GasNet controllers for a virtual robot in discrete and continuous versions of a food gathering task
Tipo
Artigo de evento
Data de publicação
2003
Periódico
Proceedings of the IASTED International Conference on Artificial Intelligence and Soft Computing
Citações (Scopus)
0
Autores
Santos C.L.R.
De Oliveira P.P.B.
De Oliveira P.P.B.
Orientador
Título da Revista
ISSN da Revista
Título de Volume
Membros da banca
Programa
Resumo
Further experiments are reported on the GasNet, a recently proposed artificial neural network model, whose units can communicate to each other also via a virtual gas that diffuses through the network (in addition to the standard electric interaction). Results from previous investigations have shown that GasNet transposition from its original application domain in evolutionary robotics to discrete benchmark problems in the literature did not yield the same good results. Here, we extend the investigations on one of those problems - a food gathering task for a simulated robot - even reaching a continuous version of the task. The results provide strong empirical evidence about GasNet's dependence on the continuous dynamics of its problem domain, but raises question on the effectiveness of its chemical substrate, when its performance is compared to one of its possible gasless versions.
Descrição
Palavras-chave
Assuntos Scopus
Evolutionary robotics , GasNet and distributed genetic algorithms