Dexhands
SearchJoin community

DEXTEROUS MANIPULATION

The world of dexterous hands.

Compare hardware. Discover applications. Share what works. An open reference and community for dexterous manipulation.

Understand the market ↗Choose a hand ↗Share an experiment ↗Find tools & datasets ↗

LEAP Hand / Carnegie Mellon LEAP Hand v1

Understand this hand

LEAP Hand v1 is an open research platform with an assembly route, development API and linked simulation work. It is a useful starting point for studying a reproducible hand setup.

What distinguishes it
The project links CAD, robot models and software. Its published build cost describes a research build, not a universal delivered-system quote.
Before you build
Pin the v1 hardware and software revisions. Confirm parts availability, build effort and the simulator used by the experiment you want to reproduce.

Editorial interpretation of public information · reviewed 2026-09-24. Manufacturer / project source ↗

Related practical guide →Understand quote scope →
LEAP Hand / Carnegie Mellon LEAP Hand v1
Evidence review · 2026-09-25

Manufacturer, author and seller statements checked against public sources. This is not an independent hardware test. Fields outside this review retain their earlier evidence dates.

Model / DOF
V1 retained as its own generation; not V2 Standard or Advanced. Existing 16-axis value retained; not freshly established by the homepage text. Source ↗
Weight
No new weight measurement or confirmed assembly boundary in this review; retain previous dated evidence. Source ↗
Price
USD 1,771.74 regular parts total, dated August 2023. Historical BOM, not a current finished-hand price; tools, labor and delivery additional. Source ↗
Software / availability
Author API exists; build platform, not a confirmed ready-to-ship commercial offer. Source ↗

Specifications

Record source date: 2026-09-02

Active / total DOF: 16 / 16

Servo position, velocity, current, voltage and temperature; no tactile skin standard

Weight basis: 595 g

Availability / status: Open hardware / buildable

Evidence limits: DIY/open-hardware project. Build-cost estimate excludes purchased assembly/service and is not a comparable commercial sale price.

Price scope: USD 1,771.74 regular parts total, dated August 2023. Historical BOM, not a current finished-hand price; tools, labor and delivery additional.

FROM A RECORD TO A RESULT

Start with LEAP Hand v1

Inspect one simulated cube-reorientation rollout before connecting hardware.

Simulation first

The authors publish a v1 API and an Isaac Lab environment. Dexhands reviewed the instructions; we have not run this simulation or tested the hand.

Sources reviewed 2026-09-30 · no hardware connection from this page
Before you start: versions and requirements

LEAP Hand v1. The environment README reports Ubuntu 22.04, Isaac Sim 4.5 and Isaac Lab 2.1.0. Record the repository commit you actually use.

  • A workstation meeting the linked Isaac Sim requirements.
  • The environment, its documented dependencies and the checkpoint expected by its playback instructions.
  • A place to save the console log, configuration and a short rollout recording.
  1. Match the environment

    Follow the environment installation instructions. Keep the v1 model separate from LEAP v2 Standard and Advanced. Record the simulator versions and commit before changing parameters.

    Environment installation
  2. Try the published simulation playback

    After the upstream setup and checkpoint requirements are satisfied, use its single-environment playback example. This is simulation; do not run the deployment scripts for this first check.

    python scripts/rl_games/play.py --task Isaac-Reorient-Cube-Leap --num_envs 1
    Playback instructions
  3. Check what actually happened

    Record whether the model loads, the policy steps and the cube moves as expected. Keep failed runs and errors. A moving simulation is a setup check, not evidence of real-world grasp reliability.

    Plan a reproducible evaluation

What counts as a useful result?

  • The selected environment starts without missing-asset or dependency errors.
  • The log identifies the model, software versions, commit and checkpoint.
  • The report separates observed behavior from expected behavior, including failures.
If something does not work
Import or asset failure
Compare your installed versions and downloaded assets with the environment README before changing the task.
Wrong hand generation
v1 and the two v2 variants have different designs. A shared brand name does not establish controller compatibility.
Simulation versus hardware
The hardware API has its own port, joint-convention and commissioning requirements. Motor current is not a calibrated fingertip-force measurement.
Original sourcesLEAP v1 projectIsaac Lab environmentv1 device API
Prepare an experiment report

Fill in observed results and failures. Keep credentials, serial numbers and confidential designs out of public reports.

Enable JavaScript to continue to the discussion editor, or copy this template manually.

A discussion is public when approved. Continuing opens an editor; it does not publish or contact a supplier. Sign in to edit the template here, or copy it without an account. Unsaved edits stay on this page until you copy them or open the editor.

Find discussions for this hand
Guide updates

· Added a simulation-first route, upstream environment versions, acceptance checks and an experiment-report template.

This is a Dexhands editorial change, not a new manufacturer release or a completed reproduction.

Software, projects & datasets

What is ready to build with?

01 · Published supportSource statement

Author-linked API repository; select this generation. No hardware compatibility test performed.

Product source ↗
02 · Developer references7 linked references

Check each link’s model scope, release, files and license. A reference is not a tested integration.

03 · Reproduced hereNot independently tested

Dexhands has not reproduced a software setup or hardware test for this record. Published support is not a test result.

Curated public references, checked 2026-09-22. Links describe a model or family relationship, not hardware-tested compatibility.

Dexterous hand URDF collection

DexSuite · Family resource

Research-maintained visual and collision models for multiple hand families.

Inspect the model directory and upstream asset license; revision, size and sensor geometry may differ from your hand.

Hand motion retargeting

DexSuite · Family resource

Human-to-robot motion optimizers with configurations for the listed hand families.

Retargeting generates joint targets; it is not a device driver. Verify model geometry and joint order.

LEAP Hand v1 API

LEAP Hand · Model-specific

Python, C++, ROS and ROS 2 interfaces for the original LEAP Hand.

For v1 hardware. LEAP v2 Standard and Advanced use separate repositories.

LEAP Hand in Isaac Lab

LEAP Hand · Model-specific

Simulation and learning environment for LEAP Hand v1.

Check the required Isaac Lab release and v1 hardware configuration.

LEAP Hand in Isaac Gym

LEAP Hand · Model-specific

Original v1 simulation code accompanying the LEAP Hand project.

Uses the older Isaac Gym stack; it is distinct from the Isaac Lab implementation.

Bimanual glove teleoperation

LEAP Hand / Bimanual Dexterity · Research use

Glove-based control examples for LEAP Hand experiments.

Gloves, calibration and the specified robot setup are required.

DexCap capture and learning pipeline

DexCap authors · Research use

Motion capture processing, LEAP retargeting and demonstration dataset tools.

The project uses its own capture rig and robot setup; raw human motion needs retargeting.

Learn to audit a model-to-hand interface →

Product sources

Original source

Original source

Suggest a correction

Start a discussion

Published posts are public. First contributions are reviewed. Only share material you have permission to disclose.

Report a post