Investigating Kinect-based fighting game AIs that encourage their players to use various skills

Abstract

This paper investigates AIs that increase their players' amount of exercise by encouraging the usage of various skills in fighting game FightingICE, recently used in a number of game AI competitions. Our research aim is to develop such AIs for promoting players' health with fighting games that use Kinect as the input interface. In our experiment, two types of AIs are used as the opponent against a human player. One of the AIs is based on the k-nearest neighbor algorithm and fuzzy control, and the other is based on UCT, a variation of Monte-Carlo Tree Search. Our results show that the players, participating in the experiment, use more different skills, thus demonstrating higher action entropy, when playing the game against the UCT AI.