Despite great advances in learning dexterity, hardware remains a major bottleneck. Most dexterous hands are either bulky, weak or expensive. I’m thrilled to present the RUKA Hand — a powerful, accessible research tool for dexterous manipulation that overcomes these limitations!

RUKA is a tendon-driven, 3D-printable robotic hand. It is fully open-source, costs under $1300, and can be assembled in under 7 hours by a first-time builder. RUKA features 15 degrees of freedom and successfully performs 31 out of 33 grasps in a standard grasp taxonomy.

To keep costs low and control both effortless and precise, RUKA leverages data-driven methods. We autonomously collect data using a MANUS glove attached to RUKA and learn a controller that maps fingertip positions to motor commands.

This approach enables precise control and teleoperation for a wide range of tasks. RUKA can be controlled at 40 Hz and teleoperated at 25 Hz using mainstream devices like motion-capture gloves or VR headsets.

To evaluate its capabilities in policy learning, we ran HuDOR on RUKA across two tasks. Given an open-loop trajectory from human demonstrations, a residual motor policy was learned in under 45 minutes and 40 epochs, successfully improving upon the open-loop and maximizing reward.

To make RUKA accessible to others, we provide a calibration script that determines motor ranges after assembly. This eliminates tensioning differences across builds. With this script, controllers trained on one RUKA hand can be used on another, with 3mm difference in fingertips.

We provide detailed assembly instructions, 3D models and a discord server for all the new time builders / users to talk and ask questions about the hand!

It was a lot of fun being part of RUKA! This project was led by me and @anyazorin, and wouldn’t have been possible without our amazing collaborators: @BillyYYan, @AadhithyaIyer, Lisa Kondrich, @NXBhattasali, and @LerrelPinto.

You can start building your own RUKA today! 📜Arxiv: 🔗 Website: 🏗️Assembly Instructions: 💬 Discord Server: Code and controller weights will be released next week — stay tuned!

Awesome

there will be a need to create increased torgue for canning jar. great steps to collect data

Why 5 fingers? People only use 3 fingers for 99.9% of their daily work. Is it worth adding extra actuators for unused fingers and increasing the cost?

awesome! Thank you for sharing. Can I get the link to the software documentation?

Hello thank you very much! :) We will share the code and the software documentation next week!

This is super cool.

Incredible work here!

Oh wow, this is awesome!

Impressive! Notable that the actuators are Dynamixel vs. feetech. Also very curious on the teleop device. Do you mind sharing specs for it? We're gonna give it a try and try to teleop with our glove:

Also curious what is the arm it is attached to in the saw and nutella demonstration video?

Impressive work.

This is awesome Irmak! :)

looks great

This is incredible!!