Customer Reviews

Probabilistic Robotics (Intelligent Robotics and Autonomous Agents series)

5 star:		(8)
4 star:		(3)
3 star:		(3)
2 star:		(0)
1 star:		(0)

 

 

The authors took 6 years to write this book. And it shows. This is a mindblowing tour through the algorithms used at the cutting edge of Robotics.

What is good

1. Every algorithm has descriptive text, mathematical derivations AND pseudo code. More importantly it all meshes into a cohesive whole.

2. The progression of chapters is excellent, starting with basic algorithms and proceeding to more advanced/refined algorithms.

3.There is a consistent practical focus with algorithms being explained in the context of solving real world problems in robotics.

4. The exercises are few in number , but are *perfect* to illuminate each chapter's ideas and encourage the reader to start thinking on his own.

5. There is a comprehensive errata page on the book's website.

6. Last but not least, the tone of the writing is very engaging. The reader is not talked down to. It is almost as if the authors were in your study carefully guiding you through an intellectual wonderland.

The bad.
Hmmm i can't think of anything. It is great book. I just wish the authors would write MORE books like this :-)



About the only caveat is that a reader should have *some* degree of mathematical insight before attempting this book. The authors do cover elementary probability theory etc in the initial chapters, and they do a good job given the space constraints. But in my opinion if you have absolutely no experience in probability theory or calculus, you should probably learn from other books and then tackle this one. This is, after all, a graduate level text.

This is really an amazing book - it more than fulfilled my expectations.
It starts from the very basics of probability theory and clearly derives
Kalman Filtering, Particle Filtering, Probabilistic Motion and Probabilistic
Perception in the first 6 chapters. From there it moves on to talk about
Localization and Mapping completely separately (which I appreciated, since
the two topics are far easier to comprehend independently) in chapters 7 and
8 and then finally introduces SLAM (the main topic of the book) in chapter
9. From there it goes on to discuss various SLAM algorithms and implementations,
and finally rounds out with planning and control (that is, the practical
application of SLAM algorithms).

I can't imagine a more well-researched academic work. Every point is backed
up with examples and illustrations, and every algorithm is derived rigorously.
Even better, the mathematical derivations are set apart from the main text
so that a more "casual" reader can skip over the derivations and still get
some benefit from the text (and believe me, the math parts of this book are
very involved!). The authors assume a working knowledge of trigonometry,
calculus and linear algebra (although you could likely make some sense of the
book even if you're rusty in any of these areas). However, since the book
is about probability, you'll probably need some background in probability
theory to get any value from this text. Chapter 2 contains a refresher on
probability theory, but I doubt it would be enough to decipher the later
chapters if you had no background in the subject. I found myself having to
go back and look up the details of Bayes Rule and multivariate conditional
probability more than once.

My only gripe with this book is that each chapter includes suggested exercises
(good) but no answers/cross-check (bad). Especially considering the open-ended
nature of the exercises, it's almost not worth attempting them (or even reading
them), since you'll never know if you got the right answer, or were even on the
right track. There's no "student supplement" (at least not as I write this),
so the exercises are fairly pointless.

However, that aside, this is one of the best academic books I've read in a very
long time. I had been struggling through academic papers from IEEE and ACM on
the topic of SLAM, and only comprehending about half of it before I picked up
"Probabilistic Robotics". After reading this book carefully (I actually had
to read it twice to get it all to sink in), I'm actually zipping through the
academic papers, and understanding everything I read. You couldn't ask for a
better introduction to probabilistic robotics and SLAM.

Robotics is a vast field. At the center of its computational side are the algorithms for spatial reasoning: mapping, localizing, and navigating. This book covers these fundamentals more thoroughly and comprehensively than any other. The algorithms described here are already the de facto starting point for cutting-edge work in computational robotics. Variants on these ideas make up a huge part of ongoing research, as evidenced by current journal articles and conference papers.

This is not an introductory text. There are many excellent choices for that kind of broad coverage of robotics, even computational robotics. Rather, this book is to robotics what Vision: A Modern Approach is to the field of computer vision. Even at the speed the field is moving, this book will be a standard for many years.

The book presents what is currently the frontier of probabilistic research in robotics. This is explained as a means of a robot coping with inadequate information from its perceptive inputs. The intent is to embed more robust control logic within the robot. Rather than having human programmers try to code for every contingency.

There are many algorithms in the text. Each is explicitly defined in pseudocode. But just as significantly, each is accompanied by extensive textual explanations and derivations. These are rounded out by the chapters having exercises that extend the ideas developed in each chapter.

Many ideas from statistics are applied here, from Markov processes to Monte Carlo samplings to Bayesian inferences.

Uncertainty is an important issue facing intelligent systems.
Thrun, Burgard, and Fox have made important contributions to
this area of research. Probabilistic Robotics is a more narrowly focused text than the title might suggest. At 650 pages perhaps it could not be broader and yet do justice to the topics the authors want to cover. Perhaps the title should have been Probabilistic Robot Navigation. My other criticism would be the lack of executables

This book has plenty of math and theory in regards to state estimation and SLAM. However, it is really lacking in details and examples of implementation. Many of the problems at the end of the chapters rely on material from the next chapters. Most of what I learned about Kalman and Extended Kalman filters in the past made more sense as I read this book. However, new material such as particle filters was difficult to understand from the text alone. I had to look elsewhere for examples of implementation/tutorials.

Think of a situation where you had an extremely good lecturer for some subject at uni. The lecturer explained everything very nicely using a host of slides, examples etc. Now, if after the lectures you go back and look at the slides again, you'll probably understand what the slides mean and remember what the lecturer said with regard to each slide. But, as is typical in any course, the slides doesn't contain all the information you learnt during the course. The explanations from the lecturer were critical in understanding the slides.

Now, think of the situation where someone who didn't attend the course, try to learn the subject, just by reading through the slides. This is generally extremely difficult. especially because the slides themselves doesn't contain all the information required to comprehend them.

This book is exactly that. It's a set of slides from a good lecture, but without the important other bits of information required to properly comprehend the ideas presented with it. I feel that the authors assume that the readers already know what they are talking about, and goes on skipping over the details required to properly grasp those concepts. If you've already learnt the topics in the book from by some other means, then you'll be able to understand what the authors are trying to say, and would end up ranking this book as an excellent book. But, my question is, if you already knew those things, why would you read this book in the first place.

BUT, it's not a worthless book. Few of the sections are explained in a very intuitive manner. Also, it's a good book for those who already know the concepts, but want to put them altogether in a coherent manner, with respect to robotics. Also, it is a good source for those who are looking for a list of state-of-the-art algorithms in mobile robotics.

This textbook is the standard reference for probabilistic robotics in the areas of navigation and mapping. One of the authors is the director of the Stanford AI lab and headed the winning entry in the DARPA Grand Challenge in 2007, which needless to say means he understands and has developed many of the techniques in the book. The algorithms are laid out and explained at different depths of understanding, which sometimes allows them to be used without reading the rigorous mathematical derivations that are included. Within the first week of having this book, I found that my method of estimating odometry in the prediction step of a Kalman filter could be improved with a different estimation. In addition, since the book provided a mathematical derivation, I could compare the two techniques and explain under what assumptions my approximation fails to do well.

I consider this book the most valuable resource in the field! If you are really interested in implementing kalman filter localization, particle filter localization or SLAM algorithms, this book really will help you. This book was my reference during my Master Thesis and the algorithms are so comprenhensive that I hadn't any problem to put them running.
I think the autors made a really good effort to explain complex mathematical concepts as clearly as possible. Great Job!

This is by far the best resource that I have found for collating a large number of internally consistent SLAM algorithms into a single volume. The book carefully leads the reader through the requirements of SLAM presenting one algorithm at a time, building upon the algorithms presented previously. This approach lends itself very well to develop-while-you-read. If you care to do so, I recommend reading it through once in its entirety and then starting over for the develop-while-you-read approach. The once through does a good job of presenting the big picture and giving you the opportunity to decide which primary SLAM path you prefer; Kalman and particle filtering are the two main approaches discussed. I'm currently implementing FastSLAM with particle filtering and have not run into any large hurdles using this book to lead the way.

The only major challenge that I've encountered is that it assumes a very good understanding of probability distributions. A good college statistics book makes a good companion for this read.

I also read Thrun's FastSLAM monograph. There's very little new information in that monograph which Probabilisitc Robotics doesn't already cover. After reading PR, Google becomes your best resource for finding the latest algorithms and code samples. Because even with the descriptive pseudo code algorithms, a perfect follow-up to this book would be "Probabilistic Robotics Implemented" with lots of code samples.