"Guide Humanoid-Gym users through PPO training, XBot-L environment
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill humanoid-gym --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Humanoid Gym?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-humanoid-gym)More formats (shields.io, HTML) on the badges page.
---
name: humanoid-gym
description: "Guide Humanoid-Gym users through PPO training, XBot-L environment
customization, and MuJoCo sim-to-sim deployment."
disable-model-invocation: true
metadata:
disco-role: operating
license: NO_LICENSE
---
# Humanoid-Gym
Use this repo skill when a future agent needs to work with Humanoid-Gym: training or evaluating the bundled `humanoid_ppo` task, changing the XBot-L environment, or validating/exporting a policy for MuJoCo sim-to-sim.
## Read this first
- `references/repo-provenance.md` if you need to compare the current checkout against the snapshot used to build this skill.
- `references/installation-and-backends.md` for Python, torch, Isaac Gym, MuJoCo, and CUDA readiness.
- `references/troubleshooting.md` when setup, import, backend, or checkpoint selection fails.
## Main routes
- `sub-skills/training-and-evaluation/` — PPO training, evaluation, checkpoint loading, and TorchScript export.
- `sub-skills/environment-customization/` — XBot-L config, rewards, terrain, and new robot/environment setup.
- `sub-skills/sim2sim-deployment/` — MuJoCo deployment of exported policies and asset/policy validation.
## Verified package facts
- Distribution name: `humanoid`
- Version: `1.0.0`
- Registered public task in this checkout: `humanoid_ppo`
- XBot-L policy dimensions: 705 observations, 219 privileged observations, 12 actions
- The public `play.py` export path writes `logs/<experiment_name>/exported/policies/policy_1.pt`
## Minimal install and check
Use a Python 3.8 environment for the repo's documented backend stack.
1. Install this checkout in editable mode from a local clone.
2. Install the runtime stack required by the selected route.
3. Confirm `import humanoid` and `python -m pip check`.
4. If you need train/play execution, confirm Isaac Gym Preview 4 is installed and importable.
Suggested smoke check:
```bash
python -c "import humanoid; print(humanoid.LEGGED_GYM_ROOT_DIR)"
```
If Isaac Gym is missing, stay in static/command-building mode and use the bundled helpers instead of claiming native training or evaluation succeeded.
## Helper script
- `scripts/inspect_humanoid_gym_install.py` — run this when you want a safe root-level install/resource/backend summary without launching Isaac Gym.
## Notes for future agents
- `humanoid/scripts/train.py` and `humanoid/scripts/play.py` are the user-facing PPO entry points, but they require the Isaac Gym backend for real execution.
- `humanoid/scripts/play.py` exports the actor only and hard-codes render/export side effects unless the source file is edited.
- `humanoid/scripts/sim2sim.py` uses MuJoCo viewer rollout and is best validated with the bundled sim2sim sub-skill.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!