Investigation of Various Online Adaptation Methods of Computer-Game AI Rulebase in Dynamic Scripting

Inthispaper,weinvestigatevariousonlineadaptationmethodsofcomputer-gameAIrulebaseinatechniquecalledDynamic Scripting (DS). DS is a technique for balancing the level of computer controlled characters that play computer gamesagainsthumanplayers. Itonlineupdatesruleweights inrulebasethatin∞uencethebehaviorofthecomputercontrolledcharacter. However,theweightupdatingmechanism ofDSisnotefiectiveifimproperinitializationoftherulebase isdone. Inourpreviouswork,weproposedacomplementary method to DS that replaces ine‐cient rules with randomly generatedrules. Inthepresentwork,weproposethreemore methods and compare them with the previously proposed one and with the original DS, using a simulator in which one adaptive character duels against one hard-coded character(HC).Ourflndingisthatthemethodthatreplacesan ine‐cient rule with the rule least similar, among the given candidates,totheine‐cientoneisofthebestperformance, in terms of the winning rate against HC, for mediocre and weakinitial rulebase conditions.