Desenvolvimento de um framework em python para programação declarativa de sistemas multi-agentes BDI com técnicas de aprendizagem

Abstract

Multi-agent systems are used in numerous research and development areas, from artificial intelligence to the social sciences. Several agent programming frameworks, such as JaCaMo, SPADE, and JADE, offer essential features for developing multi-agent systems. However, there is still a gap in the search for a platform that allows declarative agent creation, following the Belief, Desire, Intention (BDI) model, natively in Python. To address this gap, a systematic literature review was conducted, focusing on multi-agent system development platforms and learning techniques. This study revealed a lack of comprehensive solutions for declarative agents. Therefore, this work presents the development of the Multi-Agent Systems in Python Framework, or MASPY, for the declarative creation of multi-agent systems following the BDI paradigm. This work aimed to develop classes for the declarative creation of multi-agent systems, initially incorporating a reinforcement learning component. The choice to develop this framework in Python is due to the ease of implementing learning techniques in the language. Each class contains methods for managing information and aiding in designing a system that follows the BDI paradigm. This work presents a complete framework and a basic learning class. All classes, their internal workings, and practical implementations are presented and described, which are validated in detail and compared with another Python framework, SPADE. The results show that MASPY fulfils its purpose as a platform for creating BDI MASs and has advantages in simplicity and performance. Some bottlenecks and low-performance points were identified, but the framework is at a stage comparable to a recognised platform in the field and ideal for programmers seeking declarative BDI development in Python.