Vision Strategies for Robotic Manipulation of Natural Objects

Abstract

This paper describes a vision guided robotic platform that will be used for the handling of natural products in a multi-layer stack that is typical in a bin situation. It is designed to automatically recognise non-rigid objects, to estimate their pose, and to select suitable picking points using depth information from a 3D camera. We adopted a novel approach by using a topographical method to segment the objects and then with elevation contours to build volumetric data for each object. When the periphery for each object is established, we introduced strategies by which decisions will be made for task planning. These strategies identify the appropriate object to pick and the optimum picking position. Results of the experimental trials demonstrate the robustness of this approach for specific types of natural products. 1