Avaliação de técnicas de planejamento para NPCs: GOAP e UD GOAP, em um jogo de simulação de loja

Abstract

The use of planning architectures for the artificial intelligence (AI) of non-player characters (NPCs) has increased in digital game development, especially in scenarios that require adaptive and modular behavior. Among the most adopted models, Goal-Oriented Action Planning (GOAP) stands out, allowing dynamic decision-making through the formulation of an objective-based sequence of actions. However, due to design limitations of GOAP, variations such as Utility-Directed GOAP (UD GOAP) have been proposed in the literature. This work developed a game environment used as a testing platform to compare GOAP and UD GOAP, assessing both computational performance and the quality of the behavior produced by each approach. Controlled scenarios containing multiple NPCs were implemented, from which execution time, frames per second, and behavioral patterns were collected. The results revealed relevant differences and similarities between the architectures, highlighting variations in computational cost and the impact on agent behavior. The analysis contributes to understanding the advantages and limitations of each technique, supporting both academic research in game-oriented AI and practical applications in the development of interactive systems.