RL Handbook
An openly available handbook covering reinforcement learning from multi-armed bandits through modern policy-gradient and actor-critic methods, with runnable code examples for each chapter.
RL Handbook is a comprehensive, continuously updated guide to reinforcement learning and sequential decision making, starting from multi-armed bandits and Markov decision processes and progressing through value-based methods, such as dynamic programming, Monte Carlo and temporal-difference prediction, Sarsa, Q-learning, and DQN and its improvements, on-policy policy-gradient methods, such as REINFORCE, A2C/A3C, TRPO, and PPO, and off-policy policy-based methods, such as DDPG, TD3, and SAC, with model-based methods and further advanced topics planned.
Each chapter lives in its own folder holding its source, a runnable code example, and its figures, and is meant to serve students, researchers, and engineers with a resource that balances mathematical rigor against runnable code. The site is built with Next.js and a documentation framework, and welcomes contributions such as fixing typos, improving explanations, adding references, or extending examples.