Monocular Vision Based Grasping Approach for a Mobile Manipulator

Abstract

In recent years, mobile manipulators have been widely used in industry and services due to their flexibility and efficiency. However, as the key tasks of the mobile manipulators grasping, object recognition and localization in unstructured environments are still challenging. In this paper, a monocular vision based grasping approach for a mobile manipulator is proposed. This method obtains the optimal grasping pose of the robot arm by locating the marker, thereby avoiding the restrictions on the shape and texture of the object, simplifying the complexity and improving locating accuracy. Our method has three main contributions. First, we calculate the optimal grasping pose of the robot arm by locating the markers and presetting grasping position. Further, we divide the grasping task into 2D plane grasping and 3D grasping, and establish a calculation model for each part. Finally, we propose a method to improve the accuracy. The experimental results show that the 3D grasping error is less than 4mm and the 2D plane grasping error is less than 1mm.