2-Stage vision system for robotic handling of flexible objects

Abstract

Machine vision has been extensively studied during the last decades enabling the implementation of vision systems in robotic applications usually involving dispersing of material or handling of rigid objects. Nevertheless, the manipulation of deformable parts remains an open challenge since it involves numerous potential configurations of the object increasing the training effort of the algorithms. To this end, a simple and low-cost solution employing a thresholding contour and depth-based detection method of flexible objects is proposed. This algorithm calculates the coordinates of a grasping point near an edge of the object along with the required orientation of the robot's gripper, leading to effective pick-and-place operations. The suggested approach has been implemented in 2 stages utilizing a 2D vision system. The testing of the method was performed in a case study coming from the household appliances manufacturing sector, where the efficient robotic grasping of a dishwasher gasket with mirror symmetrical edges was achieved successfully.