MyoSuite-MJX [WIP]

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.

This modified version integrates MJX for efficient execution.

---

Installation

Prerequisites

Ensure you have Python 3.8 or later installed.

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

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

Step 1: Install MuJoCo

Clone and build MuJoCo:

git clone https://github.com/google-deepmind/mujoco.git
cd mujoco
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=../install/ ..
cmake --build .
cmake --install .

Step 2: Build and Install MuJoCo Python Bindings

cd ../python
bash make_sdist.sh
MUJOCO_PATH=/ABS_PATH/install MUJOCO_PLUGIN_PATH=/ABS_PATH/install/mujoco_plugin pip install dist/mujoco-3.2.8.tar.gz

Set the necessary environment variables:

export DYLD_LIBRARY_PATH=/ABS_PATH/install/lib:$DYLD_LIBRARY_PATH

Step 3: Install MJX

cd ../mjx
pip install -e .

Step 4: Install MyoSuite

pip install -e .
pip install jax mediapy

---

Running MyoSuite in MJX

To test MyoHand in the modified environment, run:

python myosuite/mjx/play.py

---

License and Attribution

This repository is based on MyoSuite by Facebook AI and its original authors: Vikash Kumar, Vittorio Caggiano, and others. This modified version integrates MJX and ensures compatibility with the latest MuJoCo versions.

Original citation:

@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},
}

For more details, refer to the original MyoSuite repository.