DeepMind Control Suite

TL;DR

The DeepMind Control Suite provides a standardized, easy-to-use set of continuous control tasks with interpretable rewards, serving as benchmarks for evaluating reinforcement learning algorithms.

Contribution

It introduces a comprehensive, accessible benchmark suite for continuous control tasks using MuJoCo, facilitating consistent evaluation of RL agents.

Findings

Benchmark results for several learning algorithms

Ease of use and modification demonstrated

Publicly available for community use

Abstract

The DeepMind Control Suite is a set of continuous control tasks with a standardised structure and interpretable rewards, intended to serve as performance benchmarks for reinforcement learning agents. The tasks are written in Python and powered by the MuJoCo physics engine, making them easy to use and modify. We include benchmarks for several learning algorithms. The Control Suite is publicly available at https://www.github.com/deepmind/dm_control . A video summary of all tasks is available at http://youtu.be/rAai4QzcYbs .