Partial view geometric reconstruction of objects with rotational and planar symmetries for grasping tasks

In this paper we present a method based on two steps for reconstruction rotational and planar symmetric objects from a single partial view for grasping tasks. In the first step the orientation of the object is determined and in the second step the symmetry axis position is found. Then shape parameters are recovered and the completed geometry is used to plan and execute grasps. We leverage the symmetry information from the reconstruction to plan grasps effectively, resulting in high ratio of good grasps vs. tested grasps. We evaluate our methods on a robot platform that can reliably perform the grasps. Also, a shape characterization is provided for which the first step of the method (finding an object´s orientation) will work, giving a starting point for more general shape reconstruction methods.