Portal de Periódicos da CAPES

Evolving a Ms. PacMan Controller Using Grammatical Evolution

2010; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-642-12239-2_17

1611-3349

Edgar Galván, John Mark Swafford, Michael O’Neill, Anthony Brabazon,

Reinforcement Learning in Robotics

In this paper we propose an evolutionary approach capable of successfully combining rules to play the popular video game, Ms. Pac-Man. In particular we focus our attention on the benefits of using Grammatical Evolution to combine rules in the form of "if then perform ". We defined a set of high-level functions that we think are necessary to successfully maneuver Ms. Pac-Man through a maze while trying to get the highest possible score. For comparison purposes, we used four Ms. Pac-Man agents, including a hand-coded agent, and tested them against three different ghosts teams. Our approach shows that the evolved controller achieved the highest score among all the other tested controllers, regardless of the ghost team used.