MyoSuite 2.12.2

Musculoskeletal environments simulated in MuJoCo

MyoSuite is a collection of musculoskeletal environments and tasks simulated with the MuJoCo physics engine and wrapped in the OpenAI gym API to enable the application of Machine Learning to bio-mechanic control problems.

Documentation | Tutorials | Task specifications

Below is an overview of the tasks in the MyoSuite.

Installations

You will need Python 3.10 or later versions.

Using uv

uv is a fast Python package manager. Install MyoSuite with:

uv sync -p 3.10

Using conda

It is recommended to use Miniconda and to create a separate environment with:

conda create --name myosuite python=3.10
conda activate myosuite
pip install -U myosuite

Using uv (recommended)

For faster installation, you can use uv, a fast Python package installer:

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create and activate a virtual environment
uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install MyoSuite
uv pip install -U myosuite

For installation from source with uv:

git clone --recursive https://github.com/myohub/myosuite.git
cd myosuite
uv pip install -e .

For advanced installation options, see here.

Test your installation using the following command (this will return also a list of all the current environments):

# With uv:
uv run python -m myosuite.tests.test_myo

# With pip/conda:
python -m myosuite.tests.test_myo

You can also visualize the environments with random controls using the command below:

# With uv:
uv run python -m myosuite.utils.examine_env --env_name myoElbowPose1D6MRandom-v0

# With pip/conda:
python -m myosuite.utils.examine_env --env_name myoElbowPose1D6MRandom-v0

NOTE: On MacOS, we moved to mujoco native launch_passive which requires that the Python script be run under mjpython:

mjpython -m myosuite.utils.examine_env --env_name myoElbowPose1D6MRandom-v0

It is possible to take advantage of the latest MyoSkeleton. Once added (follow the instructions prompted by uv run myoapi_init or python -m myosuite_init), run:

# With uv:
uv run python -m myosuite.utils.examine_sim -s myosuite/simhive/myo_model/myoskeleton/myoskeleton.xml

# With pip/conda:
python -m myosuite.utils.examine_sim -s myosuite/simhive/myo_model/myoskeleton/myoskeleton.xml

Examples

It is possible to create and interface with MyoSuite environments just like any other OpenAI gym environments. For example, to use the myoElbowPose1D6MRandom-v0 environment, it is possible simply to run:

from myosuite.utils import gym
env = gym.make('myoElbowPose1D6MRandom-v0')
env.reset()
for _ in range(1000):
  env.mj_render()
  env.step(env.action_space.sample()) # take a random action
env.close()

You can find our tutorials on the general features and the ICRA2023 Colab Tutorial ICRA2024 Colab Tutorial on how to load MyoSuite models/tasks, train them, and visualize their outcome. Also, you can find baselines to test some pre-trained policies.

License

MyoSuite is licensed under the Apache License.

Citation

If you find this repository useful in your research, please consider giving a star ⭐ and cite our arXiv paper by using the following BibTeX entrys.

@Misc{MyoSuite2022,
  author =       {Vittorio, Caggiano AND Huawei, Wang AND Guillaume, Durandau AND Massimo, Sartori AND Vikash, Kumar},
  title =        {MyoSuite -- A contact-rich simulation suite for musculoskeletal motor control},
  publisher = {arXiv},
  year = {2022},
  howpublished = {\url{https://github.com/myohub/myosuite}},
  doi = {10.48550/ARXIV.2205.13600},
  url = {https://arxiv.org/abs/2205.13600},
}