Tesis doctoral de Abdelmadjid Boudaba
Object grasping plays a key role in robotics. The problem of achieving efficient grasps is clearly of considerable importance in many areas of robotics such as feeding, fixturing, loading, assembly, and manipulation in general, and it is therefore of utmost and direct practical application. While significant progress has been made in the study of the desirable properties necessary to analyze and synthesize a grasp, more research is still needed. Dexterous manipulation requires a high level of sensory data and feedback from the robotic systems to perform grasping and manipulation tasks. As a consequence, current robotic systems have been adapted to fit the need of grasping by using the principle of sensing- planning-action. In order to grasp an object, robotic systems must obtain relevant information from sensors. the sensor information required typically includes contact point estimation, surface normal and curvature measures, and knowledge of both applied and induced forces on the fingers. Despite the advances in research, much of the work in grasping task is based on the use of cad models of the objects to be grasped, so as to predefine a set of candidate grasps, or on simplified mode of control for complex manipulation tasks. In fact, most of the existing grasp planning algorithms suffer from ineffectiveness due to their inability to perform satisfactorily in real world applications. it is worthwhile to discover more general properties of robotic systems, so that they can be capable of taking decisions, interacting with their environment and carrying out tasks successfully. What is desirable is to create and to develop a methodology for planning strategies that perform well on completely unknown objects, that incorporates vision and other sensors that are also useful for the visual guidance of a manipulator, and taking into account the constraints imposed by the working environment. this thesis addresses the problem of vision based planning for grasping unknown objects. The proposed approach does not rely on prior models of the objects to be grasped; the information obtained from vision and from tactile sensors located at the fingertip of the hand will be used as sensor data input for both grasp planning and control of grasping positions. In this approach, the following computational tasks are included: (1) the definition of a basic grasping system framework; (2) the analysis of several methods for object contour description for extracting grasp regions; (3) the development of several algorithms for the selection of the appropriate contact locations of a grasp on planar objects using visual features; (4) the development of a method based on matching tactile and visual features for controlling the finger po- sitions in order to guide the gripper towards the desired positions before the task of grasping is executed. finally, all grasp planning algorithms discussed in this thesis are implementations and their performance is evaluated on a variety of objects.
Datos académicos de la tesis doctoral «Ision-based planning for grasping unknown objects«
- Título de la tesis: Ision-based planning for grasping unknown objects
- Autor: Abdelmadjid Boudaba
- Universidad: Politécnica de catalunya
- Fecha de lectura de la tesis: 09/07/2009
Dirección y tribunal
- Director de la tesis
- Alícia Casals Gelpí
- Tribunal
- Presidente del tribunal: raúl Suárez feijoo
- rachid Alami (vocal)
- heinz Wí¶rn (vocal)
- marc Carreras pérez (vocal)