STRATEGO EXPERT SYSTEM SHELL Casper Treijtel and Leon Rothkrantz Faculty of Information Technology and Systems Delft University of Technology Mekelweg 4 2628 CD Delft University of Technology E-mail: L.J.M.Rothkrantz@cs.tudelft.nl KEYWORDS Games, A.I., Multi-agent, Expert systems, Stratego ABSTRACT The field of multi-agent systems is an active area of research. One of the possible applications of a multi-agent system is the use of distributed techniques for problem solving. Instead of approaching the problem from a central point of view, a multi-agent system can impose a new mode of reasoning by breaking the problem down in a totally different way. In this paper we investigate a distributed approach to playing Stratego. Computational agents that each have their own field of perception, evaluation and behavior represent the individual pieces of the Stratego army. A first prototype of a framework has been developed that consists of a simulation environment for the agents and an implementation of the agent’s evaluation function. The agents have a rule engine that generates behavior that is a resultant of the environment in which they live. INTRODUCTION This paper describes a first attempt to play the Stratego game with multiple agents. The Stratego game is a board game where two players battle each other with their armies of pieces. The object of the game is to capture the enemy flag, by moving pieces towards the enemy and trying to capture the enemy pieces. An interesting property of the game is that the information the players have is incomplete, because the identity of the opponent's pieces is concealed until exposed by battles between pieces. Our motivations for using the multiple agent approach are as follows. When we consider a human society from a central point of view we see that it is a very complex system. A possible attempt to understand the complex behavior of a human society is to consider it as a system that is made up of individuals that have their own characteristics, behavior patterns and interactions with each other. It is the sum of all the local actions and interactions that constitutes the overall behavior of the society. This investigation is an attempt to support this hypothesis by considering the Stratego game. Specifically we want to investigate whether a distributed way of playing this game will provide us with a means to break down the complexity of playing it. Our work is based on ideas of multiple agents as described by J. Ferber (Ferber 1999) and intelligent agents as developed by P.Maes (Maes 1995) and L.Steels (Steels 1997). DESIGN In designing the agents we want to make use of the fact that each piece in the Stratego army has a certain dedicated role. These roles originate from their specific ranks and the rules of the Stratego game. All pieces have secondary goals as well of which possibly the most important one is to stay alive. We propose to define some degrees of freedom in our model of the agent that will allow us to experiment with different types of agents in the Stratego army. Specifically we define for each agent: • The agent's perception range. Depending on the agent's role in the army the perception will be a diamond of range one to five, or an n x n square of fields. Important pieces will have wider perceptions. • The agent's ‘reactive’ behavior. For every agent we define four elementary behaviors that are executed following a reaction in various situations. These behaviors are attack, flee, random walk, and stay and do nothing. • The agent's ‘cognitive’ abilities, for example evaluate situation, compute optimal next move, form hypotheses, and make plans. In our design emotion is modeled as follows. Emotions are related to parameter settings regarding the agent's perception and behavior. For example, if an agent gets upset, afraid or stressed we shrink his field of perception (tunnel view). And if the agent is angry we increase the possibility to attack (McCauley 1998; Scheutz 2000). We designed two levels of communication among agents. One is communication by means of a blackboard that can be written to and read from by every agent. The blackboard is a container of all information of the board situation that is available. This way all agents can rely on the fact that their field of perception is in accordance with the current board- situation. The blackboard contains strictly information about the board status. Additionally the agents can use an asynchronous message-passing structure. Agents can send and receive messages to each other containing information about the Stratego battlefield. The communication structure allows sending messages to all other agents, sending messages to agents of a certain rank or sending messages to specific agents. The content of messages can either be known facts, hypotheses or requests.