Učenje strategija igranja računalnih igara

master's thesis

• Title (english): Learning Playing Strategies for Computer Games
• Author: Joel Mislav Kunst
• Mentor: Marko Čupić (mentor)
• Committee member: Marko Čupić (predsjednik povjerenstva)
• Committee member: Željka Mihajlović (član povjerenstva)
• Committee member: Domagoj Jakobović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2016-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Automatsko igranje igara je zanimljiva tema, te postoje brojni sustavi koji pokušavaju igrati igre automatski. Cilj ovog rada bio je pokušaj ostvarenja sustava koji može igrati neku grupu igri. Znači jedan sustav koji na temelju konfiguracije za pojedinu igru, uči pravila, pronalazi "dobre" strategije igranja te igra igru.

Uspostavilo se da je problem veoma složen te se na kraju za grupu podržanih igara odabrala grupa strateških igara u stvarnom vremenu koje se igraju u web-preglednicima. Igrač je vladar nekakvih posjeda (planet, selo, ..), te razvija ekonomiju, zgrade, napredovanja i vojsku kako bi se uzdigao iznad ostalih igrača po nekom od kriterija.

Primjeri takvih igara za koje je i napisana konfiguracijska datoteka za demonstraciju rada sustava su Ogame i Travian. Te igre imaju različite načine na koje igrač može igrati ovisno o svojim zamislima i željama. Razvijen sustav igra igrača "farmera". Igraču farmeru cilj je razviti ekonomsku situaciju te imati što veću produkciju resursa.

Sustav pravila igre doznaje iz konfiguracijske datoteke, a odluku donosi na temelju simuliranja poteza, odluke neuronske mreže te vrednovanje stanja u kojima se igra u tim simulacijama nalazi. Nakon što je sustav pronašao zadovoljavajuće stanje, kreće s izvođenjem koraka kako bi došao do simuliranog stanja.

Sustav automatizira igračeve akcije po web-pregledniku koristeći PhantomJS, web-preglednik koji se nalazi samo u memoriji računala i nema korisničko sučelje, ali nudi programsko sučelje, kako bi program (ovdje naš sustav) mogao posjećivati web-stranice, obavljati na njima akcije čovjeka, te čitati podatke sa stranice.

Za korištenje sustava korisnik osim podešavanja konfiguracijske datoteke treba implementirati sučelje GameHook na željeni način (ovdje korišten PhantomJS). Sučelje definira metoda ze interakciju sa stvarnom igrom.

Rezultati implementiranog sustava su veoma zadovoljavajući. U obje igre sustav je došao do TOP 100 pozicija, što se kod tih igara smatra dobrim rezultatom, te naravno može napredovati i dalje.

Abstract (english)

Automatic games playing is interesting subject, and there is a lot of systems that try to play games automatically (without human's help). The goal of this thesis was to try achieve a system that can play not just one but a group of games. One system that plays multiple games based on configuration file explaining the game's rules learns how to play, and finds "good" strategies to play the game.

It turns out that described task is very complex and in the end the group of games that the system can play was narrowed down to very restricted area of games. Choosen games are masive multiplayer online real time strategy games that are played in web browsers, and have only the data as information (no parsing images, animations, etc). In those games player is a ruler of some property (planet, village, ...) and is developing the economy, buildings, research, army, .. in attempt to rise above other players by some of the criterion.

Example of those games for which we wrote the configuration files for system demonstration are Ogame and Travian. Those games have different ways the player can play them based on player's goals and intentions. Developed system plays the role of "farmer". Player farmer has the goal of developing the economy situation and develop his resource production as much as possible.

System learns the rules from the configuration file, and the decision about what to do is based on simulation of possible moves, decisions of neural network and evaluation of states in which games end up doing those moves. After system has found the acceptable game state, it executes the action needed to get to that state.

System is automating players action in the web browser by using PhantomJS, in memory web browser. The browser does not have user interface but offers programatical interface (API) for navigating the web and parsing the pages.

To use the system a user has to implement GameHook interface. That interface defines methods for interaction with the real game. We have used PhantomJS for that.

Achieved results of the implemented system are quite good. In both games system got into TOP100 by general statistics which is a realy good result. Of course system can progress further in time.

• Keywords: mmo; web-preglednik; ogame; travian; igre; strategija; učenje; bot; neuronska mreža; umjetna inteligencija
• Keywords (english): mmo; web browser; ogame; travian; games; strategy; learning; bot; neural network; artificial intelligence
• Language: croatian
• URN:NBN: urn:nbn:hr:168:084762
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2018-02-06 10:00:02