Robotic Grasping Pose Estimation based on Point Cloud accelerated by image feature correspondence

Abstract

The pose estimation of target objects based on point cloud information is one of the mainstream schemes for robot grasping at present. However, due to the large amount of point cloud, the traditional pose estimation method based on point cloud usually takes a long time to calculate, which cannot meet the real-time requirements of robot control. To solve this problem, based on the high speed and robustness TEASER++ algorithm, we propose a new method for fast registration of point clouds by taking advantage of the correspondence between point cloud data and image feature points and the efficiency of image feature matching, which greatly improves the speed of pose estimation. Finally, the proposed method is evaluated by executing real grasping tasks using the position-based visual servo method, which shows the efficiency and robustness of the pose estimation method.