Research Analyzer
← Back ICRA 2026

HumanoidExo: Scalable Whole-Body Humanoid Manipulation Via Wearable Exoskeleton

Rui Zhong, Yizhe Sun, Junjie Wen, Jinming LI, Chuang Cheng, Wei Dai, Zhiwen Zeng, Yi Xu, Huimin Lu, Yichen Zhu

PDF

AI summary

Key figure (auto-extracted from paper)
A wearable exoskeleton system bridges the human-robot embodiment gap, enabling scalable whole-body policy learning with minimal real-robot demonstrations.
Humanoid robotics wearable exoskeleton whole-body manipulation policy learning embodiment gap data collection

Problem

Scaling humanoid policy learning is bottlenecked by the high cost and difficulty of collecting diverse real-world robot data, while existing teleoperation and simulation methods suffer from severe embodiment gaps and scalability limits.

Approach

HumanoidExo uses a lightweight wearable exoskeleton fused with LiDAR odometry to capture and retarget whole-body human motion into structured datasets, which train a hybrid Vision-Language-Action and reinforcement learning policy for stable whole-body control.

Key results

  • Enables policy generalization to novel environments
  • Learns complex whole-body tasks from just five real-robot demonstrations
  • Acquires new skills like walking using only exoskeleton data
  • Provides a scalable, in-the-wild data collection pipeline

Why it matters

It offers a practical, scalable pathway for generating diverse whole-body humanoid datasets, accelerating the development of general-purpose robots that can operate reliably in dynamic real-world environments.

Abstract

A significant bottleneck in humanoid policy learn- ing is the acquisition of large-scale, diverse datasets, as col- lecting reliable real-world data remains both difficult and cost-prohibitive. To address this limitation, we introduce Hu- manoidExo, a novel system that transfers human motion to whole-body humanoid data. HumanoidExo offers a high- efficiency solution that minimizes the embodiment gap between the human demonstrator and the robot, thereby tackling the scarcity of whole-body humanoid data. By facilitating the collection of more voluminous and diverse datasets, our approach significantly enhances the performance of humanoid robots in dynamic, real-world scenarios. We evaluated our method across three challenging real-world tasks: table-top manipulation, manipulation integrated with stand-squat mo- tions, and whole-body manipulation. Our results empirically demonstrate that HumanoidExo is a crucial addition to real- 1National University of Defense Technology, 2University of Toronto, 3East China Normal University, 4Tianjin University of Science and Tech- nology, 5Shanghai University, 6Midea Group. ∗Equal contributions. †Corresponding authors. This work was done while Rui Zhong, Yizhe Sun, Junjie Wen, Jinming Li and Yichen Zhu were at Midea Group. robot data, as it enables the humanoid policy to generalize to novel environments, learn complex whole-body control from only five real-robot demonstrations, and even acquire new skills (i.e., walking) solely from HumanoidExo data.

Index terms

Humanoid Robot Systems

Related papers