Grasping Unknown 3D Objects using the Hopfield-RBF-Hopfield (HRH) Approach

L.M. Pedro and G.A.P. Caurin (Brazil)


works, esh simplification


This work presents an approach for the determination of grasping points on the surface of unknown three dimensional (3D) objects for robotic hands and grippers. The novel approach is based on the combination of three artificial neural networks (ANN) that substitute the conventional methods used to accomplish grasping determination. The proposed ANN algorithm aims to allow the determine of 3D grasping points for unknown objects without the necessity of an extensive model data base. Three different ANNs compose the approach: 1) The first network, a Hop field competitive network, is responsible to simplify the triangular mesh that represents the object surface. Mesh simplification reduces the object representation; 2) The second ANN, a radial basis function (RBF) network, classifies the simplified object with respect to previously trained basic 3D shapes. 3) Finally, another Hopfield network, optimizes the Q∞ grasp quality from the grasping points determined with the RBF network based on. The last Hopfield net work finds optimal force-closure grasps. Simulations with 11700 different 3D objects show the algorithm capacity of determining grasping points for 3D unknown objects which performance is close to 94%.

Important Links:

Go Back