Reinforcement Learning: An Introduction, 2nd Edition – Sutton & Barto (Hardcover)

Reinforcement Learning: An Introduction, Second Edition, by Richard S. Sutton and Andrew G. Barto is the hardcover MIT Press textbook that defines the field, covering the core ideas and algorithms of reinforcement learning from bandits and dynamic programming through function approximation and policy gradients.

Almost every reinforcement learning course, paper and tutorial cites "Sutton and Barto." The book explains how an agent learns to maximize reward by interacting with an uncertain environment, and it does so with clear prose and the harder mathematics set off in shaded boxes so the main thread stays readable. Part of the Adaptive Computation and Machine Learning series.

What's included

• Reinforcement Learning: An Introduction, Second Edition, complete text
• Hardcover binding
• Publisher: MIT Press
• ISBN-13: 9780262039246
• Part I: tabular solution methods, including multi-armed bandits, dynamic programming, Monte Carlo and temporal-difference learning
• Part II: approximate solution methods with neural networks, off-policy learning and policy-gradient methods
• Part III: connections to psychology and neuroscience, and case studies including AlphaGo

Who it's for

Undergraduate and graduate students in a reinforcement learning or machine learning course. Software engineers and data scientists moving into RL for robotics, recommendation or game AI. Researchers who want the standard reference on the shelf, and self-learners following the many online courses built around this book.

Why this edition

The second edition was substantially expanded over the 1998 original. It added algorithms such as UCB, Expected Sarsa and Double Learning, new material on neural networks and the Fourier basis for function approximation, expanded coverage of off-policy and policy-gradient methods, and entirely new chapters on psychology, neuroscience and modern case studies. The closing chapter discusses the societal impact of reinforcement learning.

Frequently asked questions

Is Sutton and Barto still the best introduction to reinforcement learning?

It remains the standard text. Newer deep RL books build on it, and most courses assign it as the foundation.

How much math do I need to read this book?

Comfort with probability, basic linear algebra and calculus is enough for the main text. More advanced derivations are boxed so they can be skipped on a first pass.

Is this the hardcover MIT Press edition?

Yes. This listing is the MIT Press hardcover, ISBN 9780262039246.

Details