This repository provides experimental reinforcement learning algorithms and tools for the Stable-Baselines3 library, targeting researchers and practitioners who need access to cutting-edge or niche RL implementations. It extends Stable-Baselines3's core functionality by offering a curated collection of less mature but potentially valuable algorithms and utilities, maintaining the library's characteristic simplicity and documentation standards.

How It Works

SB3-Contrib acts as a supplementary package, housing implementations that may not meet the strict integration requirements of the main Stable-Baselines3 library. This approach allows for the inclusion of a wider range of algorithms, such as Augmented Random Search (ARS), Quantile Regression DQN (QR-DQN), MaskablePPO, RecurrentPPO, Truncated Quantile Critics (TQC), TRPO, and CrossQ, as well as utility wrappers like the Time Feature Wrapper. The goal is to offer these beyond the core library's scope while adhering to Stable-Baselines3's quality standards for code style and documentation.

Quick Start & Requirements

• Install with pip: pip install sb3-contrib
• Recommended: Install master versions of Stable-Baselines3 (pip install git+https://github.com/DLR-RM/stable-baselines3) and SB3-Contrib (pip install git+https://github.com/Stable-Baselines-Team/stable-baselines3-contrib).
• Documentation: https://sb3-contrib.readthedocs.io/

Highlighted Details

• Implements advanced RL algorithms like MaskablePPO, RecurrentPPO, and QR-DQN.
• Includes experimental algorithms and utility wrappers not found in the main library.
• Maintains Stable-Baselines3's standards for code style and documentation.
• Facilitates contributions for niche or experimental RL research.

Maintenance & Community

The project is maintained by the Stable-Baselines-Team. Contribution guidelines are available in CONTRIBUTING.md.

Licensing & Compatibility

The project is likely licensed under the MIT License, consistent with Stable-Baselines3, allowing for commercial use and integration with closed-source projects.

Limitations & Caveats

As an experimental package, implementations may be less mature or subject to change compared to the core Stable-Baselines3 library. Users should consult the documentation for specific algorithm readiness and potential breaking changes.