Has anyone ever tried to implement a Learning Classifier System like LCS, XCS or maybe even XCSF on Arduino?
Maybe using XCSlib?
I can't find anything.
My robot should learn to move through the room without colliding with the wall.
I can't find anything.
I'm not familiar with those algorithms, but my guess would be that most Arduino chips don't have enough code, and especially RAM, space to implement them. You have only 2KB-8KB of RAM to play with, in general. Maybe doable using the Due or other boards with more powerful chips. Or maybe better, raspberry Pi or Beagle board.
Thank you for your ideas.
The limitations are not a problem because the above named algorithms don't need much RAM. An minimal implementation of an lcs has by guess 200 lines of code and doesn't need more than an array of 100 strings's with a size of max. 100 characters. Should be possible for an arduino.
The challange is to connect the interfaces of the algorithms and the arduino and to find a good fitnessfunction that allows the algorithm to learn. I hoped to find an example how someone other did this. But no search result for "xcs lcs on arduino" 
Resource requirements sound fairly moderate. But 100 strings of 100 chars each is 10KB, and still a little more than any std Arduino can handle, assuming you need to change them. Looks like you're the favored man to do the project, :-). If you can find "any" source code written in C, then should be translatable to Arduino.