Monte-Carlo Tree Search Implementation of Fighting Game AIs Having Personas
Ryota Ishii, Suguru Ito, Makoto Ishihara, Tomohiro Harada, Ruck Thawonmas · 2018
In this paper, we propose a method for implementing a game AI with a persona using Monte-Carlo Tree Search (MCTS). Video games are now a powerful entertainment media not just for players but spectators as well. Since each spectator has personal preferences, customized spectator-specific gameplay is arguably a promising option to increase the entertainment value of video games streaming. In this paper, we focus on personas, which represent playstyles in the game, in particular fighting games. In order to create an AI player (character) with a given persona, we use a recently developed variant of MCTS called Puppet-Master MCTS, which controls all characters in the game, and introduce a new evaluation function, which makes each character take their actions according to the given persona, and roulette selection-based simulation to this MCTS. The results of a conducted experiment using FightingICE, a fighting game platform used in a game AI competition at CIG since 2014, show that the proposed method can make both characters successfully behave according to given personas, which were identified by participants - spectators - in the experiment.