Open Source RoboticsRobots, software & learning

Human2Humanoid (H2O/OmniH2O)

Software

Humanoid teleoperation and teacher/student learning research code with motion and hardware-specific workflows. [1] [2] [3] [4]

Installation & first result →Official websiteGetting started

For
Robotics developers evaluating the selected source workflow [1] [2] [3] [4]
Hardware
Upstream showcased Unitree H1 EDU, ZED Mini, Orin NX 16 GB, Inspire Hand and DM-J4310-2EC wrist motor configuration [1]
Source rights
Named licenses are documented by part. Includes noncommercial terms. Review licenses
Unknowns
operating system not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware
License
CC-BY-NC-4.0 (noncommercial) · Details by part

Requirements · Research details and dates

On this page

Overview

Human2Humanoid contains H2O/OmniH2O humanoid teleoperation and learning research code. Its workflows distinguish privileged teacher training from distilled student policies and physical deployment. The code is CC-BY-NC-4.0, with an explicit restriction against commercial use including advertising demonstrations; inherited libraries and external motion data retain separate terms. [1] [2]

The documented environment uses Python 3.8, IsaacGym Preview 4 and editable installations of rsl_rl, legged_gym and phc, followed by the requirements file. It specifies numpy below 1.24 and setuptools 59.5.0. Treat this as a documented historical setup, not a resolved environment or interchangeable IsaacLab installation: the README recommends numpy 1.20 while requirements.txt pins 1.23.0, torch is unpinned and two Git dependencies track master. Dependency resolution remains unverified. No simulator or dependency was installed here. [1] [4]

The README supplies distinct teacher/student train and play commands. The config_teleop, task=h1:teleop examples use teacher observation counts 913/990, versus the default student's 1665/1742 and 25-step history. Distillation requires train.dagger.load_run_dagger and train.dagger.checkpoint_dagger matched to an actual teacher run. The example motion is resources/motions/h1/stable_punch.pkl; the optional full filtered motion dataset uses its own motion.motion_file override. Retargeting additionally requires AMASS and SMPL preparation. Inspect those settings and data terms before adapting train_hydra.py. Placeholder checkpoint/run values must be replaced with actual artifacts; none were downloaded or validated here. [1] [3]

The inspected Hydra entry point creates the chosen environment and algorithm runner, saves configuration and calls learning. That is actual orchestration code, not an observed successful training run. Large environment counts in examples are workload settings, not a measured minimum compute budget or guarantee of convergence. [1] [3]

A bounded first step is the documented teacher simulation configuration, followed by its matching motion and checkpoint requirements. The documented physical system is Unitree H1 EDU with a ZED mini, Orin NX (16GB), Inspire hands and DM-J4310-2EC wrist motors, plus the tracking/input integrations and controller references described upstream. Calibration for any adapted deployment remains unresolved by this review. Upstream limits the code to its showcased system and says it will not tune it for other robots. Code access does not commission those interfaces or validate motion limits. No training, policy rollout, model loading, retargeting, physical teleoperation, latency or safety test was performed. [1] [2] [3]

Is this for you?

Known limitations
  • CC-BY-NC4 restriction includes commercial demo use; inherited LeggedGym/RSLRL and external motion-data terms separate.
  • Teacher and distilled student use different observations/history; checkpoint/config interchange is not automatic.
  • No model weights/motion corpus downloaded, training, simulator rollout or real robot teleoperation performed.
  • README NumPy 1.20 differs from requirements 1.23.0; torch and two master Git dependencies are unpinned, so environment resolution is not established.

Editorial assessment of bounded source evidence; no runtime, fit, manufacturing or compatibility proof.

[1] [2] [3] [4] · Editorial assessment

What you need

Languages / runtimes
  • Python 3.8 documented training environment with Hydra/PyTorch/IsaacGym.

Source-reported requirement/configuration, not a tested installation, compatibility matrix or measured runtime result.

[1] [3]

Dependencies
  • Python 3.8, Isaac Gym Preview 4 and matched torch environment
  • Editable rsl_rl, legged_gym and phc; numpy<1.24 and setuptools 59.5.0 documented
  • Selected motion/checkpoint and robot/teleoperation hardware dependencies

[1] [2] [3] [4]

Compute for the workload
  • Physical deployment guide identifies Orin NX 16 GB; training examples select CUDA:0 and large environment counts, not measured minimum training/inference requirements.

Source-reported requirement/configuration, not a tested installation, compatibility matrix or measured runtime result.

[1]

Installation & first result

Open the upstream getting-started guide [1] [2] [3] [4]

  • Inspect the pinned implementation and select the exact configuration described in the article.
  • Follow the chosen upstream environment/build route, resolving the listed dependencies before runtime or hardware integration.

Editorial assessment of bounded source evidence; no runtime, fit, manufacturing or compatibility proof.

[1] [2] [3] [4] · Editorial assessment

Installation

Historical Python 3.8/Isaac Gym Preview 4 route with editable rsl_rl, legged_gym, phc and requirements; NumPy 1.20 README versus 1.23.0 requirements and unpinned dependencies need resolution. No environment installed.

Source-reported requirement/configuration, not a tested installation, compatibility matrix or measured runtime result.

[1] [4]

Hardware, models & data

Supported hardware
  • Upstream showcased Unitree H1 EDU, ZED Mini, Orin NX 16 GB, Inspire Hand and DM-J4310-2EC wrist motor configuration; not verified for other robots.

Source-reported requirement/configuration, not a tested installation, compatibility matrix or measured runtime result.

[1]

What we checked

Pinned source and selected artifact structure review only; no dependency installation, upstream build, native CAD/kernel/fit/strength validation, fabrication, wiring, calibration, simulation, powered operation or performance testing. Original archives stay private.

Not yet established

  • Requirements: operating system, host computer

Updates

Upstream last updated
Not established
Latest release
Not established
Profile updated
Nov 1, 2026
Sources last checked
Nov 1, 2026
Next review
Dec 1, 2026
Reviewed revision
750f1fa

Source files and licenses

Distinct inspectable robotics software/data workflow with specific contribution and bounded next source-investigation step; not performance or physical-build certification.

LayerAvailability / license
softwareavailable · CC-BY-NC-4.0
Selected project implementation; external dependencies, hardware, weights/data and asset rights remain separate. [3] [2]
documentationavailable · CC-BY-NC-4.0
Scoped project declaration; no blanket grant over all linked material. [1] [2]

Outside the selected contribution; no separate released layer claimed.

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.7.0-draft.5

Change history