Rješavanje problema igranja video igara korištenjem algoritma NEAT

master's thesis

• Title (english): Solving the video game playing problem using the NEAT algorithm
• Author: Luka Terzić
• Mentor: Marko Đurasević (mentor)
• Committee member: Marko Đurasević (predsjednik povjerenstva)
• Committee member: Domagoj Jakobović (član povjerenstva)
• Committee member: Nikolina Frid (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-15, Croatia

Abstract

Ovaj rad bavi se analizom performansi neuroevolucijskog algoritma NEAT za potrebe učenje igranja igara Flappy Bird, Donkey Kong Country i Sonic the Hedgehog. Implementirano je grafičko korisničko sučelje za igru Flappy Bird, a za igre Donkey Kong Country i Sonic the Hedgehog koristi se grafičko korisničko sučelje biblioteke gym-retro. Na primjeru igre Flappy Bird ispitana je razlika u performansama treniranja uz korištenje sigmoidalne i arkustangensne aktivacijske funkcije. Također, ispitan je utjecaj veličine populacije na performanse treniranja. Na primjeru igre Donkey Kong Country, čiji agent kao ulaz prima sliku, ispitane su performanse treniranja uz različite faktore smanjivanja slike ulaza. Na kraju, koristeći prikupljeno znanje parametara NEAT algoritma i utjecaja faktora smanjivanja slike implementiran je agent koji uspješno igra igru Sonic the Hedgehog te je prikazana usporedba rezultata treniranja uz dvije različite funkcije dobrote.

Abstract (english)

This paper deals with the analysis of the performance of the neuroevolutionaty algorithm NEAT for the purpose of learning to play the games Flappy Bird, Donkey Kong Country and Sonic the Hedgehog. A graphical user interface has been implemented for the game Flappy Bird, and the graphical user interface of the gym-retro library is used for the games Donkey Kong Country and Sonic the Hedgehog. Using the example of the Flappy Bird game the difference in training performance was examined using the sigmoid and arctangent activation functions. Also, the influence of population size on the training performance was examined. On the example of the game Donkey Kong Country, whose agent receives an image as input, the performance of different input image reduction factors was tested. Finally, using the collected knowledge of the NEAT algorithm parameters and the influence of the image reduction factor, an agent that successfully plays the Sonic the Hedgehog game was implemented, and a comparison of the training results with two different fitness functions was presented.

• Keywords: Neuroevolucija; NEAT; Flappy Bird; Donkey Kong Country; Sonic the Hedgehog
• Keywords (english): Neuroevolution; NEAT; Flappy Bird; Donkey Kong Country; Sonic the Hedgehog
• Language: croatian
• URN:NBN: urn:nbn:hr:168:951652
• 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: Open access
• Created on: 2024-10-12 22:20:48