Thesis

General video game playing using ensemble decision systems

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2020
Thesis identifier
  • T15703
Person Identifier (Local)
  • 201555913
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • This thesis explores the application of Ensemble Decision Systems as an Artificial Intelligence (AI) agent for playing video games on the General Video Game Artificia lIntelligence (GVGAI) platform. The GVGAI offers a platform for research into developing General Video Gameplaying AI agents. Significant progress has been made over the years in the area of game playing AI agents, but it is often a trivial task for designers to propose new problems that the agents are unable to solve. Humans are typically able to solve these additional problems, making them an ideal model for an excellent general video game player and a benchmark for AI agents. One of the objectives of this thesis has been the introduction of Deceptive Games to the GVGAI, which are a class of games that are designed to deliberately deceive AI agents. Ensemble Decision Systems make use of multiple AI algorithms to make their decisions, which may make them more robust to the problems that can deceive singular AIagents.;A wide variety of Ensemble Decision Systems were developed and compared with agents from the GVGAI competition, with the aim of developing an indication of the current level of performance that agents can reach. The Ensemble Decision Systems show improved generality, being able to complete a wider range of games than other agents, at a cost of win rate in specific games. This thesis presents an Ensemble Decision System for GVGP and a suite of Deceptive Games. The Ensemble Decision System detailed in this thesis manages to out perform comparison agents in the Deceptive Games suite, with the top three positions being taken by Ensemble agents for win rate, and a wider range of games.
Advisor / supervisor
  • Levine, John R.
Resource Type
DOI
Date Created
  • 2020
Former identifier
  • 9912922291502996

Relazioni

Articoli