Designing a Computer Program for Ms. Pac-man Competition using Neural Network

Abstract

This project is to design a computer program for Ms. Pac-man competition held by IEEE World Congress on Computational Intelligence (WCCI 2010). The computer program is aimed to control Ms. Pac-man to play the game and get a score as high as possible. In this project, Neural Network, instead of rule-based hard coding, is used to implement the controller. The program collects information from the game played by the given strategies (hard-corded or human player). A training set is designed based on this information. The Neural Network is then trained by this training set. Finally, the controller program loads the trained neural network and interfaces the status of the game as the inputs of it. The program reads the output of the neural network and decides the next move of Ms. Pac-man. The result shows a little improvement after the training of the neural network. It is also observed that the behaviour of Ms. Pac-Man controlled by the resultant controller has several patterns.