Welcome to the official webpage for Principles of Robot Autonomy

Why we wrote this book

In less than a decade, autonomous systems have evolved from technologies confined largely to research labs into systems operating in everyday environments. Self-driving cars now share city streets with human drivers, delivery robots navigate sidewalks, drones autonomously inspect infrastructure, and humanoid and quadruped robots are beginning to enter warehouses and homes. The field has crossed an important threshold. Robot autonomy is no longer solely an academic pursuit, but a collection of mature, field-tested methods and tools that practitioners rely on in real-world deployments.

This transition changes what students and engineers need to learn. It is no longer enough to understand an algorithm in isolation; one must understand how it fits into a complete autonomy stack and how it interacts with the other components. We wrote this book to reflect that reality by providing both a principled foundation in the methods that have proven themselves in practice and a systems-level perspective on how those methods work together to build capable autonomous systems.

Moreover, the existing literature largely reflects the historical divisions of the field rather than the integrated nature of modern autonomy stacks. Classical robotics textbooks provide excellent treatments of modeling, control, motion planning, and state estimation, but many were written before deep learning became a central component of perception and decision-making. Conversely, recent books on machine learning and embodied AI cover modern learning techniques in depth, yet often devote comparatively little attention to the dynamics, control theory, and state estimation methods that ultimately determine whether a learned policy can operate reliably on real hardware. Practitioners building autonomous systems cannot afford this separation. A self-driving car depends on both geometric perception and learned object detection, as well as model-based trajectory optimization and learned behavior prediction. Our goal is to present these ideas within a single, coherent framework, rather than forcing readers to assemble them from disparate sources that often employ inconsistent notation, assumptions, and perspectives.

Finally, robot autonomy is fundamentally a practical discipline. Real understanding comes from implementing the algorithms, tuning them, breaking them, and watching how they respond to noisy data and edge cases. For this reason, the book is built around the principle of learning by doing: chapters are accompanied by interactive Python implementations in Jupyter notebooks, with exercises ranging from quick conceptual checks to substantial implementation projects.

We also wanted to lower the barrier to entry for readers coming from adjacent disciplines. A controls engineer interested in reinforcement learning, or an ML researcher wanting to ground their work in physical systems, should be able to jump into the relevant chapters, run the notebooks, and develop practical intuition without assembling a research-grade software stack from scratch. The notebooks are designed to bridge the gap between understanding the mathematics and building autonomous systems that work in practice.