Repository files navigation

Google DeepMind's software stack for physics-based simulation and ReinforcementLearning environments, using MuJoCo physics.

Anintroductory tutorial for this package is available as a Colaboratorynotebook:

This package consists of the following "core" components:

• dm_control.mujoco: Libraries that provide Python bindings to the MuJoCophysics engine.

• dm_control.suite: A set of Python Reinforcement Learning environmentspowered by the MuJoCo physics engine.

• dm_control.viewer: An interactive environment viewer.

Additionally, the following components are available for the creation of morecomplex control tasks:

• dm_control.mjcf: A library for composing and modifying MuJoCo MJCFmodels in Python.

• dm_control.composer: A library for defining rich RL environments fromreusable, self-contained components.

• dm_control.locomotion: Additional libraries for custom tasks.

• dm_control.locomotion.soccer: Multi-agent soccer tasks.

If you use this package, please cite our accompanyingpublication:

@article{tunyasuvunakool2020,         title = {dm_control: Software and tasks for continuous control},         journal = {Software Impacts},         volume = {6},         pages = {100022},         year = {2020},         issn = {2665-9638},         doi = {https://doi.org/10.1016/j.simpa.2020.100022},         url = {https://www.sciencedirect.com/science/article/pii/S2665963820300099},         author = {Saran Tunyasuvunakool and Alistair Muldal and Yotam Doron and                   Siqi Liu and Steven Bohez and Josh Merel and Tom Erez and                   Timothy Lillicrap and Nicolas Heess and Yuval Tassa},}

Installdm_control from PyPI by running

pip install dm_control

Note:dm_control cannot be installed in "editable" mode (i.e.pip install -e).

Whiledm_control has been largely updated to use the pybind11-based bindingsprovided via themujoco package, at this time it still relies on some legacycomponents that are automatically generated from MuJoCo header files in a waythat is incompatible with editable mode. Attempting to installdm_control ineditable mode will result in import errors like:

ImportError: cannot import name 'constants' from partially initialized module 'dm_control.mujoco.wrapper.mjbindings' ...

The solution is topip uninstall dm_control and then reinstall it withoutthe-e flag.

Starting from version 1.0.0, we adopt semantic versioning.

Prior to version 1.0.0, thedm_control Python package was versioned0.0.N,whereN was an internal revision number that increased by an arbitrary amountat every single Git commit.

If you want to install an unreleased version ofdm_control directly from ourrepository, you can do so by runningpip install git+https://github.com/google-deepmind/dm_control.git.

The MuJoCo Python bindings support three different OpenGL rendering backends:EGL (headless, hardware-accelerated), GLFW (windowed, hardware-accelerated), andOSMesa (purely software-based). At least one of these three backends must beavailable in order render throughdm_control.

• Hardware rendering with a windowing system is supported via GLFW and GLEW.On Linux these can be installed using your distribution's package manager.For example, on Debian and Ubuntu, this can be done by runningsudo apt-get install libglfw3 libglew2.0. Please note that:

  • dm_control.viewer can only be used with GLFW.
  • GLFW will not work on headless machines.

• "Headless" hardware rendering (i.e. without a windowing system such as X11)requiresEXT_platform_device support in the EGL driver. Recent Nvidiadrivers support this. You will also need GLEW. On Debian and Ubuntu, thiscan be installed viasudo apt-get install libglew2.0.

• Software rendering requires GLX and OSMesa. On Debian and Ubuntu these canbe installed usingsudo apt-get install libgl1-mesa-glx libosmesa6.

By default,dm_control will attempt to use GLFW first, then EGL, then OSMesa.You can also specify a particular backend to use by setting theMUJOCO_GL=environment variable to"glfw","egl", or"osmesa", respectively. Whenrendering with EGL, you can also specify which GPU to use for rendering bysetting the environment variableMUJOCO_EGL_DEVICE_ID= to the target GPU ID.

1. The above instructions usingpip should work, provided that you use aPython interpreter that is installed by Homebrew (rather than thesystem-default one).

2. Before running, theDYLD_LIBRARY_PATH environment variable needs to beupdated with the path to the GLFW library. This can be done by runningexport DYLD_LIBRARY_PATH=$(brew --prefix)/lib:$DYLD_LIBRARY_PATH.