dm_control: Software and Tasks for Continuous Control

TL;DR

dm_control is a comprehensive Python toolkit offering simulation, task creation, and benchmarking for reinforcement learning in continuous control environments, facilitating research and development in robotics and AI.

Contribution

It introduces a unified software package with standardized tasks, procedural model manipulation, and a suite of benchmarks for reinforcement learning research.

Findings

Provides a set of standardized tasks for benchmarking RL algorithms.

Includes tools for procedural model manipulation and task customization.

Offers high-level abstractions for locomotion and manipulation tasks.

Abstract

The dm_control software package is a collection of Python libraries and task suites for reinforcement learning agents in an articulated-body simulation. A MuJoCo wrapper provides convenient bindings to functions and data structures. The PyMJCF and Composer libraries enable procedural model manipulation and task authoring. The Control Suite is a fixed set of tasks with standardised structure, intended to serve as performance benchmarks. The Locomotion framework provides high-level abstractions and examples of locomotion tasks. A set of configurable manipulation tasks with a robot arm and snap-together bricks is also included. dm_control is publicly available at https://www.github.com/deepmind/dm_control