I will highly recommend reading this textbook along with RM Murray's "A Mathematical Introduction to Robotic Manipulation" [1]. Both of these textbooks provide very intuitive and clear cut explaination of principles behing robotic motion and planning.
I would also highly recommend completing a 5-course specialization on Coursera [2] by the same name as the first textbook. The course is conducted by the author, he uses plenty of great visualization to explain the concepts clearly and the course follows the textbook.
Whilst this is interesting, this isn't where the field is going. This is useful if, say, you want to make a CNC robot with extreme rigidity.
Replacing forward and inverse kinematics with machine learning is extremely simple, and avoids most of the problems these formulas have (for instance, they don't consider that a robot might attempt to move through itself). And while grasping and manipulation isn't simple even with machine learning, I think you'll agree that the theory in these books also very, very much isn't simple.
And the problem is that the current capabilities of robots using these formulas from books like these ... are insufficient. They need to become WAY more complex to be useful, but that's impossible: they're already damned complicated.
The outlook very much is that the only way we'll ever arrive at the required complexity for these machines is through machine learning, and if not that, then through some form of automated program writing. A human just cannot write safe and working code that deals with robots in open settings and/or where interactions might occur.
You should probably still learn the very basic parts. Chapter 9 and 11. But after that, you will likely get further with machine learning.
You can go a long long way with what's in this book. Think of factory robots, medical robotics, cruise control, aircraft components, etc. Just because it can't handle a robot octopus doesn't mean it's pointless.
> Replacing forward and inverse kinematics with machine learning is extremely simple
Could you point towards a good paper or review paper on using machine learning models to learn forward and inverse kinematics? How far can one get with plain old LSTMs?
The topology part of these textbooks is the hardest for me. Understanding the 2d representations of the shapes. Where can one get a more bottom up explanation of the crucial topology concepts? Also bad at geometry in general
Control or spatial-related DSP has a lot of geometry going on and it's really hard to get with a good working knowledge. I don't know if there's a short path. Start with a good linear algebra course maybe in parallel to actually trying to solve or hack though the type of problems you want to solve, and thinking about how the basic stuff you are learning complements with the high level one.
Besides the courses recommended here this are other online materials related to planning, location and control I found really useful and consider to be high quality:
I took a linear algebra course. Maybe I need to restudy but I would rather have links to exact chapters than reading the whole thing, as I don't see a direct path from the linear algebra I studied to the topology concepts and geometry here. Specifically there weren't any curves in linear algebra. Ex 4.10 and 4.11 here: http://planning.cs.uiuc.edu/node143.html
Are examples of things that are hard for me to understand. Especially that 2d picture in 4.11
You have to study quite a lot of linear algebra before you can tackle functional analysis, the subject which includes the study of topological vector spaces. I don't know anything about robotics, however, so I don't know how functional analysis would apply, if at all.
Functional analysis shows up all over the place. Everywhere you can take a Fourier transform, you can use functional analysis. For robotics, control theory and computer vision both come to mind.
As a start, I recommend reading about Winding Numbers. I've found that to be the most important topology concept for 2d geometry. Computational Geometry is probably deeper than you want to go, but it would answer most of your questions. For curves, I recommend a Numerical Methods reference. Honestly, wander around Wikipedia links starting from topics like winding numbers and spatial partitioning, and you'll probably get a better primer than any book.
Thanks for the link! That seems super interesting! You just made me regret publishing Weekly Robotics couple of hours ago. This link would make for a great addition!
I wouldn't assume that an introductory course would. The concepts in an introductory course will lend a lot to flight vehicles, but flight adds a lot of complications and reduces survivability when failure happens.
I have seen this so often, this top-down approach. And it never arrives at the factory floor.
All those object orientated, industry 4.0 robots, they looks so nice to the professionals on trade-shows, and then there is one example bought- and put to all the other show-off-proto-types in the company-tour room of archievments- and on the factory floor we use the same old robots, for which all the legacy software was written - that nobody can ditch anymore.
And none of those high flying concepts ever talk about those maintaining the robots. Those who never attended High School, who are okay with a 9000 line copy pasted Procedural program, and not okay with object orientation or even functional programing style.
Im so sick of all that wasted time in startups, where people like me fresh from university develop - new, idea robots, who will never take hold in a actual factory- because none of those revoluzzers wants to spend all its time in that factory, maintaining the robot- beeing the only one competent enough todo so.
Sorry if this is a rant, but i wish for once that economic reality would nuke some sense into the architects of that tomorrow that never comes. Maybee we could have small, nice things, if we where willing to do the little steps.
I will highly recommend reading this textbook along with RM Murray's "A Mathematical Introduction to Robotic Manipulation" [1]. Both of these textbooks provide very intuitive and clear cut explaination of principles behing robotic motion and planning.
I would also highly recommend completing a 5-course specialization on Coursera [2] by the same name as the first textbook. The course is conducted by the author, he uses plenty of great visualization to explain the concepts clearly and the course follows the textbook.
1. https://www.cds.caltech.edu/~murray/books/MLS/pdf/mls94-comp...
2. https://www.coursera.org/specializations/modernrobotics