Análise comparativa de agentes reativos e deliberativos no ambiente TileWorld utilizando o framework MASPY

Abstract

The development of multiagent systems has benefited from frameworks that facilitate the implementation of complex architectures. In this context, MASPY, a Python framework based on the BDI (Belief-Desire-Intention) paradigm, is a promising tool that, due to its recent development, requires validation. This work aims to test and validate MASPY through the development and comparison of agents with different architectures in solving the classic TileWorld problem. TileWorld consists of an environment where agents must push tiles to fill holes while navigating around obstacles. For this investigation, two types of agents were implemented: a reactive agent, characterized by random exploration and decision-making based on local information, and a deliberative agent, equipped with BDI architecture and A* algorithm for optimized route planning. The performance of both approaches was systematically evaluated across three distinct scenarios, using quantitative metrics such as execution time, number of reasoning cycles, final score, and number of movements performed. Data collection was conducted through MASPY's native functionalities. The obtained results demonstrated that, although deliberative agents present advantages in most tested scenarios, reactive agents can solve problems in less time when the environment has a high density of elements. In addition to confirming the technical feasibility of the framework for agent development, environment implementation, and experimental data collection, this work contributes significantly to the empirical validation of MASPY as a multiagent systems development tool.