Singulation of Irregular Objects by Adaptive Robotics and Sparse Range Imagery

Abstract

A basic problem in automated handling of materials is how to remove a single piece, of any size or shape, from a stream, stack, or heap of unknown objects, one object at a time. This classic “bin-picking” problem, has been the focus of research at GE/RCA Advanced Technology Laboratories (ATL) under a contract sponsored by the Office of Advanced Technology of the US Postal Service. This research has resulted in the development of an approach for picking or singulating irregular mail pieces from a pile using adaptive robotics and sparse range imagery. Heaps of irregular pieces are dispensed onto a carousel from a pair of conveyor belts. Sparse range images of the heaps are generated by a dynamic structured light technique. The use of sparse range data significantly decreases the image processing time required for traditional full image range data techniques. The range images are processed to determine the position and accessibility of individual pieces on the carousel. The centroid coordinates of accessible pieces are computed and passed from the vision system to a robot where a specific object is selected and serviced. The rotating disk provides recirculation of pieces that are not initially removed so that they may be picked up on succeeding revolutions. A continuous process is achieved by injecting additional objects onto the carousel when empty areas are detected by a second camera system. Sensory feedback, such as force, torque, vacuum, and pressure sensors are used to adaptively modify the actions of the robot, making the system inherently jam free.