Abstract

Hunt the Wumpus is a simple computer game dating from c.1973. It is set in a cave of interconnected rooms, one of which contains the wumpus. The object of the game is to kill the wumpus by navigating around the cave and shooting an arrow into the correct room. The player receives messages to help in this task, such as "you smell the wumpus nearby", or "you feel a draught from a nearby pit". Earlier games of this "hide-and-seek" genre had simply used a cartesian grid as the play area, but Hunt the Wumpus used a topological map, based on the vertices of a dodecahedron.

The simplicity of the game, and its ingenuity for the time, provides an interesting basis for a model. The "player" agent has access to a limited range of observables (e.g. "Can I smell a wumpus? Is there a bat nearby?") from which facts can be derived (e.g. "There is no wumpus in room 12"). We can construe the player as having a restricted set of operations in the world - at any point, the player can either move into, or shoot into, an adjacent room.

The paper discusses the use of this model in testing some of the different strategies a player could employ in the game, and how they might be hampered by unexpected things happening that are not represented in the game's rules, such as the unexpected movement of the wumpus to a different location, or the impromptu reconfiguration of the cave's network. In so doing, the application of Empirical Modelling to the simulation and exploration of a strategy game will be demonstrated.

I intend to discuss means of investigating the effectiveness of different strategies given different rules, and the robustness of these strategies if something unexpected occurs. Therefore, this is partly a study in Artificial Intelligence. The experimental approach I intend to take with the model, however, suggests that it is also relevant to the area of Educational Technology, as the user of the model is encouraged to learn about the game and how to play it well by experimenting with strategies.