Genetski pristop k strateškim igram

KRISTL, ČRT (author), Fijavž, Gašper (mentor), Brodnik, Andrej (co-mentor)

V delu razvijemo in implementiramo genetski algoritem s katerim optimiziramo nabor znanj, opreme in lastnosti akcijskih junakov v strateški igri Warcraft III, oziroma njeni različici The Kingdom of Kaliron. Izbiro optimalnega nabora, takega pri katerem je akcijski junak kar se da uspešen v bojevanju z nasprotniki, zapišemo kot problem kombinatorične optimizacije, za njegovo reševanje pa uporabimo pristop z genetskim algoritmom. Za oceno sposobnosti junaka smo implementirali simulacijo, za katero smo potrebovali natančno poznavanje mehaničnih in programskih principov igre Warcraft III. Med drugim smo uporabljali metode vzvratnega inženirstva. Konvergenco genetskega algoritma smo zagotovili z uporabo otokov, ločenih podpopulacij s šibko medsebojno interakcijo, in s pazljivo izbiro parametrov genetskega algoritma. Poleg tega smo genetskemu algoritmu dodali spomin, ki pri ustvarjanju novih populacij pripomore k boljšem začetnem stanju osebkov. Ustrezno časovno učinkovitost pa smo pridelali s paralelizacijo metode.

• Language: English
• Keywords: genetski algoritem, računalniške igre, optimizacija, paralelizacija, simulacija
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2016
• PID: 20.500.12556/RUL-80202
• Publication date in RUL: 11.02.2016

Secondary language

• Language: Slovenian
• Title: Genetic Approach to Strategic Games

Abstract

In this thesis we develop and implement a genetic algorithm to optimize a set of talents, equipment and sub-attributes of characters in the game Warcraft III and its modification The Kingdom of Kaliron. Finding the optimal set where a character performs the best in fights against enemies is a combinatorial problem for which we use a genetic algorithm to solve. To be able to evaluate a character, we implemented a simulation that required deep knowledge of game mechanics and programming principles of Warcraft III. We also used reverse engineering as a tool. We ensured convergence of a genetic algorithm with the use of population islands, which are disjoint subpopulations with weak mutual interactions, and with careful choosing of genetic algorithm parameters. We also implemented genetic algorithm memory, which helps create better initial individuals when creating new populations. Finally, we used parallelization to reduce the running time of the algorithm.

• Keywords: genetic algorithm, computer games, optimization, parallelization, simulation