Installation

Install IsaacSim 4.5.0 and IsaacLab 2.2.1, then install this repository:

pip install -e .

Place it under humanoid_isaac/assets. It should look like

humanoid_isaac/
├─ assets/
│  ├─ robots/
│  ├─ scenes/
│  └─ ...
├─ isaac_utils/
├─ robots/
└─ ...

Humanoid Benchmark Environments

Train a policy:

python scripts/rsl_rl/train.py --task {TASK_NAME}_{ROBOT_TYPE}-v0

Supported tasks (TASK_NAME, 11 total): Stand, Walk, Run, Hurdle, Slide, Stair, Crawl, Push, Package, Truck, Door

Supported robots (ROBOT_TYPE):

G1
H12 (H1-2): limited support — physics, rewards and termination functions may not be fully tuned.

Training options:

Flag	Description
`--headless`	Run without GUI
`--num_envs {int}`	Number of parallel environments
`--video`	Record training videos
`--resume`	Resume from the latest checkpoint
`--checkpoint {str}`	Path to a specific checkpoint to resume from

Evaluate a trained policy:

python scripts/rsl_rl/play.py --task {TASK_NAME}_{ROBOT_TYPE}-v0 --checkpoint {CHECKPOINT_PATH}

Evaluation options:

Flag	Description
`--headless`	Run without GUI
`--num_envs {int}`	Number of parallel environments
`--video`	Record evaluation videos

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
.github		.github
.vscode		.vscode
config		config
docs		docs
humanoid_isaac		humanoid_isaac
motions		motions
scripts		scripts
.dockerignore		.dockerignore
.flake8		.flake8
.gitattributes		.gitattributes
.gitignore		.gitignore
.pre-commit-config.yaml		.pre-commit-config.yaml
README.md		README.md
aggregate_etri_tasks.py		aggregate_etri_tasks.py
aggregate_etri_tasks_lift.py		aggregate_etri_tasks_lift.py
cube_push_collect.sh		cube_push_collect.sh
cube_push_dagger.sh		cube_push_dagger.sh
cube_push_dagger_play.sh		cube_push_dagger_play.sh
cylinder_lift_collect.sh		cylinder_lift_collect.sh
cylinder_lift_dagger.sh		cylinder_lift_dagger.sh
cylinder_lift_dagger_play.sh		cylinder_lift_dagger_play.sh
pyproject.toml		pyproject.toml
setup.py		setup.py
test_lerobot_data.ipynb		test_lerobot_data.ipynb