Návrh a vývoj počítačové hry s adaptivním chováním řízeným umělou inteligencí

Thesis focuses on the development of a 3D action RPG game titled Arcane Souls in the Unity game engine, with a primary focus on implementing modern machine learning methods for creating game artificial intelligence. The main objective of the thesis is to create an adaptive opponent that serves as an advanced alternative to traditional, often predictable finite state machines. As part of the practical component, a game featuring a combat system and multiplayer was designed using Unity Netcode and the Steamworks API. The core of the research lies in the design and training of artificial intelligence models using Reinforcement Learning within the Unity ML-Agents toolkit. The thesis presents a hybrid AI architecture where finite state machines control movement logic outside of combat, while the actual combat and real-time decision-making are entrusted to neural networks. The thesis further analyzes the technical complexity of training and, based on the data obtained, offers recommendations for the effective integration of machine learning into future game projects.