Browse Books

Reviewer: Goran Trajkovski

The extremely appealing title of this book is misleading: it indicates only that it is different from most other robotics textbooks in its infusion of Bayesian and Markovian concepts into most of the standard algorithms that robotics has used in the past decade or two. In fact, this is an average textbook in robotics for undergraduates, or for remedial courses for graduate students interested in robotics research. It outlines traditional artificial intelligence (AI)-based robotics, in which the robotic agent is more or less spoon-fed information about the environment, and there is too much interference from the designer (meta-agent or us, the designers of the experiments). Except for the discrepancies between the exact intention of the algorithm, and, say, movements that are not executed accurately in the real world, in this volume there is not much more space to infuse probabilistic concepts, let alone further infuse possibilities and necessities, plausibilities, belief in general, or fuzzy measures. It is true that some of these terms are in the book, such as "a priori" and "a posteriori" beliefs, but their use is miles away from the Dempster-Shaffer definitions. Apart from the paradigmatic direction the authors take, this textbook is a good one, offering a broad overview of the traditionally valued results in the field of robotics. The book has 647 pages, arranged into 17 chapters, structured into four parts. The first part, "Basics," consists of six chapters that introduce the engineering tools necessary to follow the rest of the text. These chapters focus on recursive state estimation, Gaussian and nonparametric filters, and robot motion and perception. The book's second part, "Localization," contains two chapters focusing on Markov and Gaussian, and grid and Monte Carlo approaches to this topic. The third part, "Mapping," emphasizes algorithms for occupancy mapping and simultaneous localization and mapping (SLAM). The final part, "Planning and Control," studies the Markov decision processes in depth. Students should read, or at least browse though, this volume, as it provides a good perspective on how we present and compare results in traditional robotics, and offers knowledge of the benchmark problems that were used in the past. Novel approaches in the post-AI sciences (cognitive science, neuroscience, and so on, which are not covered in this textbook) do indeed make problems of classical robotics easier to solve, or even obsolete. This textbook outlines the efforts that brought us there. As a bonus, this is all done with a painfully correct and mathematically detailed engineering approach. Online Computing Reviews Service