Development of werewolf match system with analysis of human gaze motion
Daisuke Katagami, Satoshi Nira · 2017
Recently the progress of AI in a complete information game is remarkable. Incomplete information games are attracting attention as the next research theme. Among them, communication games called werewolf game are attention. Nonverbal information is important in werewolf game, and It is expected to be able to natural discussion can be made by incorporating non-verbal information when a life-like agent performs werewolf game. In this research, we aim to develop a werewolf match system that can perform werewolf game naturally by introducing appropriate gaze motion to life-like agent. As an approach, we analyzed the gaze in the werewolf games played by humans, classified the gaze motion in the werewolves, and developed a 6-screen werewolf match System with classified gaze motions. Using the developed system, we created a werewolf video that gives gaze motion and a werewolf video that does not give gaze motion, and conducted an experiment for validating the effect by gaze. As a result, when giving the gaze motion to the agent, experiment participants who noticed the gaze motion in particular have found that the game tends to look naturally.