Open Source RoboticsRobots, software & learning

ASAP

Software

Humanoid motion tracking and delta-action adaptation workflows, with separate simulator training and G1 policy-playback environments. [1]

Installation & first result →Official websiteGetting started

For
Humanoid learning researchers studying sim-to-real adaptation [1]
Source rights
Some licenses are unresolved. Review licenses
Unknowns
operating system, compute requirements not fully established. Review gaps

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

Licenses & related hardware
License
Research use only (README notice), MIT · license not fully assessed for software, datasets · Details by part

Requirements · Research details and dates

On this page

Overview

ASAP combines humanoid motion tracking with a delta-action model intended to reduce simulation-to-real mismatch. Its release contains training, retargeting and G1 policy-playback workflows built on HumanoidVerse and Human2Humanoid. A motion-tracking policy and a delta-action adaptation experiment are distinct stages: the latter needs motion records that also contain an action key. [1]

Is this for you?

Known limitations
  • Delta-action training needs motion data containing an action key.
  • Root MIT license conflicts with package BSD-3-Clause metadata.
  • Models have a separate research-use notice.

Editorial assessment from cited documentation; no runtime testing.

[1] [2] [7] · Editorial assessment

What you need

Dependencies
  • Training route: Python 3.8 and Isaac Gym Preview 4
  • Deployment route: Python 3.10, ROS 2 Humble, Unitree Python SDK, MuJoCo and ONNX Runtime
  • Separate simulator environments are documented

[1] [4]

Installation & first result

Open the upstream getting-started guide [1]

  • Prepare the documented Python 3.10 deployment environment and dependencies.
  • From sim2real start sim_env/base_sim.py with the G1 history configuration.
  • Start the companion policy with matching locomotion and mimic model paths using the README command.

Editorial assessment from cited documentation; no runtime testing.

[1] · Editorial assessment

Installation

Choose one documented environment; simulation playback uses asap_deploy with ROS 2 and Unitree SDK dependencies, separately from training.

[1]

Hardware, models & data

Inputs and outputs

Reference motions and recorded actions support policy/delta-model training; exported policies produce locomotion or imitation in the deployment workflow.

[1]

Project notes

Choose playback or training before installing

This profile reviews commit df5320cc47dd8cad97961bdfabfe402dd62ad999, with no GitHub release listed. The Isaac Gym training route uses Python 3.8 and Preview 4. The separate deployment route uses Python 3.10, ROS 2 Humble, the Unitree Python SDK, MuJoCo and ONNX Runtime. The README also documents separate Isaac Sim and Genesis environments; it does not establish that their current releases can be substituted without checking versions. [1] [4] [6]

For simulation playback, prepare the deployment environment and follow the README's paired simulator/policy commands from sim2real. Start python sim_env/base_sim.py --config=config/g1_29dof_hist.yaml, then run the companion policy command with the supplied locomotion ONNX path and mimic-model directory. Preserve the working directory because the robot configuration uses relative asset paths. [1] [3]

The reviewed configuration selects the Linux loopback interface lo and enables a virtual elastic band. Those are simulation settings, not evidence of unsupported free-standing behavior on a physical G1. The README distinguishes policy activation and switching keys, including ] for locomotion and [ for motion tracking. Keep simulation and physical network configurations separate. [1] [3]

There is additional dependency work to verify: the policy imports pygame and sshkeyboard, but the inspected simulation package dependency list does not declare either. Installing that package alone is not a complete proven playback environment. No installation or policy execution was performed here. [4] [5]

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
Jan 6, 2026 [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 / df5320cc47dd8cad97961bdfabfe402dd62ad999 · software repository

Source files and licenses

Available humanoid-learning research source; package license conflict, model restrictions and environment boundaries explicitly disclosed.

LayerAvailability / license
softwareavailable · License not fully assessed
Root MIT versus setup.py BSD-3-Clause declaration remains unresolved. [2] [7]
model weightsavailable · Research use only (README notice)
README supplies model paths and research-use restriction; no physical deployment validated. [1]
datasetsavailable · License not fully assessed
Motion paths supplied; separate motion/SMPL/AMASS rights unresolved; external datasets not acquired. [1]
documentationavailable · MIT
Repository materials only; external components retain their own terms. [7]

Rights and adaptation data remain separate

The root license and README specify MIT, while the main setup.py declares BSD-3-Clause. That conflict remains unresolved in this profile. The README separately restricts supplied models to research use; SMPL and AMASS assets retain their own terms. A successful demonstration would still not reproduce the full delta-action experiment, which depends on the additional action-bearing data and subsequent policy fine-tuning. [1] [2] [7]

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