Application of neural networks on robot grippers

Abstract

A new-generation general-purpose robot gripper system which applies an artificial neural network to guide a three-finger gripper has been designed. The simulation of the core part of the whole system, i.e. optimally placing three fingers for a stable grasp using the Hopfield net, has been conducted. The results obtained show that this scheme behaves in a promising fashion. The actual computation time is usually within several seconds if implemented in an analog neural net, making the real application attractive