Abstract: When valve core workpieces are stacked in disordered order,it is difficult for the robotic arm to automatically grasp the upper exposed valve cores.A point cloud segmentation algorithm based on surface curvature weighted fitting was proposed by processing the point cloud information collected by binocular camera. First,filter downsampling was performed and the redundant planar point cloud was removed using the Random Sample Consensus (RANSAC) algorithm,and the noise points were removed by clustering. Then,in order to solve the problem of sticking points at the edge of valve cores,the local plane was fitted using the least squares method and the dynamic curvature threshold was obtained based on the weighted average of surface curvature to remove the sticking points between the adjacent valve cores. Finally,the bottom area of the Oriented Bounding Box (OBB) of the valve core was calculated by the principal component analysis algorithm,and the interval test was performed to realize the iterative Euclidean distance threshold,so as to accurately segment the valve cores to be grasped by clustering. The experimental results show that the accuracy of valve core point cloud segmentation is over 95 %,the center of mass of the segmented valve core is obtained,and the grasping position is determined by combining with the linear fitting algorithm,and the Dobot robotic arm is able to accurately recognize and grasp the valve core through hand-eye calibration with the binocular camera.

Key words: valve core grasping, point cloud segmentation, RANSAC, curvature estimation, OBB, hand-eye calibration