• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.18.1999.tde-07062024-161643
Document
Author
Full name
Carlos Magno de Oliveira Valente
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 1999
Supervisor
Committee
Caurin, Glauco Augusto de Paula (President)
Araújo, Aluízio Fausto Ribeiro
Mucheroni, Mario Francisco
Title in Portuguese
Fixação de objetos de formato geométrico desconhecido utilizando redes neurais artificiais
Keywords in Portuguese
.
Abstract in Portuguese
Desenvolvimento de uma garra robótica de três dedos, capaz de fixar objetos de formato arbitrário. Para manipular estes objetos, propõe-se um sistema composto por dois estágios: processamento de imagem e cálculo neural dos pontos de contato do objeto. O sistema de visão captura imagens de topo da cena e utiliza o algoritmo do vizinho mais próximo para identificar os pontos que definem o contorno do objeto. No segundo estágio, dois modelos de redes neurais foram implementados para planejar a fixação, definindo os pontos de contato entre a garra e a peça. A primeira rede neural (Rede Competitiva de Hopfield) realiza uma aproximação poligonal sobre o conjunto de pontos de contorno, simplificando a representação deste. O segundo modelo de rede é responsável pelo cálculo efetivo dos três pontos de contato. Diversas configurações de redes Multi-layer Perceptron (MLP) e Redes de Funções de Base Radiais (RBF) foram testadas a fim de definir o método mais adequado. Através desta análise, a rede RBF treinada pelo algoritmo Global Ridge Regression apresentou uma maior qualidade de resposta e um desempenho compatível com aplicações em tempo real
Title in English
.
Keywords in English
.
Abstract in English
This work presents a robot gripper with three fingers which is able to capture objects of arbitrary shape. To handle arbitrary shaped objects, we propose a solution in two stages - image processing and object contact points definitíon. The vision system captures the top image and uses the Nearest-Neighbor Method to define a set of points representing the object outline. In the second stage, two neural network architectures work together selecting three contact points for the gripper on the outline. A first neural network (Competitive Hopfield Network) realizes a polygonal approximation over the sei ofpoints, reducing the number ofpoints to be analyzed. A second supervised neural network computes the three contact points from the approximated polygon. Several configurations of Multi-layer Perceptron (MLP) ana Radial Basís Function (RBF) networks were tested to define the most suitable configuration. Through this analysis, the RBF network trained by the Global Ridge Regression method presented the best response accuracy and a performance compatible with on-line applications.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2024-06-07
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.