Configuration of a Paraconsistent Artificial Neural Network for the Learning from Demonstration Method applied to a Robotic Arm

• Autor(a) principal: Gomes, Paulino Machado
• Data de Publicação: 2022
• Outros Autores: Fernandes, Cláudio Luís Magalhães, Silva Filho, João Inácio da, Silveira, Rodrigo Silvério da, Santo, Leonardo do Espírito, Mario, Mauricio Conceição, Rosa, Vitor da Silva, Torres, Germano Lambert
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/29720
• topic: Aprendizagem por demonstração; Ensino; Inteligência Artificial; Lógica paraconsistente anotada; Rede neural artificial paraconsistente.; Paraconsistent Annotated Logic; Learning from demonstration; Artificial Intelligence; Teaching; Paraconsistent artificial neural network.; Lógica Paraconsistente Anotada; Aprendizaje por Demostración; Enseñando; Inteligencia Artificial; Red neuronal artificial paraconsistente.

The Annotated Paraconsistent Logic - LPA is a non-classical logic, based on concepts that allow, under certain conditions, to accept the contradiction in its foundations, without invalidating the conclusions. Mathematical interpretations in its associated lattice make it possible to obtain equations and algorithm constructions, which form efficient paraconsistent analysis networks, in treating signals simulating learning. The algorithm used in this research is called Paraconsistent Artificial Neural Cell of Learning (CNAPap), and was created from equations based on LPA. With standardized signals repeatedly applied to its input, CNAPap is capable of gradually storing this information, increasing or decreasing its level of response at the output with asymptotic variation, controlled by a Learning Factor (FA). To run the tests, a set of five CNAPaps forming a learning Paraconsistent Artificial Neural Network (RNAPap), was implemented in an ATMEGA 328p microcontroller and several tests were carried out to validate its operation, acting on learning by demonstration (LfD) in a Robot Manipulator. Considering the fragile mechanical structure of the Robot Manipulator, and the sensor devices adapted to respond to the standards, the laboratory results obtained in the various tests presented were satisfactory, and the microprocessed system built responded efficiently, where the levels of correct answers corresponded to between 75 % to 90%, at all stages of the LfD method. The results of comparative studies showed that RNAPap has dynamic properties capable of acting both in the demonstration learning method and in the imitation method.