Yifei Yao
InternW0-Δ: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data
A world action model combining predictive dynamics, semantic guidance, and geometric priors, pretrained on more than 20,000 hours of open data.
InternW0-Δ integrates video and action experts in a Mixture-of-Transformers architecture. Causal Imprint supplies predictive representations to the action expert without future-video rollout at inference.
I contributed as a coauthor. Please refer to the full author list and technical report for the complete collaboration.
Reference
BibTeX
@misc{yao2026internw0delta,
title = {InternW0-$\Delta$: A World Action Model Bridging Predictive Dynamics and Actions with 20K+ Hours of Open Data},
author = {Miao, Xingyu and Li, Zizun and Fang, Baole and Song, Kaiwen and Wang, Tenghui and Zhang, Hanxue and Wang, Yating and Li, Xudong and He, Yuping and Wei, Xueyuan and Gao, Chao and Yang, Xijie and Xu, Yingxiang and Ren, Kerui and Guo, Wenqi and Zhou, Jianjun and Wang, Xinzhe and Zhao, Weiguang and Yang, Ni and Cai, Zetao and Xue, Yufei and Li, Hengjie and He, Zeyu and Zhou, Yuanzhen and Fu, Rong and Zhang, Jianyang and Cui, Siwei and Huang, Fuxian and Zhou, Yunsong and Gao, Xing and Yao, Yifei and Yu, Qiaojun and Li, Kailin and Zhou, Ming and Huang, Mu and Li, Xinyue and Cui, Wenze and Jiang, Bingqi and Zhu, Xueyue and Dong, Junting and Guo, Haoyu and Lu, Tao and Yu, Mulin and Zhou, Bowen and Zhao, Bin and Xue, Tianfan and Zhang, Weinan and Shen, Chunhua},
year = {2026},
url = {https://arxiv.org/abs/2609.31394},
eprint = {2609.31394},
archivePrefix = {arXiv},
primaryClass = {cs.RO}
}Metadata checked October 27, 2026. Original source ↗