Available at: http://digitalcommons.calpoly.edu/theses/1593
Date of Award
MS in Computer Science
Companion or support characters controlled by Artificial Intelligence (AI) have been a feature of video games for decades. Many Role Playing Games (RPGs) offer a cast of support characters in the player’s party that are AI-controlled to various degrees. Many First Person Shooter (FPS) games include semi-autonomous or fully autonomous AI-controlled companions. Real Time Strategy (RTS) games have traditionally featured large numbers of semi-autonomous characters that collectively help accomplish various tasks (build, attack, etc.) for the player. While RPGs tend to focus on a single or a small number of well-developed character companions to accompany a player controlled main character, the RTS games tend to have anonymous and replaceable workers and soldiers to be micromanaged by the player.
In this paper we present the MimicA framework, designed to govern AI companion behavior based on mimicking that of the player. Several features set this system apart from existing practices in AI-managed companions in contemporary RPG or RTS games. First, the behavior generated is designed to be fully autonomous, not partially autonomous as in most RTS games. Second, the solution is general. No specific prior behavior specifications are modeled. As a result, little to no genre, story or technical assumptions are necessary to implement this solution. Even the list of possible actions required is generalized. The system is designed to work independently of game representation. We further demonstrate, analyze and discuss MimicA by using it in Lord of Towers, a novel tower defense game featuring a player avatar. Through our user study we show that a majority of participants found the companions useful to them and liked the idea of this type of framework.