This repository is a curated, categorized list of reinforcement learning (RL) environments, serving as a comprehensive resource for researchers and practitioners. It aims to simplify the discovery and selection of suitable environments for various RL tasks, from robotics and games to autonomous driving and text-based challenges.

How It Works

The list is organized into broad categories such as Robotics, Games, Multi-Task Learning, Navigation, and Autonomous Driving, with sub-categories for specific applications. Each entry provides a brief description of the environment's capabilities, supported tasks, and key features, often highlighting integrations with popular RL frameworks or simulators like OpenAI Gym, Unity, and Gazebo.

Quick Start & Requirements

This is a curated list, not a runnable software package. To use any of the listed environments, refer to their individual project pages for installation and usage instructions.

Highlighted Details

• Extensive coverage across diverse domains including robotics (e.g., Assistive-gym, Meta-World, RLBench), games (e.g., Gym Retro, VizDoom, StarCraft II), and specialized areas like autonomous driving (e.g., CARLA, DeepGTAV) and text-based games (e.g., Jericho, TextWorld).
• Includes environments focused on generalization (e.g., Procgen, DMControl Generalization Benchmark) and safety (e.g., DeepMind AI Safety Gridworlds, Safety Gym).
• Features environments with complex, long-horizon tasks and realistic physics simulations (e.g., IKEA Furniture Assembly, RAISIM, Holodeck).
• Provides links to related collections like "Environment Zoo" and "Awesome Deep RL" for further exploration.

Maintenance & Community

Maintained by Andrew Szot and Youngwoon Lee. The project encourages community contributions for missing or miscategorized environments via GitHub issues or pull requests.

Licensing & Compatibility

The licensing varies per environment listed. Users must consult the individual project licenses for usage terms, compatibility, and restrictions.

Limitations & Caveats

This is a reference list; it does not provide a unified API or installation method. Users must individually install and configure each environment, which may involve complex dependencies or specific hardware requirements.