robotwin_dual_system_joint_self_attention_wan22_vae

robotwin_dual_system_joint_self_attention_wan22_vae

An OpenWAM-Study checkpoint from the Q3 visual-encoder comparison, run on RoboTwin 2.0. The encoder is always frozen, so swapping it changes which latent the world stream is asked to predict. The baseline is robotwin_dual_system_joint_self_attention.

Encoder: Wan2.2-VAE, a reconstructive encoder trained on a pixel-reconstruction objective.

• Paper: https://arxiv.org/abs/2609.07398
• Code: https://github.com/OpenWAM-Official/OpenWAM
• Project page: https://openwam-official.github.io/

Citation

@article{wang2026openwam,
  title   = {OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining},
  author  = {Yuran Wang and Siqiao Huang and Mingleyang Li and Chenhao Zhang and Jiaqi Liang and Weiyang Jin and Yue Chen and Xuemin Chi and Donghao Zhou and Qize Yu and Yu-Kai Wang and Yuhan Rui and Shenzhe Yao and Zhen Yuan and Zhenhao Shen and Kefei Zhu and Zijie Zhu and Ning Gao and Xiaowei Chi and Guanqi He and Shanghang Zhang and Hao Dong and Lin Shao and Hang Zhao},
  year    = {2026},
  journal = {arXiv preprint arXiv: 2609.07398}
}