Artificial Intelligence Methods for Creating a Computer Player in a Game

U sklopu ovog rada implementirana je video igra po uzoru na postojeću naziva "Light Riders". Cilj rada je bio izraditi računalnog igrača za tu video igru. Korištene su razvojne tehnike podržanog učenja i neuroevolucije povećavajućih topologija i tehnike planiranja. Sve korištene tehnike su opisane. Igrači izrađeni tehnikama planiranja korišteni su i kao dio razvojnih tehnika, bilo kao okruženje agenta u podržanom učenju, bilo kao dio funkcije dobrote u neuroevoluciji. Korištena je i tehnika učenja igranjem sam protiv sebe. Implementacija razvojnih tehnika uključivala je dizajniranje stanja igre i nagrađivanja / funkcije dobrote. Igrači su izrađeni pod različitim hiperparametrima i njihovom usporedbom su se podržale određene tvrdnje.In this thesis the video game modeled by the one called "Light Riders" was implemented. The goal of thesis was to develop a computer player for this video game. Development techniques that were used were reinforcement learning and neuroevolution of augmenting topologies (NEAT). Also, planning techniques were used. All techniques were described. Planning players were used as a part of development techniques, either as an agent's environment in reinforcement learning, either as a part of fitness function in neuroevolution. Self-play technique was also used. Implementation of development techniques included the design of game state and reward function / fitness function. Players were developed with different hyperparameters and by their comparison certain statements were supported