The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #92159

Title:Finding Robot Grasping Points from Range Data
Date: 1992
Abstract:This dissertation addresses part of the problem of identifying robot grasping positions on objects. It attempts to find the individual surfaces on an object which could be grasped by a robot gripper. These surfaces must be smooth enough, large enough and of a suitable shape. The system uses range data from a laser striper and then segments the image using an automatic segmentation tool. Each segment is checked for smoothness by assuming that the shape can be described by a biquadratic fit, and any bumps/dents are found. Erosion/expansion is then used to simulate gripper placement over each surface, to find where grasps may be made. Every patch which is large enough is then described and put into one of four classes. These classes are designed to provide the information needed to plan a grasping strategy. Finally, the program rejects patches which are of an unsuitable shape. Erosion/expansion worked well and the biquadratic fitting provided for easy classification. In order for the biquadratic fitting to work well as a smoothness check, it was found that the segmentation method must be changed. It is likely that the surface fitting and segmentation must be combined for the best results.

[Search These Pages] [DAI Home Page] [Comment]