Otimização de agentes inteligentes no jogo Ludo por meio de um Algoritmo Genético

Abstract

Genetic Algorithms (GAs) are optimization techniques inspired by the principles of natural evolution, widely used in solving complex problems characterized by large search spaces and high variability. In the context of board games, especially those involving randomness and interaction between multiple players, the manual definition of effective strategies becomes a significant challenge. This work investigates the application of Genetic Algorithms in optimizing the decision parameters of an intelligent agent in the game Ludo, a stochastic environment in which probabilistic factors, such as dice rolls, directly influence the course of the games. To this end, a logic simulation environment and an agent based on weighted heuristics were implemented, whose strategy is encoded by a chromosome composed of five real genes. These genes represent weights associated with different strategic aspects of the game, such as progress, attack, safety, risk, and completion. The Genetic Algorithm is used to evolve these weights, and systematic experimental analyses are performed on different evolutionary configurations, including population size, number of generations, selection methods, elitism, crossover, and mutation. The results obtained demonstrated that the Genetic Algorithm is capable of evolving agents with consistent performance superior to that of random agents, exhibiting stable convergence and low variability between independent runs. The analysis of the winning DNAs reveals recurring strategic patterns, indicating the emergence of coherent and interpretable behaviors. Thus, the study highlighted the potential of Genetic Algorithms as an effective approach for optimizing intelligent agents in stochastic games, in addition to contributing to the practical understanding of the impact of genetic operators in this type of problem.