humanoidrobots.systems

stanford/umi-gripper · spec sheet

Stanford UMI Gripper (Universal Manipulation Interface) specs

Handheld, trigger-activated 3D-printed parallel-jaw gripper with a GoPro and side mirrors for in-the-wild demonstration collection, with policies deployable on robot-mounted grippers.

Maker
Stanford / Columbia / Toyota Research Institute
Class
Gripper
Status
Research
Availability
DIY build
Country
United States, Stanford
Level
L2 Sim-verified
01

Media

3 items · credited and licensed
UMI Gripper (Universal Manipulation Interface) studio render, three-quarter front viewrender
Studio render of the published MJCF, the model's zero pose. Geometry only; colours are the model's own materials.Rendered by Hyperspawn from mujoco_menagerie/umi_gripper (MIT)license: Hyperspawn rendergithub.com
UMI Gripper (Universal Manipulation Interface) studio render, side viewrender
Side view of the same MJCF render, the model's zero pose.Rendered by Hyperspawn from mujoco_menagerie/umi_gripper (MIT)license: Hyperspawn rendergithub.com
4 s turntable of the MJCF studio render.Rendered by Hyperspawn from mujoco_menagerie/umi_gripper (MIT)license: Hyperspawn rendergithub.com
02

Specifications

2/12 fields sourced
Stanford UMI Gripper (Universal Manipulation Interface) specifications
Heightspecs.heightCmunsourced
Massspecs.massKgunsourced
Degrees of freedomspecs.dofunsourced
Payloadspecs.payloadKgunsourced
Top speedspecs.speedMpsunsourced
Runtimespecs.runtimeMinunsourced
Battery capacityspecs.batteryWhunsourced
Pricespecs.priceUsdunsourced
Actuatorsspecs.actuatorsunsourced
Computespecs.computeunsourced
Sensorsspecs.sensorsGoPro camera with side mirrors
Year introducedyear2024
Licenselicense.hw / .sw / .docsSoftware: MIT

Note · Presented at RSS 2024. Repo provides a pretrained checkpoint (cup arrangement). A simplified MJCF is in MuJoCo Menagerie (umi_gripper). The build guide (BOM, CAD, print settings, deployment) is the UMI Hardware Guide Google Doc; assembly is shown in video only. Code is MIT; the gripper mechanism and soft finger are adapted from third-party designs whose licenses are not stated.

03

Openness

method
85 / 100Open hardware
ArtifactPublicWeightPoints
CADopenness.cadyes2525
Bill of materialsopenness.bomyes2020
Sim model (URDF/MJCF)openness.simyes2020
Firmwareopenness.firmwareno150
Softwareopenness.softwareyes1010
Trained policiesopenness.policiesyes1010
score = Σ weight × public85
04

Build package

v0.1.0 · full guide
BOM lines
26
Parts (qty)
24.52
Priced BOM
$946
Fab parts
10
Print time
—
Material
—
Build steps
21
Est. assembly
—
Safety steps
2
Vendors
4

BOM by category · USD from priced lines

CategoryLinesQtyUSD
Sensors55569
Electronics55245
Cabling2340
Structure22.237
Other43.0724
Printed20.7520
Fasteners21.511
Actuators22—
Power11—
Compute11—

Offers in EU · GLOBAL · US from 4 vendors

Subsystems · assembly groups

GroupStepsMinSafety
Fabrication2——
Handheld gripper4——
Camera1——
Software1——
Data collection4——
Processing1——
Policy1——
Deployment7—2

Fabrication: FDM print ×10

Firmware

Targets
GoPro HERO9 (GoPro Labs firmware + UMI settings QR), Weiss WSG-50 (Lua startup script umi/real_world/cmd_measure.lua), UR5e / UR5-CB3 (stock controller, remote control mode)
Repo
github.com/real-stanford/universal_manipulation_interface
Calibrate
Per session: record a gripper calibration video (open/close 5 times) and a mapping video with the 16 mm table marker; `python run_slam_pipeline.py <session>` runs 04_detect_aruco and 05_run_calibrations. GoPro intrinsics ship as example/calibration/gopro_intrinsics_2_7k.json (calibrated with OpenImuCameraCalibrator).
Notes
SLAM uses the authors' ORB_SLAM3 fork (github.com/cheng-chi/ORB_SLAM3) via the chicheng/orb_slam3 Docker image. Time sync uses the umi-gripper.github.io/qrlocal QR timecode.

Policies

  • In-the-wild cup arrangement (rotate espresso cup, place on saucer) · Diffusion Policy (UMI), fine-tuned CLIP ViT-L backbone · real.stanford.eduTrained on handheld UMI data from 30 locations; deployed on UR5e + WSG-50 with a GoPro. Real-world checkpoint, about 3.05 GB. README warns it does not work well in direct sunlight.

Caveats · stated by the package

  • BOM prices are the Hardware Guide's Jan 2024 Amazon US figures; some quantities are fractions of a pack (0.5 PLA spool, 0.25 TPU spool, 0.07 grip tape roll). Mirror, grease, double-sided tape and tag quantities are blank upstream and set to 1 here.
  • Every gripper vendor link is Amazon US (reduced here to /dp/ASIN); there are no EU, CN or IN sources upstream. The guide specifies a GoPro HERO9 with Max Lens Mod 1.0; HERO9 and Max Lens Mod 1.0 availability may be limited now.
  • BOM is per gripper; bimanual collection needs two grippers with different ArUco tag sets.
  • Assembly steps come from the assembly video; there is no written step-by-step assembly doc, so screw placement per step is not documented in text.
  • Deployment hardware is expensive and partly unpriced: UR5e is by quote, the WSG-50 is €3875 (EUR, not converted). Franka and ARX X5 deployments are documented in separate files and repos.
  • The README's download command fetches cup_wild_vit_l_1img.ckpt but its eval command passes cup_wild_vit_l.ckpt (that URL returns 404).
  • SLAM (ORB_SLAM3) is the most fragile stage per the README; scene choice and marker size matter.
  • The software is only tested on Ubuntu 22.04. Code is MIT; the gripper mechanism is adapted from a Thingiverse Push/Pull Gripper and the soft finger from TRI, whose licenses are not stated in the repo.
  • The MuJoCo Menagerie umi_gripper model is a simplified model and is not referenced by the UMI repo.

Package stanford/umi-gripper@0.1.0 · upstream real-stanford/universal_manipulation_interface@d095ba9590df789df5189eea5ee7e431689038a6 (2026-05-26); UMI Hardware Guide (Google Doc) and UMI Data Collection Instruction (Notion) as of 2026-09-25; pretrained checkpoint cup_wild_vit_l_1img.ckpt (Last-Modified 2024-03-16) · 13 package sources

05

Models, repositories, links

5 links
06

Videos

1 tagged · 0 from the maker's channel
UMI - 3D Printing TutorialRobotics & Embodied Artificial Intelligence LabOfficial embedyoutube.comStanford UniversityStanford UMI Gripper (Universal Manipulation Interface)
07

News and papers

0 stories · 0 papers

News mentioning UMI Gripper (Universal Manipulation Interface)

The news feed is not reachable right now; this page regenerates every 30 minutes.

Papers using UMI Gripper (Universal Manipulation Interface)

The papers feed is not reachable right now; this page regenerates every 30 minutes.
08

Timeline

industry timeline

20241

  1. Research

    Universal Manipulation Interface (UMI) published

    Researchers presented UMI, a handheld gripper and policy-learning framework for collecting manipulation demonstrations in the wild without a robot.

    Stanford UniversityStanford UMI Gripper (Universal Manipulation Interface)arxiv.org
09

Sources

last checked 2026-09-25
  1. [1]MuJoCo Menagerie umi_grippergithub.com/google-deepmind/mujoco_menagerie/tree/main/umi_gripperchecked 2026-09-24
  2. [2]UMI project pageumi-gripper.github.io/checked 2026-09-24
  3. [3]arXiv paperarxiv.org/abs/2402.10329checked 2026-09-24
  4. [4]Code repo (MIT)github.com/real-stanford/universal_manipulation_interfacechecked 2026-09-24
  5. [5]UMI Hardware Guide (BOM, CAD, print settings, deployment)docs.google.com/document/d/1TPYwV9sNVPAi0ZlAupDMkXZ4CA1hsZx7YDMSmcEy6EU/edit?usp=sharingchecked 2026-09-25
  6. [6]UMI pretrained checkpointsreal.stanford.edu/umi/data/pretrained_models/checked 2026-09-25
10

Suggest a correction

Corrections need a public source. They go to the Hyperspawn intake queue and are reviewed before the record changes; the page updates on the next data release.

now: —
POST hyperspawn.io/api/v1/reports
11

Related

Nearest by size (height, mass)

Needs a sourced height or mass before size neighbours can be computed.

Also by Stanford / Columbia / Toyota Research Institute

Compare

Compare pages cover humanoid-form robots only.

Compare UMI Gripper (Universal Manipulation Interface) with up to three others

12

Cite this page

humanoidrobots.systems. “Stanford UMI Gripper (Universal Manipulation Interface) specs.” Hyperspawn, data revision 2026-10-08. https://humanoidrobots.systems/stanford/umi-gripper

Machine-readable: /data.json (record stanford/umi-gripper).