Environments and simulators for Learning Algorithms

Collection of environments, simulators and competitions for training & benchmarking Reinforcement Learning and AI algorithms.

Collections of environments

Gym. Collection of classic environments for benchmarking RL, such as Atari, MuJoco, etc (OpenAI).

• Link: https://github.com/openai/gym
• Gym-compatible

Gym Universe. Huge collection of various environments for benchmarking RL (OpenAI).

• Link: https://github.com/openai/universe
• Gym-compatible

ALE. Arcade Learning Environment with Atari games (Marc Bellemare).

• Link: https://github.com/mgbellemare/Arcade-Learning-Environment

Pycolab. Customized Grid-World env (DeepMind).

• Link: https://github.com/deepmind/pycolab

Vehicle Simulation

Carla (Intel, Toyota).

• Link: https://github.com/carla-simulator/carla

AirSim. Realistic autonomous vehicle simulator (Microsoft).

• Link: https://github.com/Microsoft/AirSim

Navigation

MineRL. Minecraft Gym-friendly RL environment along with human player dataset for imitation learning (CMU).

• Link: https://minerl.io/

Deepmind Lab. 3D Navigation in Labyrinths (Deepmind).

• Link: https://github.com/deepmind/lab

VizDoom. 3D Shooting and Navigation Doom game.

• Link: https://github.com/mwydmuch/ViZDoom
• Competition (May): http://vizdoom.cs.put.edu.pl/

Project Malmo. 3D Navigation and Quest Solving in Minecraft game (Microsoft).

• Link: https://github.com/Microsoft/malmo
• Competition (April): https://malmo-leaderboard.azurewebsites.net/

AI2Thor. Home indoor 3D Navigation.

• Link: https://github.com/allenai/ai2thor
• Competition (June): https://competitions.codalab.org/competitions/16929#learn_the_details

HoME Platform. Home indoor 3D Navigation. Based on SUNCG dataset.

• Link: https://github.com/HoME-Platform/home-platform
• Gym-compatible

MINOS. Home indoor 3D Navigation. Based on SUNCG and Matterport3D datasets (Intel).

• Link: https://github.com/minosworld/minos
• Gym-compatible

House3D. Home indoor 3D Navigation & Visual Question Answering. Based on SUNCG dataset (Facebook).

• Link: https://github.com/facebookresearch/House3D

GibsonEnv. Home indoor 3D Navigation & Locomotion. Based on Gibson, SUNCG, Stanford 2D3DS and Matterport 3D datasets (Stanford).

• Link: https://github.com/StanfordVL/GibsonEnv

Gym-Maze. 2D navigation in customizable mazes.

• Link: https://github.com/zuoxingdong/gym-maze
• Gym-compatible

Strategies

PySC2. Starcraft II strategy learning environment (Deepmind, Blizzard).

• Link: https://github.com/deepmind/pysc2

TorchCraft. Starcraft I strategy learning environment (Facebook).

• Link: https://github.com/TorchCraft/TorchCraft/
• Competitions:
  • SC AI Competition: https://www.cs.mun.ca/~dchurchill/starcraftaicomp/
  • Student SC AI Tournament: https://sscaitournament.com

Locomotion

Roboschool. Locomotion, replicates proprietary MoJoCo environments with additional improvements; OpenAI

• Link: https://github.com/openai/roboschool
• Gym-compatible

Control Suite. Set of locomotion environments based on MuJoCo physics engine (DeepMind).

• Link: https://github.com/deepmind/dm_control
• arXiv: https://arxiv.org/abs/1801.00690

Multi-Agent RL

PommerMan. Multi-Agent (up to 4 players) "Bomberman"-like game.

• Link: https://github.com/MultiAgentLearning/playground
• Competition: https://www.pommerman.com/competitions