Video Generation Model (Stage 1 Pretrained)

Stage 1 pretrained WAN 2.2 (5B) video generation model for Motus. This checkpoint provides the video generation backbone trained on Multi-Robot Task Trajectory, Synthetic Robot Data, and Egocentric Human Videos.

Homepage | GitHub | arXiv | Feishu | WeChat

---

Table of Contents

• WAN 2.2 Video Generation Model (Stage 1 Pretrained)
  • Table of Contents
  • Model Details
    • Architecture
    • Training Details
  • Hardware & Software Requirements
  • Usage in Motus
    • Configuration
    • Download
    • Note on VAE
  • Citation

---

Model Details

Architecture

Training Details

• Stage: Stage 1 (VGM Training)
• Training Data: Multi-Robot Task Trajectory, Synthetic Robot Data, Egocentric Human Videos
• Objective: Text-conditioned image-to-video generation (TI2V)

---

Hardware & Software Requirements

---

Usage in Motus

Configuration

Update your Motus config file (e.g., configs/robotwin.yaml):

model:
  wan:
    checkpoint_path: "./pretrained_models/Motus_Wan2_2_5B_pretrain"  # This checkpoint
    config_path: "./pretrained_models/Motus_Wan2_2_5B_pretrain"
    vae_path: "./pretrained_models/Wan2.2-TI2V-5B/Wan2.2_VAE.pth"  # Local VAE (not included)
    precision: "bfloat16"

Download

# Using Hugging Face CLI
huggingface-cli download motus-robotics/Motus_Wan2_2_5B_pretrain --local-dir ./pretrained_models/Motus_Wan2_2_5B_pretrain

# Or using Git LFS
git lfs install
git clone https://huggingface.co/motus-robotics/Motus_Wan2_2_5B_pretrain

Note on VAE

The WAN VAE (Wan2.2_VAE.pth) is not included in this repository. You need to:

1. Download the original WAN 2.2 VAE from Wan-Video/Wan2.2
2. Set the vae_path in your config to point to the local VAE file

---

Citation

@misc{bi2025motusunifiedlatentaction,
            title={Motus: A Unified Latent Action World Model}, 
            author={Hongzhe Bi and Hengkai Tan and Shenghao Xie and Zeyuan Wang and Shuhe Huang and Haitian Liu and Ruowen Zhao and Yao Feng and Chendong Xiang and Yinze Rong and Hongyan Zhao and Hanyu Liu and Zhizhong Su and Lei Ma and Hang Su and Jun Zhu},
            year={2025},
            eprint={2512.13030},
            archivePrefix={arXiv},
            primaryClass={cs.CV},
            url={https://arxiv.org/abs/2512.13030}, 
}