Open Source RoboticsRobots, software & learning

AMO

Software

Interactive MuJoCo playback of a humanoid whole-body policy with keyboard control of torso posture, velocity and optional arm motion. [1]

Installation & first result →Official websiteGetting started

For
Researchers examining humanoid motion control in a local simulator [1]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
operating system not fully established. Review gaps

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

Licenses & related hardware

Requirements · Research details and dates

On this page

Overview

AMO's public repository offers an interactive MuJoCo demonstration of a humanoid whole-body controller. Keyboard inputs change torso posture, travel velocity and yaw; a toggle adds arm motion. The useful starting point is inspecting the supplied policy's behavior in simulation. The published setup does not provide a training recipe for reproducing the paper from scratch. [1] [3]

Is this for you?

Known limitations
  • Published workflow is interactive model playback, not a documented training recipe.
  • README restricts supplied models to research use.
  • The modeled control interval is not a measured wall-clock performance result.

Editorial assessment from cited documentation; no runtime testing.

[1] [3] · Editorial assessment

What you need

Dependencies
  • Python 3.8 setup
  • MuJoCo 3.2.3
  • PyTorch 2.3.1
  • NumPy 1.23.5
  • mujoco-python-viewer 0.1.4 and GLFW 2.9.0

[1] [2]

Compute for the workload
  • The entry point selects CUDA if available, otherwise CPU; a graphical viewer is required.

Editorial assessment from cited documentation; no runtime testing.

[3] · Editorial assessment

Installation & first result

Open the upstream getting-started guide [1]

  • Check out the reviewed commit and install its requirements in the amo environment.
  • Run python play_amo.py from the repository root.
  • Use the documented keyboard controls in the simulation viewer.

Editorial assessment from cited documentation; no runtime testing.

[1] [3] · Editorial assessment

Installation

Create the documented Python 3.8 environment and install requirements.txt from the pinned checkout.

[1]

Hardware, models & data

Inputs and outputs

Keyboard posture and velocity commands feed the bundled learned policy and adapter, producing simulated humanoid motion.

[1] [3]

Project notes

Run from the checkout so the model paths resolve

This review covers commit 34caaf943660e6f9420e35f64e86dd56fb51dd0e, with no GitHub release listed. Create the documented Python 3.8 environment and install requirements.txt, which pins MuJoCo 3.2.3, PyTorch 2.3.1, NumPy 1.23.5, GLFW 2.9.0 and mujoco-python-viewer 0.1.4. Run python play_amo.py from the repository root. [1] [2] [6]

The script uses relative paths for g1.xml, amo_jit.pt, adapter_jit.pt and adapter_norm_stats.pt. Keep those assets in place. Its entry point chooses CUDA when available and otherwise CPU, but a graphical viewer remains part of the workflow. No throughput or minimum-memory requirement has been verified here. [3]

Use Z/X for torso height, J/U for torso yaw, K/I for pitch and L/O for roll. W/S and Q/E adjust forward and lateral velocity, A/D changes yaw, and T toggles arm actions. Upstream labels its listed command ranges as in-distribution and also invites out-of-distribution experiments; that is not a guarantee that all command combinations are stable. [1]

The simulation timestep is 0.002 seconds with control decimation 10, giving a modeled 0.02-second control interval. This is a configuration value, not evidence that a particular CPU or GPU sustains 50 updates per wall-clock second. [3]

Separate source terms from model-use restrictions

The source license notice specifies Apache-2.0. The README separately says the supplied models are for research use only and warns against casual physical deployment. That model restriction is retained here rather than treating the source license as commercial clearance for the policy. The demo, training and physical hardware were not run for this review. [1] [7]

What we checked

Source researched only; no installation, training, benchmark or physical hardware test performed. No affiliate links or reused media.

Not yet established

  • Requirements: operating system, host computer
  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Nov 19, 2025 [6]
Latest release
Not established
Profile updated
Oct 17, 2026
Sources last checked
Oct 17, 2026
Next review
Nov 17, 2026
Reviewed revision
commit snapshot; no GitHub releases listed / 34caaf943660e6f9420e35f64e86dd56fb51dd0e · software repository

Source files and licenses

Useful simulation demo with public implementation and bundled model; separate research-only model notice disclosed.

LayerAvailability / license
softwareavailable · Apache-2.0
Repository materials only; external components retain their own terms. [7]
model weightsavailable · Research use only (README notice)
Supplied policy/adapter files; README research-use restriction is separate from source Apache-2.0 notice. [1] [3]
documentationavailable · Apache-2.0
Repository materials only; external components retain their own terms. [7]

The documented interactive playback does not require a training dataset.

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