Real-time team-mate AI in games: A definition, survey, & critique

Abstract

Many contemporary games are team-based and there is a growing interest in, and need for, advances in team-mate AI for games. However, although there have been surveys of agent AI in games, to date there has been no survey of work on team-mate AI. Furthermore, the concept of "team-mate AI" is not currently well delineated to distinguish between work on independently-acting agents that happen to be on the same side from work on agents the coordinate their behaviors and decision-making in terms of their teammates behaviors, intentions, and the like. Also, it is important to distinguish between game AI that is used as an optimization technique from real-time game AI, so this paper proposes a definition for real-time team-mate AI (highlighted with examples by game genre), reviews work to date to implement real-time team-mate AI for games in terms of a number of AI research areas (e.g., coordinated action, prediction, learning), and concludes with a brief discussion of significant issues about the state of the art. © 2010 ACM.