A high-accuracy positioning method for mobile robotic grasping with monocular vision and long-distance deviation

Abstract

It is difficult for mobile robot to position and grasp large-sized objects accurately because the vision acquisition camera deviates from the rotation axis of the robot with a long distance, which leads to the magnification of the calibration error. A method of identifying target objects by positioning marked targets with monocular vision was developed. The marked target was placed in a fixed position within the field of view, which was no longer limited by the specific size of the target object. And the corresponding relation between the displacement of the robot relative coordinate system and the image coordinate system was established. Based on the surface fitting method of Gaussian process regression (GPR), the error compensation model was constructed to effectively compensate and correct the operating errors of the robot once again, so that the accuracy of the error compensation model can meet the goal of accurate positioning and grasping for large-sized objects. Specifically, the error data of each positioning was collected by comparing the real value of the robot's positioning with the estimated value of vision, and the error distribution model of the whole field of vision can be obtained. More than 1000 times of printed circuit board (PCB) board accurate loading and unloading tests were carried out with PCB board moving rack and mobile robot operating platform. The results showed that it could achieve precise visual positioning, which met the industrial application requirements for the mobile robot operating platform.