Yifei Yao
InternW0: A Foundational Physical World Model for Efficient Real-World Interactions
An asymmetric video-action architecture couples future visual dynamics with continuous robot control, supporting asynchronous prediction and contact-aware interaction.
InternW0 uses a high-capacity video expert for longer-horizon context and a lightweight action expert for faster control. It is trained on heterogeneous robot and egocentric data and evaluated in simulation and real-world scientific workflows.
I contributed as a coauthor. This entry is a collaborative technical report.
Reference
BibTeX
@misc{yao2026internw0,
title = {InternW0: A Foundational Physical World Model for Efficient Real-World Interactions},
author = {Cai, Jisong and Mu, Yao and Yang, Ganlin and Cao, Zhe and Tu, Zhangzheng and Gao, Xing and Li, Kailin and Zhan, Xinyu and Yang, Lixin and Zhu, Yangkun and Ma, Haoxiang and Zhou, Ming and Yu, Qiaojun and Xue, Yufei and He, Liqun and Yao, Yifei and Zhu, Yifan and Ling, Long and Jiang, Bingqi and Guo, Haoyu and Zhu, Xueyue and Zhou, Bowen and Zhao, Bin and Xue, Tianfan and Shen, Chunhua and Zhang, Weinan},
year = {2026},
url = {https://arxiv.org/abs/2609.27656},
eprint = {2609.27656},
archivePrefix = {arXiv},
primaryClass = {cs.RO}
}Metadata checked October 27, 2026. Original source ↗