Image-based visual servoing for grasping unknown objects

Abstract

In this paper, we present an image-based visual servoing approach to the problem of grasping unknown objects by adopting visual vectors. A CCD camera mounted on the robot md-effector constructs the visually guided servo control system and the control scheme lends itself to task-level specification of manipulation goals by the visual vectors. The grasping task is to control the robot so that the vectors of the end-effector's finger and of a target object's grasp coincide. These vectors can be used to represent the work of a stable grasping of an object that is presented in an unstructured manner. These visual vectors in image frame are obtained by analyzing the object's image and projection. The objective of vector processing is to estimate the error between the finger and grasp vectors, and to control the robot in order to eliminate kinematic errors.