INVESTIGATION INTO A LOW COST ROBOTIC VISION SYSTEM

Abstract

Industrial robots are of great economic and technological importance within the manufacturing industry due to the wide range of benefits that they provide over their human counterparts.In recent times, technological developments driven by industry demands have led to the development of more autonomous robots.However, the increased strain on production cycles and the demand for higher quality products outlines the need for these robots to be more flexible and adaptable.Existing robot vision systems are generally very expensive (A top of the range industrial vision system from Bosch is priced at $120,000) and would be unaffordable to many small to medium sized businesses.This has created a further need for a more cost effective option which encompasses both the heightened productivity of an industrial robot with the adequate amount of dexterity required to complete its tasks.This paper presents an investigation into a low-cost solution for real-time pick & place operations.A modular approach was taken to complete this model in which it was broken down and its constituent partssensing/perceptual capabilities, 3D image processing/manipulation and robotic manipulatorswere connected using the Robot Operating System (ROS), a comprehensive tool used to provide an extensible inter-process communication framework.The sensory capabilities were provided by a depth sensor, in the form of a Microsoft Kinect; which utilised the OpenNI drivers and Point Cloud Libraries to perform segmentation and clustering techniques.These techniques enabled the system to detect and isolate individual objects from 3D point clouds.Further to this, motion planning algorithms were developed within the ROS Industrial platform in order to drive an ABB IRB 120 robotic manipulator and a servo-electric end effector was incorporated within the system to provide the ability of grasping the objects to perform real time pick and place operations on the objects outlined by the Kinect.