• 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
 
 
Doctoral Thesis
DOI
10.11606/T.18.2010.tde-25012011-120324
Document
Author
Full name
Valdinei Luis Belini
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 2010
Supervisor
Committee
Caurin, Glauco Augusto de Paula (President)
Becker, Marcelo
Kato, Edilson Reis Rodrigues
Romero, Roseli Aparecida Francelin
Wolf, Denis Fernando
Title in Portuguese
Manipulação de objetos orientada pela aplicação: abordagem assistida por uma rede neural artificial
Keywords in Portuguese
Manipulação
Robótica
Tarefa
Abstract in Portuguese
Motivado pelo fato de que a escolha correta da forma preênsil para mãos robóticas antropomórficas representa aspectos relevantes no planejamento de estratégias de manipulação de objetos, este trabalho: desenvolve um algoritmo de pegas de objetos compatíveis com a tarefa a ser realizada e um método computacional de ajuste de superelipsóides a superfícies de objetos sintéticos 3D para modelar as primitivas resultantes do estágio de segmentação manual do objeto. Essa representação volumétrica contempla as atrativas propriedades de um modelo compacto, controlável e intuitivo, capaz de modelar uma ampla variedade de geometrias simétricas básicas a partir de 11 parâmetros. Tendo em vista que a acomodação de superelipsóides a nuvens de pontos caracteriza desafios típicos de mínimos quadrados, um breve estudo sobre o método numérico de Levenberg-Marquardt, utilizado na minimização da função objetivo adotada, se mostrou necessário para maximizar a sua eficiência. O desempenho e a eficiência do programa computacional desenvolvido foram demonstrados na modelagem experimental de cinco primitivas organizadas em diferentes configurações. As pequenas distorções, consideradas aceitáveis para a proposta do trabalho, evidenciaram a robustez da representação de formas elementares empregando superelipsóides sem deformações. Encerrada a etapa de modelagem das primitivas, um método de auto-aprendizagem supervisionada interpreta as informações de forma e tamanho de tais formas básicas para sugerir a pega de uma das primitivas que compõem o objeto e a forma preênsil compatível com a tarefa desejada.
Title in English
Object grasping oriented by the application: an approach assisted by an artificial neural network
Keywords in English
Manipulation
Robot
Task
Abstract in English
Motivated by the fact that the correct choice of prehensile form for anthropomorphic robotic hand describes relevant aspects in planning strategies of object manipulation, this work develops an algorithm of grasping objects to match the task requirements and a computational method capable of fitting superellipsoids to surfaces of synthetic 3D objects to model the resulting primitives from the manual segmentation stage of the object. This volumetric representation describes the attractive properties of a compact, controllable and intuitive model, which is capable of modeling a wide variety of standard geometric solids by using eleven parameters. Since the superellipsoid fitting process characterizes typical least squares problems, a brief review about Levenberg-Marquardt numerical method, employed for the adopted objective function minimization, was necessary to maximize its efficiency. The performance and efficiency of the developed program were demonstrated in the experimental modeling of five symmetrical primitives shapes organized in several configurations. The small distortions, considered acceptable to the work proposal, confirm the robustness of representing elementary shapes employing superellipsoids without deformations. After finishing the primitive modeling stage, a supervised learning method interprets the size and shape information of such basic shapes to suggest the grasp of one constituent single parts and the task-compatible prehensile form.
 
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
2011-04-13
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
Centro de Informática de São Carlos
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2020. All rights reserved.