Skip to content
RobothecaBeta
Log in

Universal Manipulation Interface (UMI)

  • Software
  • Humanoids and robot learning
  • Framework
  • Runs on hardware you can build

UMI is a handheld gripper and software pipeline for collecting robot manipulation demonstrations and training visuomotor policies from them, deployable on UR5 and ARX X5 arms.

Made by Columbia AI and Robotics Lab, now the Stanford REAL lab, United States (checked on github.com/real-stanford (opens in a new tab)).

Source on GitHub: github.com/real-stanford/universal_manipulation_interface (opens in a new tab). Facts read at commit d095ba9 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit d095ba9 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Runs a SLAM pipeline on GoPro footage of handheld demonstrations to recover gripper trajectories.
  • Converts processed demonstration sessions into a replay buffer dataset for training.
  • Trains Diffusion Policy models on that dataset, with single-GPU and multi-GPU options.
  • Provides a real-world evaluation script that drives a UR5 arm and WSG50 gripper with a SpaceMouse.
  • Includes a pretrained cup arrangement policy and a processed in-the-wild cup dataset.

Written by Robotheca from the project's README (opens in a new tab) at commit d095ba9.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
Tested on Ubuntu 22.04 with Docker installed. Single-GPU training was tested on an RTX3090 with 24GB; multi-GPU training is supported. Deployment needs a UR5 arm with teach pendant, a WSG50 gripper with web interface, a GoPro with HDMI capture card, and a 3Dconnexion SpaceMouse; ARX X5 arms are also

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
26 May 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots on Robotheca that use it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists software as a reference: nothing is hosted or run here, and every fact links to where it was read.