M. S. Thesis Abstract
A Practical Vision-Guided Part-Feeding Algorithm for Flexible Manufacturing Automation
One area which is critical to the development of a flexible automated manufacturing system is part-feeding. Many industrial operations currently use robots to feed parts, but these systems often lack flexibility. As an example, a typical system may involve pre-locating parts onto a high precision pallet such that the robot can move to a taught location and pick up the part.
This research attempts to east the requirement of precision a-priori knowledge of part location by incorporating sensing capabilities. The research involves the development of a robotic part-feeding system based on retro-reflective vision sensing. The sensing technique provides high contrast between the part and its environment. The images produced are two dimensional silhouettes which are processed using simple and efficient algorithms. Three dimensional data is collected by constraining the part to lie in a known plane. The six parameters to determine the position and orientation of the part are derived from the two dimensional image by employing the results of off-line calibrations. Schemes whereby a human operator teaches the system how to grip different parts are developed. This allows the system to pick up parts of generic shape.
The algorithms have been successfully integrated into a system comprised of a Puma 760 robot with a camera mounted on the end effector. The system is capable of locating and grasping parts of a known type from an arbitrary position and orientation within a plane. The system is evaluated with respect to its flexibility, reliability, accuracy, and speed.