Дослідження методів прийняття рішень у відеоіграх за допомогою поведінкових дерев

Abstract

This work analyzes Utility-based and Learning Behavior Trees to improve non-player character decision-making, focusing on the balance between rapid reasoning and adaptability. Mathematical modeling indicates that Utility systems offer swift choices for simple agents but suffer decision degradation as complexity increases, whereas Learning systems maintain consistent cognitive performance regardless of the action repertoire size. Consequently, the research establishes that Utility approaches are optimal for basic choices, while Learning architectures provide the necessary decision-making scalability for sophisticated adversaries in dynamic gaming environments.