Artificial Intelligence in Strategic Games

Rad razmatra korištenje metoda umjetne inteligencije na području igranja igara. U sklopu rada je razvijena jednostavna strateška igra za 2 igrača koju je moguće igrati protiv razvijenih računalnih agenata. Igra je implementirana koristeći Unity te je dostupna na operacijskim sustavima Windows i Android. Razvijeni su i implementirani algoritmi Monte Carlo pretraživanja stabla, Minimax pretraživanja stabla te algoritam Q-učenja, podvrsta podržanog učenja. Algoritam Minimax koristi posebno razvijenu heuristiku specifičnu za ovu igru te alfa-beta skraćivanje koje značajno ubrzava pretragu. Eksperimentalno se evaluiraju međusobni odnosi svih razvijenih agenata, uspoređujući udio pobjeda i poraza, kao i prosječno vrijeme odabira poteza.This paper examines usage of artificial intelligence methods on game playing. As a part of this paper, there was developed a simple 2 player strategy game which could be played against developed computer agents. The game is implemented in the Unity game engine and is available for Windows and Android operating systems. Algorithms that were developed include Monte Carlo tree search, Minimax tree search and Q-learning. Minimax tree search is using alpha-beta pruning and custom heuristic function developed specifically for this game. Experimental evaluations include comparison of win and loss rate as well as average turn time for each of the agents