The Design and Implementation of a Testbed for Comparative Game AI Studies

Abstract

An essential component of realism in video games is
the behavior exhibited by the non-player character (NPC) agents
in the game. Most development efforts employ a single artificial
intelligence (AI) method to determine NPC agent behavior during
gameplay. This paper describes an NPC AI testbed under
development which will allow for a variety of AI methods to
be compared under simulated gameplay conditions. Two squads
of NPC agents are pitted against each other in a game scenario.
Multiple games using the starting same AI assignments will form
an epoch. The testbed allows for the testing of a variety of AI
methods in three dimensions. Individual agents can be assigned
different AI methods. Individual agents can use different AI
methods at different times during the game. And finally, the
AI used by one type of agent can be made to differ from the
AI used by another agent type. Extensive data is collected for
all agent actions in all games played in an epoch. This data will
form the basis of the comparative analysis.