M S P AC -M AN VERSUS G HOST T EAM CEC 2011 Competition Philipp Rohlfshagen School of Computer Science and Electronic Engineering University of Essex Colchester CO4 3SQ, UK Email: prohlf@essex.ac.uk Simon M. Lucas School of Computer Science and Electronic Engineering University of Essex Colchester CO4 3SQ, UK Email: sml@essex.ac.uk Abstract—Games provide an ideal test bed for computational intelligence and significant progress has been made in recent years, most notably in games such as GO, where the level of play is now competitive with expert human play on smaller boards. Recently, a significantly more complex class of games has received increasing attention: real-time video games. These games pose many new challenges, including strict time constraints, simultaneous moves and open-endedness. Unlike in traditional board games, computational play is generally unable to compete with human players. One driving force in improving the overall performance of arti- ficial intelligence players are game competitions where practition- ers may evaluate and compare their methods against those sub- mitted by others and possibly human players as well. In this pa- per we introduce a new competition based on the popular arcade video game MS PAC-MAN: MS PAC-MAN VERSUS GHOST TEAM. The competition, to be held at the Congress on Evolutionary Computation 2011 for the first time, allows participants to develop controllers for either the Ms Pac-Man agent or for the Ghost Team and unlike previous MS PAC-MAN competitions that relied on screen capture, the players now interface directly with the game engine. In this paper we introduce the competition, including a review of previous work as well as a discussion of several aspects regarding the setting up of the game competition itself. Index Terms—Computational Intelligence, Games, Game Com- petition, Ms Pac-Man, Predator-Prey I. I NTRODUCTION The field of computational intelligence (CI) has had no- ticeable success in recent years in developing computational tools that may compete with human expertise in a variety of domains. One such domain is games, including traditional boardgames such as CHESS or GO and real-time video games such as UNREAL TOURNAMENT. Games pose an interesting challenge, both academically and commercially and have been subject to a long-established research effort. Academically, games provide an ideal test bed for the development and testing of new techniques and technologies: games are defined by an explicit set of rules and the goal of playing a game is usually defined unambiguously by the game’s score or outcome. Games are also immensely flexible and vary greatly in complexity from single player puzzles to two-player boardgames to massively multi-player real- time video games. Furthermore, it is important to note that techniques developed specifically for game playing may often be transferred easily to other domains, greatly enhancing the scope with which such techniques may be used. Monte Carlo Tree Search [9], for example, allowed for a breakthrough in human-competitive play in the classic board game GO; recently, the same technique has been applied successfully to other domains such as scheduling [13]. Undoubtedly, there is significant commercial interest in developing strong game AI as well. The video game software industry in the USA alone is worth an annual turnover of US$ 4.9 billion (2009) with a growth rate of 10.6% for the period 2005-2009 (the growth rate for the US economy as a whole was 1.4% for the same period) [20]. Here, the goal of AI agents is usually not to achieve the strongest possible play but to optimise the overall playability of the game: human players need to be engaged at the right level of difficulty to make the game appealing, a task that has proven difficult as Yannakakis and Hallam [26, p 119] point out: “the increasing number of multi-player online games (among others) is an indication that humans seek more intelligent opponents and richer inter-activity.” A solution to this dilemma is the development of stronger non-player characters (NPCs) that do not rely primar- ily on classical game artificial intelligence (AI) methods such as scripting, triggers and animations. 1 Game com- petitions provide an ideal testbed for practitioners to fur- ther the development of NPCs that play a game intelli- gently and in this paper we introduce a new game com- petition based on the popular arcade game MS PAC-MAN: previous MS PAC-MAN-competitions required participants to develop AI controllers for the Ms Pac-Man charac- ter. The MS PAC-MAN VERSUS GHOST TEAM-competition allows participants for the first time to also develop multi- agent controllers for the ghost team. This paper outlines the scope, rules and technical specifications of this competition. First we introduce the game MS PAC-MAN in section II 1 As Ahlquist and Novak [1, p 4] point out, computer science AI and game AI are only distantly related: the former is about substance, whereas the latter is about appearances. The limitations of game AI have are discussed, for instance, in [23].