robotwin_dual_system_joint_self_attention_mutual

An OpenWAM-Study checkpoint from the Q4 video↔action interaction comparison, run on RoboTwin 2.0. Only the two cross-modality attention blocks change; intra-modality attention stays first-frame causal for video and bidirectional for actions. The fourth mode, action-sees-video, is the baseline robotwin_dual_system_joint_self_attention.

Mask: mutual. Both cross-modality blocks are enabled, so video and action tokens attend to each other. This is the mode OpenWAM-α adopts.

• 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}
}