A hybrid agent architecture for learning good cooperative behaviours for game characters

Date
2012
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Creating intelligent game agents is a difficult problem, cspcci??1lly in non-dctenninistic games where the outcome of an action cannot be determined with certainty. When we add in the complexity of team based games and the need for cooperation between the agents the problem can become even more complex. In this work, we provide a hybrid agent architecture that can be used to create and train agents to play team based games. Our architecture uses role based agent pools, communicated intentions, reinforcement learning, and evolutionary learning to train a team of agents. We test our architecture by applying it to the turn-based strategy game Battle for Wesnoth and demonstrate that we can train effective agents that can work together in teams to win against the built in \i\Tesnoth artificial intelligence.
Description
Bibliography: p. 92-95
Some pages are in colour.
Keywords
Citation
Paskaradevan, S. (2012). A hybrid agent architecture for learning good cooperative behaviours for game characters (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/4816
Collections