A Real-time Robotic Calibration Method for Vision-based Bin-picking

Abstract

This research proposed an automatic approach for robotic calibration while the robot is completing the bin-picking task. Since the performance of vision-based bin-picking task can be affected by calibration inaccuracies or weak scene processing, we proposed a method to do calibration during the bin-picking task in real time. First, we use fast Graspability measures to predict the two-finger gripper grasp pose on acquisitions captured from a 3D depth sensor. We use another camera to get visual feedback of robot gripper after every grasp execution. Then, we track the differences between actual grasp pose and detected grasp pose. By updating calibration matrix of the robot system in real time, we increase the success rate in bin-picking task e ciently.