This is a list of Gym environments, including those packaged with Gym, official OpenAI environments, and third party environment.
For information on creating your own environment, see Creating your own Environment.
The code for each environment group is housed in its own subdirectory gym/envs. The specification of each task is in gym/envs/__init__.py. It's worth browsing through both.
These are a variety of algorithmic tasks, such as learning to copy a sequence.
import gym
env = gym.make('Copy-v0')
env.reset()
env.render()The Atari environments are a variety of Atari video games. If you didn't
do the full install, you can install dependencies via pip install -e '.[atari]' (you'll need cmake installed) and then get started as
follows:
import gym
env = gym.make('SpaceInvaders-v0')
env.reset()
env.render()This will install atari-py, which automatically compiles the Arcade
Learning Environment. This
can take quite a while (a few minutes on a decent laptop), so just be
prepared.
Box2d is a 2D physics engine. You can install it via pip install -e '.[box2d]' and then get started as follows:
import gym
env = gym.make('LunarLander-v2')
env.reset()
env.render()These are a variety of classic control tasks, which would appear in a
typical reinforcement learning textbook. If you didn't do the full
install, you will need to run pip install -e '.[classic_control]' to
enable rendering. You can get started with them via:
import gym
env = gym.make('CartPole-v0')
env.reset()
env.render()MuJoCo is a physics engine which can do very
detailed efficient simulations with contacts. It's not open-source, so
you'll have to follow the instructions in
mujoco-py
to set it up. You'll have to also run pip install -e '.[mujoco]' if
you didn't do the full install.
import gym
env = gym.make('Humanoid-v2')
env.reset()
env.render()These environments also use MuJoCo. You'll have to also run pip install -e '.[robotics]' if
you didn't do the full install.
import gym
env = gym.make('HandManipulateBlock-v0')
env.reset()
env.render()You can also find additional details in the accompanying technical report and blog post. If you use these environments, you can cite them as follows:
@misc{1802.09464,
Author = {Matthias Plappert and Marcin Andrychowicz and Alex Ray and Bob McGrew and Bowen Baker and Glenn Powell and Jonas Schneider and Josh Tobin and Maciek Chociej and Peter Welinder and Vikash Kumar and Wojciech Zaremba},
Title = {Multi-Goal Reinforcement Learning: Challenging Robotics Environments and Request for Research},
Year = {2018},
Eprint = {arXiv:1802.09464},
}
Toy environments which are text-based. There's no extra dependency to install, so to get started, you can just do:
import gym
env = gym.make('FrozenLake-v0')
env.reset()
env.render()16 simple-to-use procedurally-generated gym environments which provide a direct measure of how quickly a reinforcement learning agent learns generalizable skills. The environments run at high speed (thousands of steps per second) on a single core.
Learn more here: https://github.com/openai/procgen
Gym Retro lets you turn classic video games into Gym environments for reinforcement learning and comes with integrations for ~1000 games. It uses various emulators that support the Libretro API, making it fairly easy to add new emulators.
Learn more here: https://github.com/openai/retro
We recommend using the PyBullet Robotics Environments instead
3D physics environments like Mujoco environments but uses the Bullet physics engine and does not require a commercial license.
Learn more here: https://github.com/openai/roboschool
The gym comes prepackaged with many many environments. It's this common API around many environments that makes Gym so great. Here we will list additional environments that do not come prepacked with the gym. Submit another to this list via a pull-request.
3D physics environments like the Mujoco environments but uses the Bullet physics engine and does not require a commercial license. Works on Mac/Linux/Windows.
Learn more here: https://docs.google.com/document/d/10sXEhzFRSnvFcl3XxNGhnD4N2SedqwdAvK3dsihxVUA/edit#heading=h.wz5to0x8kqmr
3D procedurally generated tower where you have to climb to the highest level possible
Learn more here: https://github.com/Unity-Technologies/obstacle-tower-env
Platforms: Windows, Mac, Linux
PGE is a FOSS 3D engine for AI simulations, and can interoperate with the Gym. Contains environments with modern 3D graphics, and uses Bullet for physics.
Learn more here: https://github.com/222464/PGE
gym-inventory is a single agent domain featuring discrete state and action spaces that an AI agent might encounter in inventory control problems.
Learn more here: https://github.com/paulhendricks/gym-inventory
gym-gazebo presents an extension of the initial OpenAI gym for robotics using ROS and Gazebo, an advanced 3D modeling and rendering tool.
Learn more here: https://github.com/erlerobot/gym-gazebo/
A simple 2D maze environment where an agent finds its way from the start position to the goal.
Learn more here: https://github.com/tuzzer/gym-maze/
A human musculoskeletal model and a physics-based simulation environment where you can synthesize physically and physiologically accurate motion. One of the environments built in this framework is a competition environment for a NIPS 2017 challenge.
Learn more here: https://github.com/stanfordnmbl/osim-rl
A minimalistic gridworld environment. Seeks to minimize software dependencies, be easy to extend and deliver good performance for faster training.
Learn more here: https://github.com/maximecb/gym-minigrid
MiniWorld is a minimalistic 3D interior environment simulator for reinforcement learning & robotics research. It can be used to simulate environments with rooms, doors, hallways and various objects (eg: office and home environments, mazes). MiniWorld can be seen as an alternative to VizDoom or DMLab. It is written 100% in Python and designed to be easily modified or extended.
Learn more here: https://github.com/maximecb/gym-miniworld
The environment consists of transportation puzzles in which the player's goal is to push all boxes on the warehouse's storage locations. The advantage of the environment is that it generates a new random level every time it is initialized or reset, which prevents over fitting to predefined levels.
Learn more here: https://github.com/mpSchrader/gym-sokoban
A lane-following simulator built for the Duckietown project (small-scale self-driving car course).
Learn more here: https://github.com/duckietown/gym-duckietown
GymFC is a modular framework for synthesizing neuro-flight controllers. The architecture integrates digital twinning concepts to provide seamless transfer of trained policies to hardware. The OpenAI environment has been used to generate policies for the worlds first open source neural network flight control firmware Neuroflight.
Learn more here: https://github.com/wil3/gymfc/
AnyTrading is a collection of OpenAI Gym environments for reinforcement learning-based trading algorithms with a great focus on simplicity, flexibility, and comprehensiveness.
Learn more here: https://github.com/AminHP/gym-anytrading
An implementation of the board game Go
Learn more here: https://github.com/aigagror/GymGo
An environment for simulating a wide variety of electric drives taking into account different types of electric motors and converters. Control schemes can be continuous, yielding a voltage duty cycle, or discrete, determining converter switching states directly.
Learn more here: https://github.com/upb-lea/gym-electric-motor
The environment is fully-compatible with the OpenAI baselines and exposes a NAS environment following the Neural Structure Code of BlockQNN: Efficient Block-wise Neural Network Architecture Generation. Under this setting, a Neural Network (i.e. the state for the reinforcement learning agent) is modeled as a list of NSCs, an action is the addition of a layer to the network, and the reward is the accuracy after the early-stop training. The datasets considered so far are the CIFAR-10 dataset (available by default) and the meta-dataset (has to be manually downloaded as specified in this repository).
Learn more here: https://github.com/gomerudo/nas-env
gym-jiminy presents an extension of the initial OpenAI gym for robotics using Jiminy, an extremely fast and light weight simulator for poly-articulated systems using Pinocchio for physics evaluation and Meshcat for web-based 3D rendering.
Learn more here: https://github.com/Wandercraft/jiminy
An environment for behavioural planning in autonomous driving, with an emphasis on high-level perception and decision rather than low-level sensing and control. The difficulty of the task lies in understanding the social interactions with other drivers, whose behaviours are uncertain. Several scenes are proposed, such as highway, merge, intersection and roundabout.
Learn more here: https://github.com/eleurent/highway-env
gym-carla provides a gym wrapper for the CARLA simulator, which is a realistic 3D simulator for autonomous driving research. The environment includes a virtual city with several surrounding vehicles running around. Multiple source of observations are provided for the ego vehicle, such as front-view camera image, lidar point cloud image, and birdeye view semantic mask. Several applications have been developed based on this wrapper, such as deep reinforcement learning for end-to-end autonomous driving.
Learn more here: https://github.com/cjy1992/gym-carla
The OpenModelica Microgrid Gym (OMG) package is a software toolbox for the simulation and control optimization of microgrids based on energy conversion by power electronic converters.
Learn more here: https://github.com/upb-lea/openmodelica-microgrid-gym