Grasping in Depth maps of time-of-flight cameras

Abstract

The recognition and localization of objects in space is of fundamental interest in many robot vision applications, especially in those that are supposed to provide services to human beings. The trivial example of any such task is manipulation, i.e., providing a robot the means of handling objects. In this work, we discuss the chances and the problems experienced when using a time-of-flight camera as the only measurement device of an object recognizer. The localization is based on the best-fit of the geometric primitives within the objects considered (such as planes, cylinders, cones, or spheres). Such shapes are of special interest not only in recognition but also in grasp planning. We use a time-of-flight camera SwissRanger SR-3000 and localize a selected set of objects in order to grasp them with a dexterous robotic hand.