dm_control is Google DeepMind's official Python library for physics-based simulation and reinforcement learning (RL) using the MuJoCo physics engine. It provides a comprehensive software stack for building, composing, and benchmarking continuous-control RL environments. The library is structured into several layers: low-level MuJoCo bindings (dm_control.mujoco) with ctypes wrappers and named indexing; a programmatic MJCF model builder (dm_control.mjcf) for constructing and composing MuJoCo XML models; a composer framework (dm_control.composer) with reusable entities, arenas, tasks, observation pipelines, and variation systems for rapid environment prototyping; and a suite of standard benchmark tasks (dm_control.suite) including classic domains like cartpole, cheetah, humanoid, and quadruped. Additionally, it includes specialized locomotion tasks (dm_control.locomotion) with biologically inspired walkers (rat, fruit fly, dog, CMU humanoid), motion-capture tracking tasks, multi-agent soccer, and dexterous manipulation tasks (dm_control.manipulation). The library also provides an interactive viewer (dm_control.viewer) with GLFW/OpenGL rendering, a Blender add-on for exporting 3D models to MuJoCo, and automatic binding generation from MuJoCo C headers.