Comparação entre controladores fuzzy e neural desenvolvidos via simulação e transferidos para ambientes reais no âmbito da robótica evolutiva

Detalhes bibliográficos
Ano de defesa: 2018
Autor(a) principal: Farias, Weslley Alves
Orientador(a): Freire, Eduardo Oliveira
Banca de defesa: Não Informado pela instituição
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Não Informado pela instituição
Programa de Pós-Graduação: Pós-Graduação em Engenharia Elétrica
Departamento: Não Informado pela instituição
País: Não Informado pela instituição
Palavras-chave em Português:
Palavras-chave em Inglês:
Área do conhecimento CNPq:
Link de acesso: http://ri.ufs.br/jspui/handle/riufs/9569
Resumo: One of the greatest limitations of Evolutionary Robotics is when transfering controllers evolved by simulation to real environments. This limitation is mainly caused by model simplifications and difficulties to represent dynamic characteristics, whether from the robot or the environment. And this results in performance degradation of the evolved controller after the transfer, a phenomenon called reality gap. Because this problem is a limitation for practical and complex applications of evolutionary robotics, many solutions have been proposed since the 90s. Until now, most of the research use control strategies based on artificial neural networks because they allow algorithms to be evolved with less designer influence. On the other hand, fuzzy logic can also be used for the development of controllers in the field of evolutionary robotics because it also allows emulating human intelligence. Therefore, this dissertation investigates whether fuzzy control systems are more robust than neural control systems, both optimized by a genetic algorithm in simulation and later transferred to a real robot in physical environment in the task of autonomous navigation while avoiding obstacles. The results show that in the analyzed conditions, fuzzy controllers present better transfer characteristics, mainly considering the smoothness of the executed trajectory, and an equivalent performance, when compared with neural controllers.