Reinforcement Learning and Optimal Control

Widely considered the foundational text of the field, this book provides the essential theoretical framework that complements Bertsekas's work. It offers a more accessible entry point while maintaining the same rigorous focus on algorithmic design and optimal control principles.

As a companion to his other works, this two-volume set delves deeper into the mathematical foundations of the problems addressed in his reinforcement learning text. It is perfect for readers who appreciate Bertsekas's specific pedagogical style and desire a more comprehensive grasp of the underlying optimization theory.

This book bridges the gap between abstract reinforcement learning theory and practical, real-world application in robotics. Fans of Bertsekas will appreciate the rigorous mathematical treatment of uncertainty and the systematic approach to solving complex control problems.

While it focuses on neural networks rather than control theory, this text is the definitive resource for the deep learning techniques that power modern reinforcement learning. It shares the same high-level academic tone and exhaustive, structured approach to complex technical subjects.

This concise volume provides a theoretical analysis of reinforcement learning algorithms, focusing on convergence and sample complexity. It appeals to the same audience that values the formal, proof-heavy style found in Bertsekas's writing.

For readers interested in the classical control theory roots of modern reinforcement learning, this classic text offers an unparalleled deep dive into linear quadratic systems. It matches the technical depth and historical context often touched upon in Bertsekas's broader discussions.

This book provides a rigorous, modern treatment of the multi-armed bandit problem, a fundamental sub-problem of reinforcement learning. It shares the same commitment to precision and mathematical elegance that defines Bertsekas's pedagogical approach.

This seminal work is the direct predecessor to modern deep reinforcement learning, laying the groundwork for approximating optimal control. Readers familiar with Bertsekas's style will find this an essential historical and technical bridge to current methodologies.

This text provides the optimization backbone necessary to understand the linear programming approaches often used in dynamic programming. It is ideal for readers who want to master the mathematical tools that Bertsekas frequently employs across his entire bibliography.

This is the definitive reference for the theory of Markov Decision Processes, which forms the mathematical core of reinforcement learning. It offers a comprehensive and formal treatment that will satisfy readers looking for the same level of rigor found in Bertsekas's work.