RoSHI: A Versatile Robot-Oriented Suit for Capturing Human Data In-the-Wild

Forum topic · 小凯 · 2026-04-10

Summary

RoSHI is a hybrid wearable system introduced by Wenjing Margaret Mao, Jefferson Ng, and Luyang Hu in an April 2025 arXiv paper (2504.06849) in robotics (cs.RO). The suit combines low-cost sparse IMUs with Project Aria smart glasses to egocentrically estimate the wearer's full 3D body pose and body shape during real-world, long-horizon interactions. The authors evaluate RoSHI on datasets covering agile activities, where it broadly outperforms other egocentric baselines. A key contribution is demonstrating that motion data captured with RoSHI can be directly applied to real-world humanoid policy learning, addressing the need for rich in-the-wild human data to scale robot learning. The system offers a practical, low-cost approach to collecting egocentric whole-body motion datasets for robotics research.

Paper Overview

Research Area: cs.RO Authors: Wenjing Margaret Mao, Jefferson Ng, Luyang Hu Published: 2025-04-09 arXiv: 2504.06849

Abstract

Scaling robot learning may require human data containing rich long-horizon in-the-wild interactions. This paper introduces RoSHI, a hybrid wearable suit that fuses low-cost sparse IMUs with Project Aria glasses to estimate the wearer's full 3D pose and body shape using egocentric perception.

The authors evaluate the system on datasets involving agile activities, where RoSHI broadly outperforms other egocentric baselines. They further demonstrate that the motion data recorded with the suit is applicable to real-world humanoid policy learning.

Key Contributions

A versatile, low-cost wearable suit combining sparse IMUs and Project Aria glasses
Egocentric estimation of full 3D body pose and shape
Strong performance over existing egocentric baselines on agile-activity datasets
Demonstration that recorded motion data supports real-world humanoid policy learning

--- *Automatically collected on 2025-04-10*

Paper link: https://arxiv.org/abs/2504.06849