Razvoj programskih agenata za složene videoigre

master's thesis

• Title (english): Programming software robots for complex video games
• Author: Mislav Vuletić
• Mentor: Mirko Randić (mentor)
• Committee member: Mirko Randić (predsjednik povjerenstva)
• Committee member: Bruno Blašković (član povjerenstva)
• Committee member: Luka Humski (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-07, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Kako umjetna inteligencija postaje sve popularnija nove metode potpornog učenja se neprestano razvijaju.
Istraživanje je fokusirano na povećanje apstrakcije zadržavajući ili poboljšavajući postignut rezultat i/ili vrijeme treniranja prošlih agenata.
Nadalje, biblioteke strojnog učenja programskih jezika omogućuju izradu agenata umjetne inteligencije s lakoćom.
Mogu li pojedinci razviti vlastite modele potpornog učenja bez mnogo domenskog znanja i trenirati agente na tipičnom igraćem računalu u razumnom vremenu te nadigrati ugrađene agente igara koji koriste metode pretraživanja prostora stanja i stabla odluka?
U ovom radu istražujemo tehnike potpornog učenja, složenost igara te je dan opis izrade agenta potpornog učenja za pojednostavljenu verziju igre DOTA 2 koristeći programsko sučelje Project Breezy za interakciju s igrom.

Abstract (english)

As artificial intelligence is becoming increasingly more popular new methods and techniques of applying reinforcement learning are constantly emerging.
Research focuses on abstraction improvement while retaining or even outperforming scores and/or learning speed in previously made agents.
Furthermore, machine learning libraries for programming languages make it possible to build AI agents with ease.
Can individuals build their own RL models without much domain knowledge and train agents on typical gaming hardware in a reasonable time frame to outperform pathfinding or decision tree agents found in most video games?
In this paper we explore RL techniques, game complexity and build a RL agent for a simplified version of a popular video game DOTA 2 using Project Breezy to interact with the game client.

• Keywords: DOTA; DOTA 2; Breezy; SF; Shadow Fiend; 1v1; potporno učenje; tensorflow
• Keywords (english): DOTA; DOTA 2; Breezy; SF; Shadow Fiend; 1v1; reinforcement learning; tensorflow; DDQN; DQN
• Language: croatian
• URN:NBN: urn:nbn:hr:168:851621
• 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: 2023-01-13 14:17:09