EgoX transfers manipulation skills across robot embodiments
Yaru Niu and 21 coauthors submitted the EgoX paper to arXiv on November 12. The framework learns from demonstrations by humans, real robots, and simulated agents, using a Modularized Cross-Embodiment Transformer with Embodiment Dreaming (MXT+). Across 13 task-embodiment pairs, the authors report that cross-embodiment pretraining increased average normalized task progress from 0.63 to 0.80 compared with training from scratch. Under out-of-distribution conditions, average task progress increased from 0.53 to 0.77.
The paper reports measurable gains from cross-embodiment pretraining across six robot embodiments, including five physical robots and one simulated humanoid, providing a framework for improving generalist manipulation policies with heterogeneous demonstrations.
Computer Science > Robotics
arXiv:2611.16728 (cs)
[Submitted on 12 Nov 2026]
Title:EgoX: Egocentric Cross-Embodiment Manipulation with Embodiment Dreaming
Authors:Yaru Niu, Binghong Chen, Changwei Yao, Shangtao Li, Zhenlong Fang, Revanth Krishna Senthilkumaran, Shuai Zhou, Yuemin Mao, Kalie Ching, Po-Ju Chao, Yudong Luo, Peiqi Yu, Heng Zhang, Hao Zhang, Bingqing Chen, Chen Qiu, H. Eric Tseng, Ye Yuan, Umar Iqbal, Changliu Liu, Jonathan Francis, Ding Zhao
View a PDF of the paper titled EgoX: Egocentric Cross-Embodiment Manipulation with Embodiment Dreaming, by Yaru Niu and 21 other authors