Skip to content

MyoSuite is a collection of environments/tasks to be solved by musculoskeletal models simulated with the MuJoCo physics engine and wrapped in the OpenAI gym API.

License

Notifications You must be signed in to change notification settings

MyoHub/myosuite

Repository files navigation

Support Ukraine PyPI Documentation Status PyPI - License PRs Welcome Downloads Open In Colab Slack Twitter Follow

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.

TasksALL

Installations

You will need Python 3.8 or later versions.

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

conda create --name myosuite python=3.8
conda activate myosuite

It is possible to install MyoSuite with:

pip install -U myosuite

for advanced installation, see here.

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

python -m myosuite.tests.test_myo

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

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 python -m myosuite_init), run:

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: Open In Colab

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 Open In Colab ICRA2024 Colab Tutorial Open In Colab 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}},
  year =         {2022}
  doi = {10.48550/ARXIV.2205.13600},
  url = {https://arxiv.org/abs/2205.13600},
}