A robot that recognises a doorway still has several problems to solve before it can pass through it. It needs an estimate of its own position, a feasible route and a controller that keeps the actual machine close to that route. Recognition is one component of autonomy, not a substitute for the rest. That relationship is the useful organising idea behind Principles of Robot Autonomy, a freely accessible online textbook highlighted by Stanford on 22 October.[1][3]
The book grows out of a course begun in 2017 and brings together Daniele Gammelli, Joseph Lorenzetti, Katie Luo, Gioele Zardini and Marco Pavone. Its value as a resource is the connected curriculum: readers can move between the physical system, its observations and the decisions made from those observations rather than treating each as an isolated AI topic.[1][2]
Three useful routes into the material
The 19-chapter contents begin with dynamics, trajectory optimisation, control and motion planning. Later sections cover cameras and perception, localisation and filtering, sensor fusion, sequential decision-making, reinforcement learning and imitation learning. Deep learning therefore appears inside an autonomy stack rather than replacing the stack’s other elements.[3]
Choose the question before choosing the chapter
| Reader’s question | Starting point | Useful exercise |
|---|---|---|
| Why does a robot miss its planned path? | Dynamics and closed-loop tracking | PID control or linear MPCSource 3Source 4 |
| Where is the robot, given noisy observations? | Localisation and filtering | Discrete Bayes or approximate filteringSource 3Source 4 |
| How does a policy choose an action? | Sequential decisions and reinforcement learning | Reinforcement-learning notebooksSource 3Source 4 |
The supporting collection lists 40 notebooks across 15 chapters, with links to Colab and source files. Examples include cart-pole control, trajectory tracking, camera calibration, object detection and localisation. We inspected the resource listings, not the execution of every notebook, so this is a guide to the available materials rather than a compatibility or reproducibility certification.[4]
A useful way to study with such examples is to change one assumption at a time. In an illustrative localisation exercise, a learner could compare a reliable observation with a noisier one, then examine how the position estimate changes. The educational question is not simply whether the final answer looks reasonable. It is whether the learner can explain which information the estimator trusted and why.
The AI tutor is optional
The website also offers chapter-aware Ask the Book assistance. It distinguishes the text from the probabilistic assistant and warns that generated answers can be wrong. Using the assistant requires a provider API or a suitable local setup; free access to the book does not imply that every external model call or computing resource is free.[2]
That separation is sensible for a technical subject. A tutor can help formulate a question or walk through an explanation, but a plausible account of a control equation is not the same thing as a correct derivation. Readers can return to the authored text, inspect the relevant notebook and compare the consequences of a changed parameter. The three formats have different jobs.
For someone evaluating a robotics demonstration, the same curriculum offers a more demanding set of questions. Which part is learned from data? Which part is an explicit model? How is uncertainty represented? What happens when the robot is not where the perception system thinks it is? These are editorial questions suggested by the book’s structure, not claims that the book independently validates a particular robot.
The most useful result would be a reader who can follow an error across the whole system: from an imperfect observation to a mistaken position estimate, then to an unsuitable command. That is a stronger foundation for understanding physical AI than knowing only the name of the latest model.
Sources & further reading
- Stanford robotics textbook and curriculum Stanford University · 22 Oct 2026 · Accessed 27 Oct 2026
- Principles of Robot Autonomy Principles of Robot Autonomy authors · Accessed 27 Oct 2026
- The Book: chapter index Principles of Robot Autonomy authors · Accessed 27 Oct 2026
- Exercises and Supporting Code Principles of Robot Autonomy authors · Accessed 27 Oct 2026