Building Multi-Agent Systems With Reinforcement Hierarchical Neuro-Fuzzy Models

Título: Building Multi-Agent Systems With Reinforcement Hierarchical Neuro-Fuzzy Models

Autores: Corrêa, Marcelo França; Vellasco, Marley; Figueiredo, Karla

Resumo: Multi-Agent Systems (MAS) are an emergent area of Computational Intelligence that provides the tools for the construction of complex systems involving multiple agents and the coordination mechanisms amongst them. Because of its properties, MAS has been used into a great variety of fields, such as distributed control, robotic teams, navigation control, route planning control, lift programming, cargo balancing, automatic trading among agents, etc. This paper presents a new MAS that works though hierarchical neuro-fuzzy hybrid models. The main advantage of this class of system is the autonomous capacity to create rules, expand their own rule structure, and extract knowledge from the direct interaction between the agent and the environment. The new model was tested through the Pursuit game benchmark application. Some preliminary results have shown promising. The tests demonstrated that the developed system has shown a good capacity of convergence and coordination among the intelligent agents.

Palavras-chave: Multi-Agent Systems (MAS); hierarchical neuro-fuzzy; intelligent agents; reinforcement learning

Páginas: 8

Código DOI: 10.21528/CBIC2011-12.2

Artigo em pdf: st_12.2.pdf

Arquivo BibTex: st_12.2.bib