google-deepmind/dm_control
DeepMind's physics sandbox for training control agents
It wraps MuJoCo physics into ready-made reinforcement learning environments, from simple control tasks to multi-agent soccer.
Not currently ranked — collecting fresh signals.
star history
What it does
dm_control is DeepMind’s Python stack for physics-based reinforcement learning built on the MuJoCo engine. It ships with Python bindings, a standard set of continuous control environments, an interactive viewer, and libraries for composing custom robots and tasks. There is even a multi-agent soccer environment if you want to train robotic midfielders.
The interesting bit
The stack treats rendering as a deployment detail rather than an afterthought, letting you pick between GLFW for local windows, EGL for headless GPU servers, or OSMesa for pure software rendering. The composer and mjcf libraries also let you build complex environments and modify robot models in Python rather than hand-editing MuJoCo XML files.
Key highlights
- Ready-made RL environments (
suite), Python MuJoCo bindings, and an interactive viewer mjcfcomposes and modifies MuJoCo models programmatically instead of editing raw XMLcomposerassembles rich environments from reusable, self-contained componentslocomotionlibraries and a multi-agent soccer task for advanced experiments- Three rendering backends: windowed GLFW, headless hardware-accelerated EGL, and software OSMesa
Caveats
- Editable installs are explicitly broken due to legacy components auto-generated from MuJoCo headers
- The interactive viewer is tied to GLFW and will not work on headless machines
- Pre-1.0.0 versioning was an arbitrary internal revision number that increased with every commit
Verdict
A staple for anyone doing continuous control research in Python. Everyone else can safely admire it from a distance.
Frequently asked
- What is google-deepmind/dm_control?
- It wraps MuJoCo physics into ready-made reinforcement learning environments, from simple control tasks to multi-agent soccer.
- Is dm_control open source?
- Yes — google-deepmind/dm_control is open source, released under the Apache-2.0 license.
- What language is dm_control written in?
- google-deepmind/dm_control is primarily written in Python.
- How popular is dm_control?
- google-deepmind/dm_control has 4.6k stars on GitHub.
- Where can I find dm_control?
- google-deepmind/dm_control is on GitHub at https://github.com/google-deepmind/dm_control.