Planar Robot Grasping Planning Based on Symmetry

Abstract

One of the main challenges in robotic manipulation is to decide how to grasp objects. Symmetry has great significance for characterizing planar shapes for recognition, handling etc. Most of the existing methods use an approach suited for a class of shapes. In the proposed method the axes of symmetry are simultaneously searched using two approaches: as eigenvectors of the moment of inertia covariance matrix of the object and as directions that minimize a symmetry index of the contour of the object in polar coordinates. The best candidate is selected by a symmetry measure in Cartesian coordinates. The method was tested on various shapes, and then implemented in a robotic system with artificial vision. It proved to be efficient even in the case of slightly asymmetric shapes. The proposed method proved to be well suited for the symmetric objects usually grasped by the robots. The implementation must be adjusted for different classes of objects and lightning conditions using two threshold values.