MEGA Hub

HuGe: Scaling Humanoid Generalist Vision-Language-Action Models with Egocentric Experience

Authors

Do you know Yang Liu?You can claim authorship or link another user.Do you know Jiacheng Liu?You can claim authorship or link another user.Do you know Hengtao Li?You can claim authorship or link another user.Do you know Yuheng Huang?You can claim authorship or link another user.Do you know Zijian Yang?You can claim authorship or link another user.Do you know Ji Dai?You can claim authorship or link another user.Do you know Maoxin Wang?You can claim authorship or link another user.Do you know Yi Gao?You can claim authorship or link another user.Do you know Qishen Wang?You can claim authorship or link another user.Do you know Baohua Yin?You can claim authorship or link another user.Do you know Peiyang Li?You can claim authorship or link another user.Do you know Shuanghao Bai?You can claim authorship or link another user.Do you know Mingyang Sun?You can claim authorship or link another user.Do you know Pengxiang Ding?You can claim authorship or link another user.Do you know Jianqin Yin?You can claim authorship or link another user.Do you know Lixin Yang?You can claim authorship or link another user.Do you know Donglin Wang?You can claim authorship or link another user.

Abstract

Extending Vision-Language-Action models (VLAs) to humanoid robots requires coordinated whole-body control and diverse demonstrations that are costly to collect in real-world scenarios. Although egocentric human demonstrations provide a complementary source of task experience, differences in visual embodiment and missing robot proprioception complicate transfer to humanoid policies. To address these challenges, we introduce HuGe, a family of Humanoid Generalist VLAs pre-trained on over 2,000 hours of egocentric human and robot data. Specifically, HuGe employs three existing representative VLA architectures to predict latent motion tokens for a pretrained whole-body controller, together with hand-control signals. Based on this, we introduce a three-stage training pipeline that combines egocentric pre-training, humanoid adaptation, and downstream fine-tuning. To facilitate transfer across embodiments, we construct pseudo-proprioceptive states through simulation playback during pre-training and modify arm appearance in egocentric observations through re-rendering during downstream fine-tuning. With this pipeline, HuGe achieves the state-of-the-art on simulation benchmark HumanoidArena. Furthermore, real-world experiments on the Unitree G1 indicate that augmenting robot demonstrations with visually adapted egocentric demonstrations significantly increases success on a pick-and-drop task.

Community

00