Featured Works

Three ivory, sage, and terracotta flows representing vision, language, and action merge into an open eye shape

Z-sight: Unifying Vision-Language-Action and Latent World Modeling

Junjie Ye*, Yiqi Zhao*, Muchen Xu*, Celina Shiyu Wang, Hao Dong, Ruohai Ge, Rong Xue, Benjamin J Kolligs, Aditya Dutt, Jyotirmoy V. Deshmukh, Yue Wang, Yuliang Guo

* Equal contribution (co-first authors)

A unified robot model that perceives the present, anticipates the future, and acts within a shared visual representation. Predicting future visual features instead of video pixels enables efficient foresight: 62% average success across five humanoid manipulation tasks, with 110 ms inference per action chunk.

Publishing
Minimal 3D robotic arm holding a terracotta sphere, representing control

Ψ0.1: A Steerable Humanoid VLA Model for Long-Horizon Loco-Manipulation

Hongyi Jing, Jionghao Wang, Songlin Wei, Weiduo Yuan, Liang Heng, Haoyi Lin, Yunwan Fu, Junsoo Kim, Andrew Park, Xiawei Liu, Muchen Xu, Xin Li, Wenping Wang, Di Huang, Yue Wang

Turning imagined goals into whole-body action. A world model predicts goal images from language instructions and a time horizon, guiding a humanoid policy through up to five minutes of autonomous movement and manipulation.

Publishing
A ceramic robot with closed eyes dreaming of a terracotta cube in a thought bubble

WE-Dream: Generalizable Humanoid World-Action Models with Latent-Action Pretraining and Routed Attention

Boqian Li, Liang Heng, Weiduo Yuan, Hongyi Jing, Songlin Wei, Jionghao Wang, Xiawei Liu, Muchen Xu, Haoyi Lin, Henghui Bao, Haonan Chang, Di Huang, Yue Wang

Learning humanoid control from human experience. WE-Dream transfers up to 10,000 hours of egocentric video into a 16B-parameter policy. Routed attention enables action-only inference without generating future video, supporting a continuous 30 Hz whole-body action stream on a single RTX 5090.

Publishing
Two 3D robotic grippers jointly holding a sphere, representing collaboration

DUET: Dual-Robot Understanding via Efficient Teaching

Yiqi Zhao, Ruohai Ge, Celina Shiyu Wang, Junjie Ye, Muchen Xu, Minhao Li, Sergey Zakharov, Basile Van Hoorick, Vitor Campagnolo Guizilini, Leonidas Guibas, Gaurav S. Sukhatme, Jyotirmoy V. Deshmukh, Yue Wang

Teaching two different robots to collaborate by learning from people. DUET combines human-human demonstrations with a small set of robot demonstrations to learn coordinated mobile manipulation, improving task performance while reducing data-collection effort.