A model-based approach for visual guided grasping with uncalibrated system components

Abstract

In this paper, a complete framework for the task of grasping a naturally textured object is being presented. The lack of calibration in the components of the eye-in-hand system is overcome by an image trajectory based visual servoing scheme (ITBVS). Using an object model reconstructed off-line, consecutive rough pose estimation and refinement of the estimate by scale optimized 3D model to object registration solves the feature correspondence problem in the absence of artificial landmarks. Experimental results using a zoom camera and a 7-DOF serial manipulator prove the effectiveness of the approach.