Universal Manipulation Exoskeleton: Learning Compliant Whole-body Policies with Real-time Torque Feedback

A research team has introduced the Universal Manipulation Exoskeleton (UME), a lightweight upper-limb device that captures joint torque and arm configuration data while providing real-time haptic feedback to a human operator, according to a paper submitted to the arXiv preprint repository on 12 Jun 2026 [1][2]. The system is designed to address a gap in current robot data-collection pipelines, which the authors say largely lack the ability to record force and torque information needed to train active compliant policies [2]. UME is described as low-cost, lightweight, and portable, and it incorporates an embedded inertial measurement unit to support teleoperation for mobile manipulation tasks [2]. A universal retargeting algorithm allows the exoskeleton to control robots with different kinematic configurations, including the 7DoF OpenArm, 7DoF Franka, and 6DoF X-ARM [2]. The paper reports that the torque feedback is transparent enough to let a blindfolded operator unsheathe kinematically constrained objects [2]. The researchers used data collected through UME to train autonomous policies for bimanual, whole-body, and active compliant behaviors in tightly constrained spaces [2]. The resulting policies achieved high success rates on tasks that included long-horizon mobile manipulation, force-mediated box flipping, visually occluded box pushing, and space-constrained tabletop manipulation [2]. The paper appeared on arXiv, an open-access repository that hosts electronic preprints across physics, mathematics, computer science, and related fields [6]. arXiv does not conduct peer review before posting, though submissions undergo moderation [6]. As of November 2024, the repository was receiving roughly 24,000 new articles per month and had surpassed two million total articles by the end of 2021 [6]. The UME paper is listed under the computer-science subcategory of robotics and includes links to supplementary videos and code [1][2]. On its abstract page, the paper is accompanied by a set of experimental community tools hosted through arXivLabs, a framework that lets third-party developers integrate features such as citation explorers, code finders, and recommender systems directly on the site [4][5]. arXivLabs collaborators operate under guidelines that require adherence to arXiv’s values of openness, community, excellence, and user-data privacy, and they receive only minimal, anonymized user data necessary for their tools to function [4]. The program is currently on a temporary hiatus for new proposals while the arXiv development team focuses on migrating systems to the cloud, though existing Labs and already-submitted proposals are unaffected [3].