Intelligent Behavior for Computer Game Characters Using Prediction and Learning

This report details the work done and knowledge learned during the first semester of the author’s thesis (CS 297). Although video game graphics has improved tremendously, artificial intelligence in games has taken a much smaller step forward. One of the most promising new prospects in game AI is for the computer to learn and adapt to the environment, and in particular to the player. The thesis aims to create a fighting game with an AI system capable of learning and adapting to the player. Four deliverables were attempted this first semester. All were completed with good results with the exception of deliverable three, which will be completed before next semester. Deliverables one and four explored three AI techniques: n-grams, string-matching prediction, and Hidden Markov Models in detail. Deliverable two was to generate a document recording the concept and ideas for the game, called Alpha Fighter. Deliverable three was to create character models and animations. From the work completed this semester, the author is equipped with the assets and knowledge necessary for implementation of the game next semester.