An environment centered around the original [Doom](https://en.wikipedia.org/wiki/Doom_(1993_video_game)) game, focusing on visual control (from image to actions) at thousands of frames per second. ViZDoom supports depth and automatic annotation/labels buffers, as well as accessing the sound. The Gym wrappers provide easy-to-use access to the example scenarios that come with ViZDoom. Since 2016, the [ViZDoom paper](https://arxiv.org/abs/1605.02097) has been cited more than 600 times.
### [ flappy-bird-gym: A Flappy Bird environment for OpenAI Gym](https://github.com/Talendar/flappy-bird-gym)
A simple environment for single-agent reinforcement learning algorithms on a clone of [Flappy Bird](https://en.wikipedia.org/wiki/Flappy_Bird), the hugely popular arcade-style mobile game. Both state and pixel observation environments are available.
This is a 3v3 MOBA environment where you train creatures to fight each other. It runs entirely on the GPU so you can easily have hundreds of instances running in parallel. There are around 15 items for the creatures, 60 "senses", 5 actions, and roughly 23 tweakable rewards. It's also possible to benchmark an agent against other agents online. It's available for free for training for personal use, and otherwise costs money; see licensing details on the website
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.
### [ SlimeVolleyGym: A simple environment for single and multi-agent reinforcement learning](https://github.com/hardmaru/slimevolleygym)
A simple environment for benchmarking single and multi-agent reinforcement learning algorithms on a clone of Slime Volleyball game. The only dependencies are gym and NumPy. Both state and pixel observation environments are available. The motivation of this environment is to easily enable trained agents to play against each other, and also facilitate the training of agents directly in a multi-agent setting, thus adding an extra dimension for evaluating an agent's performance.
Supported fork of gym-retro with additional games, states, scenarios, etc. Open to PRs of additional games, features, and platforms since gym-retro is no longer maintained
GymFC is a modular framework for synthesizing neuro-flight controllers. The architecture integrates digital twinning concepts to provide a seamless transfer of trained policies to hardware. The OpenAI environment has been used to generate policies for the world's first open-source neural network flight control firmware [Neuroflight](https://github.com/wil3/neuroflight).
An environment for simulating the classical optimal control problem where the thrust of a vertically ascending rocket shall be determined such that it reaches the maximum possible altitude while being subject to varying aerodynamic drag, gravity, and mass.
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.
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.
A simple environment using [PyBullet](https://github.com/bulletphysics/bullet3) to simulate the dynamics of a [Bitcraze Crazyflie 2.x](https://www.bitcraze.io/documentation/hardware/crazyflie_2_1/crazyflie_2_1-datasheet.pdf) nanoquadrotor.
Mars Explorer is an openai-gym compatible environment designed and developed as an initial endeavor to bridge the gap between powerful Deep Reinforcement Learning methodologies and the problem of exploration/coverage of an unknown terrain.
3D physics environments like the Mujoco environments but uses the Bullet physics engine and do not require a commercial license. Works on Mac/Linux/Windows.
gym-carla provides a gym wrapper for the [CARLA simulator](http://carla.org/), which is a realistic 3D simulator for autonomous driving research. The environment includes a virtual city with several surrounding vehicles running around. Multiple sources of observations are provided for the ego vehicle, such as front-view camera image, lidar point cloud image, and bird-eye view semantic mask. Several applications have been developed based on this wrapper, such as deep reinforcement learning for end-to-end autonomous driving.
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.
An environment for behavioral 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 behaviors are uncertain. Several scenes are proposed, such as highway, merge, intersection and roundabout.
An environment for the stochastic longitudinal control of an electric vehicle. It is intended to be a descriptive and comprehensible example of a continuous real-world problem within the field of autonomous driving.
A Gym for solving motion planning problems for various traffic scenarios compatible with [CommonRoad benchmarks](https://commonroad.in.tum.de/scenarios), which provides configurable rewards, action spaces, and observation spaces.
A set of environments from control tasks: Acrobot, CartPole, and LunarLander with various types of anomalies injected into them. It could be very useful to study the behavior and robustness of a policy.
### [ CARL](https://github.com/automl/CARL)
Configurable reinforcement learning environments for testing generalization, e.g. CartPole with variable pole lengths or Brax robots with different ground frictions.
Environments for hyperparameter configuration using RL. Includes cheap surrogate benchmarks as well as real-world algorithms from e.g. AI Planning, Evolutionary Computation and Deep Learning.
The Gridworld package provides grid-based environments to help simulate the results for model-based reinforcement learning algorithms. The initial release supports single agent system only. Some features in this version of the software have become obsolete. New features are being added to the software like windygrid environment.
An environment that implements a typical [multi-armed bandit scenario](https://en.wikipedia.org/wiki/Multi-armed_bandit) where an [ad server](https://en.wikipedia.org/wiki/Ad_serving) must select the best advertisement to be displayed in a web page. Some example agents included: Random, epsilon-Greedy, Softmax, and UCB1.
AnyTrading is a collection of OpenAI Gym environments for reinforcement learning-based trading algorithms with a great focus on simplicity, flexibility, and comprehensiveness.
This package describes an OpenAI Gym interface for creating a simulation environment of reinforcement learning-based recommender systems (RL-RecSys). The design strives for simple and flexible APIs to support novel research.
2D Transportation Puzzles. The environment consists of transportation puzzles in which the player's goal is to push all boxes to 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 overfitting to predefined levels.
In our paper "A Reinforcement Learning Environment for Mathematical Reasoning via Program Synthesis" we convert the DeepMind Mathematics Dataset into an RL environment based on program synthesis.https://arxiv.org/abs/2107.07373
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](https://arxiv.org/abs/1808.05584). 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](https://github.com/gomerudo/meta-dataset)).
[NLPGym](https://arxiv.org/pdf/2011.08272v1.pdf) provides interactive environments for standard NLP tasks such as sequence tagging, question answering, and sequence classification. Users can easily customize the tasks with their datasets, observations, features and reward functions.
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.
Musculoskeletal Models in OpenSim. 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.
### [ PGE: Parallel Game Engine](https://github.com/222464/PGE)
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.
This is a list of environments for quantum architecture search following the description in [Quantum Architecture Search via Deep Reinforcement Learning](https://arxiv.org/abs/2104.07715). The agent designs the quantum circuit by taking actions in the environment. Each action corresponds to a gate applied on some wires. The goal is to build a circuit U that generates the target n-qubit quantum state that belongs to the environment and is hidden from the agent. The circuits are built using [Google QuantumAI Cirq](https://quantumai.google/cirq).
PyBullet based CartPole and Quadrotor environments—with [CasADi](https://web.casadi.org) (symbolic) *a priori* dynamics and constraints—for learning-based control and model-based reinforcement learning.
ABIDES (Agent-Based Interactive Discrete Event Simulator) is a message-based multi-agent discrete event-based simulator. It enables simulation of complex multi-agent systems for different domains. ABIDES has already supported work in [equity markets simulation](https://arxiv.org/abs/1904.12066) and [federated learning](https://dl.acm.org/doi/abs/10.1145/3383455.3422562).
[ABIDES-Gym](https://arxiv.org/abs/2110.14771) (ACM-ICAIF21 publication) is a new wrapper built around ABIDES that enables using ABIDES simulator as an Open AI Gym environment for the training of Reinforcement Learning algorithms.
We apply for this work by specifically using the market's extension of ABIDES/ABIDES-Markets and developing two benchmark financial market Gym environments for training daily investor and execution agents. As a result, these two environments describe classic financial problems with a complex interactive market behavior response to the experimental agent's action.
Unofficial implementation of the Go-Explore algorithm presented in [First return then explore](https://arxiv.org/abs/2004.12919) based on [stable-baselines3](https://github.com/DLR-RM/stable-baselines3).
Gym-TORCS is the reinforcement learning (RL) environment in TORCS domain with OpenAI-gym-like interface. TORCS is the open-source realistic car racing simulator recently used as an RL benchmark task in several AI studies.
Python implementation of Elastica, open-source software for the simulation of assemblies of slender, one-dimensional structures using Cosserat Rod theory.
This is the repository accompanying the paper [OstrichRL: A Musculoskeletal Ostrich Simulation to Study Bio-mechanical Locomotion](https://arxiv.org/abs/2112.06061).
