The immensely popular video game Pac-Man has challenged players for nearly 30 years, with the very best human competitors striking a highly honed balance between the games two key factors; the 'chomping' of pills (or pac-dots) throughout the level whilst avoiding the ghosts that haunt the maze trying to capture the titular hero. We believe that in order to achieve this it is important for an agent to plan-ahead in creating paths in the maze while utilising a reactive control to escape the clutches of the ghosts. In this paper we evaluate the effectiveness of such a look-ahead against greedy and random behaviours. Results indicate that a competent agent, on par with novice human players can be constructed using a simple framework.
|Title of host publication||IEEE Symposium Computational Intelligence and Games, 2008|
|Subtitle of host publication||CIG '09|
|Place of Publication||Piscataway NJ|
|Number of pages||6|
|Publication status||Published - 1 Dec 2008|
Thompson, T., McMillan, L., Levine, J., & Andrew, A. (2008). An evaluation of the benefits of look-ahead in Pac-Man. In IEEE Symposium Computational Intelligence and Games, 2008: CIG '09 (pp. 310-315). Piscataway NJ: IEEE. https://doi.org/10.1109/CIG.2008.5035655