Investigation of reinforcement learning methods

OData support
Engedy István Tamás
Department of Measurement and Information Systems

Nowadays the artificial intelligence plays significant role in people’s lives. Apart from medical, industrial and home environment it also appears in different games. One of its common fields is the reinforcement learning, which provides some kind of intelligence mostly to different games. The reinforcement learning is a learning type, in which the agent moves, acts and observes in a given environment while gathers experience. With this experience it increases its knowledge and improves its behavior, to get along in the given environment the best possible way. The reinforcement learning has several fields and methods. In this thesis I’ve dealt with two of them deeper: with Q-learning and with ADP. I’ve made up my own problem to implement, observe, test and display these algorithms, which is very similar to the known Pacman game. The project takes place in labyrinth-like field. One of the entities in the field is conscious and its goal is to get out of the field, through a goal-field, while avoiding to get caught by the other (non-conscious) entities. The calculations and the evaluation of the results were made in Matlab.


Please sign in to download the files of this thesis.