Um framework baseado em contêineres para sistemas multiagente abertos

Abstract

Multi-agent Systems (MAS) are formed by agents that interact with each other and can develop autonomous and cooperative behavior. These types of systems are used to solve problems where entities are decentralized. A specific type of MAS allows interaction between agents that participate in different models, such as the Open Multi-agent Systems (OMAS). In OMAS, heterogeneous agents inserted in different environments or models can migrate from one system to another, taking their attributes and knowledge. However, complex problems arise when developing OMAS applications compared to MAS. For example, there may be implementation issues, where agents and models may be developed by different teams, in different program ming languages, or even on different agent platforms/architectures. Furthermore, opening MAS is challenging due to the uncertainties and all the dynamic behavior that the exchange of agents entails. Thus, it is necessary to formulate techniques to analyze this complexity and facilitate the understanding of the system’s global behavior. Given the context presented, it is necessary to make a research effort to reduce the complexity of the OMAS. This research proposes a framework to assist the development of Open Multi-agent Systems based on containers to facilitate the migration of agents between different models that can execute in heterogeneous scenarios of hardware and software. Simulation scenarios were developed based on the Sugarscape 2 Constant Growback (NetLogo) and Gold Miners (JaCaMo) models to analyze the feasibility of implementing the framework. The tests cover local, remote (cloud), and hybrid executions. The tests showed that the framework allows the exchange of agents between different models that can be developed on different platforms.