A neural network for tactile sensing: the Hertzian contact problem

Abstract

A neural network has been developed for use with local closed-loop gripper control using a tactile sensor array. The specific task considered is to solve part of the inverse Hertzian contact problem. Backpropagation is used to train a single hidden-layer network to recognize the angle of contact between a cylindrical finger with an embedded sensor array and a cylindrical rod. Simulation results are presented and discussed.<>