Mimicking human strategies in fighting games using a Data Driven Finite State Machine

摘要

Multiplayer fighting videogames have become an increasingly popular over the last few years, especially with the introduction of online play, making for a more competitive experience. Multiplayer fighting games give players the opportunity to utilize particular strategies and tactics to win, allowing them to use their own signature style. As a player can only play against a particular opponent who is actively participating in the game themselves, they cannot practice combating the opponent's style if the opponent is not participating in the game. This paper presents a novel approach for an avatar to learn and mimic the style of a player. It does this by recording and analyzing the data before splitting it up into two tiers; tactical data and strategic data. The approach uses a Naïve Bayes classifier to classify the tactics to particular states, and a Data Driven Finite State Machine to dictate when certain tactics are used. Statistics recorded during an experiment involving the approach are discussed, which indicate that the architecture of the Artificial Intelligence is fit for purpose, but does require refinement. Limitations of the architecture are discussed, including that such an approach may not provide accurate results when more parameters are considered.