Application of Q-learning algorithm to maze solving

Building

You will need ant and java.

$ git clone https://github.com/PFGimenez/QLearning.git
$ cd QLearning
$ ant

You can verify it works by running an example:

$ java QLearning --file maze3.txt

Usage

java QLearning [-g gamma] [-i nbIter] [-s sleep] [--no-arrows] [--no-backtrack] [-G [16|32]] [-S strategy [param strategy]] --file maze[1|2|3].txt
java QLearning [-g gamma] [-i nbIter] [-s sleep] [--no-arrows] [--no-backtrack] [-G [16|32]] [-S strategy [param strategy]] --gen sizeX sizeY nbEntry nbExit nbTrap nbPortal

Strategies:
	FullRandomExplo: random walk
	FullSmartExplo: explore the maze (not completely at random)
	FullExploit: full exploitation
	Epsilon: epsilon-strategy with FullSmartExplo and FullExploit. Mandatory parameter : epsilon (recom : 0.99)

Screenshot

Sprites designed by Nevanda and DragonDePlatino for the game NetHack.