Continuous evolution of neural modules for autonomous robot controllers

In recent years, research on techniques for developing controllers for autonomous robots has been conducted. Evolutionary Algorithms are among the most popular tools used in this type of problem, mostly for its capacity to adapt to the environment. Nevertheless, they are usually applied to produce a...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Vinuesa, Hernán Luis, Osella Massa, Germán Leandro, Corbalán, Leonardo César, Lanzarini, Laura Cristina
Formato: Objeto de conferencia
Lenguaje:Inglés
Publicado: 2007
Materias:
Acceso en línea:http://sedici.unlp.edu.ar/handle/10915/23177
Aporte de:
Descripción
Sumario:In recent years, research on techniques for developing controllers for autonomous robots has been conducted. Evolutionary Algorithms are among the most popular tools used in this type of problem, mostly for its capacity to adapt to the environment. Nevertheless, they are usually applied to produce a controller that will not continue its adjustment after concluding this process. This causes trouble to a controller when it is used in a dynamic environment. In this paper, the combination of a state-of-the-art modular neuro-evolution algorithm with a specific evolutionary algorithm is proposed. The former method is used to generate the controller while the later is used to adjust it during its operation. As a result, an adaptable controller based on a minimal topology neural network is obtained. The method proposed was tested in a goal-reach problem with satisfying results. Finally, conclusions are presented.