Probabilistic Robotics

File: 2of3.png (873KB, 955x606px) Image search: [Google]

873KB, 955x606px

[math] bel_{\mathrm out}[bel_{\mathrm in}](x) := N^*W_z(x)\int_A K_u(x,x')\,bel_{\mathrm in}(x'){\mathrm d} [/math]

File: 3of3.png (551KB, 974x426px) Image search: [Google]

551KB, 974x426px

p(x|y)

bump
ill take a closer look at the exercises this weekend

File: KrisPime.jpg (53KB, 677x960px) Image search: [Google]

53KB, 677x960px

In the line of doing it in baby steps, I'm gonna do an intermezzo today and read this chapter
"THE BAYESIAN ALGORITHM"
(from a known book on Econometrics)

https://www.blackwellpublishing.com/content/BPL_Images/Content_store/Sample_chapter/9781405117197/Lancaster_sample%20chapter_Intro%20to%20modern%20Bayesian%20Econometrics.pdf

If anybody wants to join, we can discuss later

Is Bayes rule a mathematical theorem or more of a principle we want to be true about a theory of probabilities?

>>8443270
its a proven theorem to calculate the probability of x given y
so it doesnt give you a definite answer but rather a probability distribution

>>8440526
I think you'd be better off looking for people on IRC. On sci you don't get notified for threads or replies or anything and it's not a place people really stick around.

>>8444125
I've never seen an IRC channel that reasonably sticks to the topic AND has most posts be at least somewhat worth reading. People there aren't keeping their horses before typing random stuff that would make that work.

It's not the format and maybe not even the user base. The situation would just be different if the web were like a class, where people felt any incentive to do some 10+ minutes calculations themselves.
A reading group on StackExchange (where people are further down the educational road) would get just as many flakes

File: the book cover 009.png (3MB, 1352x1528px) Image search: [Google]

3MB, 1352x1528px

https://youtu.be/WWSXHw_Pb5g

<The tangent on Bayes algorithm>

Is the Frequentist interpretation of probability theory simply more restrictive than the Bayesian one, or is there more to it?

>>8444603
no

holy shit this is JUST probability, not your fedora called "probabilistic robotics"

if you literally can't pass a community college first year course in probability (which is what that question is) you need to commit to some sudoku puzzles

File: people_walking_darker-copy.jpg (210KB, 1800x674px) Image search: [Google]

210KB, 1800x674px

>>8445772
It's the first exercise in a 600 page book, chill and contribute

File: IMG_20161030_130837836.jpg (2MB, 3264x1836px) Image search: [Google]

2MB, 3264x1836px

>>8442187
I see that Chapter 1 text, on p.13, asks you to generate random numbers and do basic plotting of regression data.

What are the goto C++ libraries for random number generation, distributions, regression and then plotting?
(Also, you may bump with some pigs)

>>8445887
>c++
boost

>>8445887
Yes.
#include<programthemyourselfyoulazybastard.h>

>>8445891
Thx, will use

>>8445895
Yes, I will.
You don't seem to get the point og this. It's not about havong the code, and getting results from other anons, but about having done it.

>>8445887
dlib
http://dlib.net/

>>8440526
DESU from the stuff I've read in the past, probabilistic robotics/"autonomous agents" etc seem more like a fun way to learn statistics and mess with multi-dimentional data and high-level applied math rather than a useful robotics branch.

You have to hand-type all environment interactions and the control mechanisms, which is really boring and non-generalizeable.

ANN is where it's at tbqh famlad.

File: IMG_20161018_232812133.jpg (1MB, 3264x1836px) Image search: [Google]

1MB, 3264x1836px

>>8446014
Will go into it after the basic graphing stuff, thanks

>>8446022
I was motivated by learning the Kalman filter and the book was recommended.
What's ANN?

Artificial Neural Networks? An advisor of me (doing Chemical Kinetics experiements now) did this some 20 years ago and I'm interested in learning more of it.
(However, I'm at a quest to pull out some \sci\ participation and for that I'd have to dig into something more basic I think. On the other hamd I could try to go into category theory and clarify some basics to make the threads about it here more accessible)

>>8446094
Yeah I was refering to artificial neural networks.

They are resurfacing in all fields, in my uni we also had a few such papers published in the past years by professors. They're about ANN applications in combustion kinetics, predicting reaction species in complex systems etc.

It's a really interesting tool/field but it's out of my reach for now (in terms of available time and mathematical background).

Regarding /sci/ participation, you probably shouldn't get your hopes up.
/sci/ is slow as it is already, the chance of finding people with the appropriate background and enough interest is pretty slim.

If you want real participation you should post here in parallel with some other specialized forum.

File: the book cover 010.png (3MB, 1352x1528px) Image search: [Google]

3MB, 1352x1528px

https://youtu.be/v6H4x7gjBxw

File: the book cover 011.png (3MB, 1352x1528px) Image search: [Google]

3MB, 1352x1528px

https://youtu.be/x7oAByOWyIg

File: the book cover 012.png (3MB, 1352x1528px) Image search: [Google]

3MB, 1352x1528px

https://youtu.be/27_tlcWQIh0

File: 150430_a.jpg (52KB, 640x1136px) Image search: [Google]

52KB, 640x1136px

>>8450789
bump is an unsexy bump. Go at exercise 2 :)

Exercise 1:
The probability that it is broken given n measurements is:

1/(1+99/3^n)

You get there by setting up a decisiontree and observing that the absolute probability of it being broken is 0.01 and the probability of getting n measurements under 1m is 0.99/3^n.

Normalising gives the above answer.
Am I correct?

File: Screenshot_20161102-121459.png (190KB, 1280x720px) Image search: [Google]

190KB, 1280x720px

>>8450922
Yes! I got there schematically using Bayes theorem in pic related, pt. 11.

p_n = 1 / (1 + (1/p_init - 1) * r^n)

That's the same logic.
I didn't discuss exercise 2 or any beyond it, and there are several programming tasks open from thw weeks that we cooked up

>>8450934

I have not read any of the book but I took a look at your summary page. You wrote "the book does classical mechanics but with probability density functions". Do you mean classical mechanics in the sense of hamiltonians and lagrangians?

File: Path_Integral_Quantum_Mechanics_Picture.jpg (20KB, 400x233px) Image search: [Google]

20KB, 400x233px

>>8451143
No, I meant that the math feels like as if you do rigid body mechanics, and they keep it on the algorithmic side of it.

In any case, since only 2 people here read along in the book, I think I'll focus more on the programming stuff now - there's more people responding to this sort of thing

File: IMG_20161103_111835.jpg (1MB, 1836x3264px) Image search: [Google]

1MB, 1836x3264px

Pretty sure I'm gonna buy the Stroustroup book (not the translation, thought)
Then again, it's pretty thick and feels more like half a reference.

Any criticism or, bettee, alternative recommendation?

File: Screen Shot 2016-11-03 at 14.33.46.png (127KB, 486x162px) Image search: [Google]

127KB, 486x162px

>>8452663
update to that post:
I found this stackoverflow consensus
http://stackoverflow.com/questions/388242/the-definitive-c-book-guide-and-list

the "Accelerated C++" book may be more straight forward to go through (and to be found as pdf):
https://www.amazon.com/dp/020170353X/?tag=stackoverfl08-20

Is someone interested going there as a project (more than 2 people, that is)?