Open Source Environments

MuJoCo Soccer environment

MuJoCo is a a challenging competitive multi-agent soccer environment with continuous simulated physics. 

View paper  View source on GitHub    View Sample Gameplay

Hanabi Learning Environment

Hanabi is a 2-5 player cooperative game with imperfect information where players must learn to infer the knowledge of their teammates and communicate their own. The challenge is both to learn in self-play and to play with new teammates, including collaborating with human players. 

View paper  View source on GitHub

DeepMind Lab

DeepMind Lab is a 3D customisable game-like platform tailored for agent-based AI research. It is observed from a first-person viewpoint, through the eyes of the simulated agent. 

View blog post    View source on GitHub

PySC2 - StarCraft II Learning Environment

PySC2 is DeepMind's Python component of the StarCraft II Learning Environment (SC2LE). It exposes Blizzard Entertainment's StarCraft II Machine Learning API as a Python RL Environment. This is a collaboration between DeepMind and Blizzard to develop StarCraft II into a rich environment for RL research. PySC2 provides an interface for RL agents to interact with StarCraft 2, getting observations and sending actions.

View blog post    View source on GitHub

DeepMind Control Suite

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. 

View paper   View source on GitHub

AI Safety Gridworlds

This is a suite of reinforcement learning environments illustrating various safety properties of intelligent agents. These environments are implemented in pycolab, a highly-customisable gridworld game engine with some batteries included.

View blog post • View paper  View source on GitHub