Ghost direction detection and other innovations for Ms. Pac-Man

Pac-Man was developed in the 1980s, becoming one of the most popular arcade games of its time. It still has a significant following today and has recently attracted the attention of artificial intelligence researchers, in part, due to the fact that the agent must react in real time in order to navigate its way through the maze. This paper forms an entry to the 2010 IEEE Conference on Computational Intelligence and Games Ms. Pac-Man competition, where the objective is to reach the highest score possible without any human intervention. That is, the Pac-Man is under the control of a computer algorithm which must make real time decisions and control the way the Pac-Man moves. Our Pac-Man algorithm includes detecting the location of the in game objects in relation to the map and creating a grid based graph to represent the game state. Our Pac-Man follows a number of rules, utilising Dijkstra´s algorithm and a tree search algorithm. A further contribution of this paper is effective detection of the ghosts´ direction, which we believe has not been done before in the way that we propose. The world record, for the Ms. Pac-Man competition, is currently help by ICE Pambush 3 with a score of 30,010 (achieved at the 2009 IEEE Congress on Computational Intelligence and Games). Our algorithm consistently achieves a score of over 20,000. The highest score we have recorded is 30,930. We hope that we can replicate these achievements at the 2010 conference where we will present our entry.