Handroid

Bridging Dexterous Hand and Humanoid

Ruogu Li¹, Chenyang Ma¹, Sikai Li¹, Zhenyu Wei¹, Yunchao Yao¹, Haochen Shi², C. Karen Liu², Shuran Song², Mingyu Ding¹

¹UNC Chapel Hill ²Stanford University

Paper Code CAD BOM

A dex hand whenmanipulation matters.

A humanoid whenlocomotion matters.

Dexterous hand
embodiment

Handroid keeps a human-hand-like DoF layout for manipulation, teleoperation, and policy learning.

Shared mechanical
grammar

The same compact modules are reassigned across fingers, arms, legs, and base actuation.

Humanoid
embodiment

The platform crosses into locomotion and loco-manipulation without becoming a different robot.

Design

Fully 3D-Printed · Modular · Easy to Fabricate

Demos

Manipulation, locomotion, and long-horizon behavior.

Long Horizon

Mode switching across an extended task

Humanoid

Pick-and-Place

Humanoid

Push-up and Pull-up

Dex-Hand

Grasping with Diffusion Policy

In-hand Reorientation

Trained in simulation with PPO and deployed on the real Handroid hand, the policy runs at 30 Hz and maintains a 3D-printed cube in hand.

Finger Teleop

Finger-opposition and fist-closing teleoperation validate the hardware and provide motion priors for learning.

Bottle Cap Placement

Handroid picks up a flat cap and places it at a target location, testing thin-object control.

Cup Pick-and-Place

Handroid performs pick-and-place with a metal cup weighing approximately 270 g.

Cup Stacking

Stacking deformable cups demonstrates potential for force-aware manipulation and compliant contact.

Glove Removal

Handroid extracts a rubber glove from the box, showing dexterity with flexible objects.

Dual-Object Grasping

Handroid grasps two objects simultaneously, demonstrating adaptability in a single action.

Paper Grasping

Handroid picks up a sheet of paper and places it into a basket, highlighting fine manipulation.

Pouring Water

Handroid grasps a water-filled bottle and pours water into a cup during dynamic manipulation.

Locomotion Commands

Squatting, moving forward and backward, and turning validate basic locomotion capability.

Push-up

Independent push-up motions demonstrate the support and payload capabilities of the arm structure.

Abstract

Morphology is a key factor that shapes a robot's functional boundaries, task space, and research paradigm. To break the morphological boundary between dexterous hands and humanoid robots and enhance robotic adaptability across diverse tasks, we introduce Handroid, a desktop-scale robotic system that integrates dexterous-hand and humanoid embodiments within a single platform. Handroid has 27 DoFs, a compact size of 0.33 m, and a weight of 2.05 kg. Through teleoperation and reinforcement learning experiments, we demonstrate Handroid's hardware capabilities and application potential in dexterous manipulation, humanoid locomotion, and loco-manipulation tasks that combine both embodiments.

Structure

27 DoF morphology map

Finger joints double as humanoid limbs, making the mapping legible instead of ornamental.

Annotated mechanical structure of Handroid.

Electronics

Integrated control board

Power, sensing, status, and actuator management are packaged for repeatable desktop deployment.

Citation

BibTeX

@misc{li2026handroid,
  title        = {{Handroid}: Bridging Dexterous Hand and Humanoid},
  author       = {Li, Ruogu and Ma, Chenyang and Li, Sikai and Wei, Zhenyu and Yao, Yunchao and Shi, Haochen and Liu, C. Karen and Song, Shuran and Ding, Mingyu},
  year         = {2026},
  howpublished = {Project page},
  url          = {https://handroid.org}
}