Primjena algoritma učenja s podrškom u razvoju računalnih igara

In this paper, reinforcement learning was examined by creating a Python puzzle video game and implementing a Q-learning algorithm. Having full control of the development process, it was possible to track, explain, and visualize every step of the algorithm. Along with the theoretical background, this thesis serves as an introductory paper to understand the inner workings of simple machine—or reinforcement—learning algorithms while applying them in video game environments. Reinforcement learning is an industry standard tool in the development of video games, and having a firm grasp of the methods used to implement such tools is highly pertinent within the field. I also analyze how tweaking various parameters of the Q-learning formula yields different degrees of optimization. An attempt is made to find a more optimal parameter combination, which itself is contextually dependent on the environment the agent is training in