This repository provides Python code, PDFs, and curated resources for a series of blog posts on Reinforcement Learning (RL). It's aimed at RL practitioners and researchers looking for practical implementations and theoretical explanations of core RL concepts, algorithms, and applications. The project offers a structured learning path from Markov Decision Processes to Actor-Critic methods and function approximation.

How It Works

The project is organized into distinct posts, each covering specific RL topics with accompanying code and documentation. The core approach involves implementing foundational RL algorithms and environments in standalone Python files, avoiding external dependencies like OpenAI Gym for its custom environments. This design choice simplifies setup and allows for direct integration into projects. Key algorithms covered include Value and Policy Iteration, Monte Carlo methods, TD learning (SARSA, Q-Learning), Actor-Critic methods, and Evolutionary Algorithms.

Quick Start & Requirements

• Install: No specific installation is required beyond Python.
• Prerequisites: Numpy is the primary dependency. Matplotlib is needed for visualization and animations in some examples.
• Environments: Custom, standalone RL environments (Grid World, Multi-Armed Bandit, Inverted Pendulum, Mountain Car, Drone Landing) are provided as Python files that can be directly imported.
• Resources: Links to external software (DeepMind Lab, OpenAI Gym), books, articles, and other learning materials are included in the README.

Highlighted Details

• Comprehensive coverage of RL topics across eight blog posts.
• Standalone, self-contained RL environments mimicking OpenAI Gym conventions.
• Includes practical examples for various RL applications like drone landing and inverted pendulum control.
• Curated list of essential RL resources, including seminal papers and online tools.

Maintenance & Community

The repository is maintained by Massimiliano Patacchiola. Contributions are welcomed via pull requests to update the resource list.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

The code is presented as part of a blog series and may not represent the most up-to-date or production-ready implementations. Some advanced topics or environments might be simplified for pedagogical purposes.