Structured learning in feedforward neural networks with application to robot trajectory control

Abstract

The authors propose a method for structured learning in feedforward neural networks (FFNNs) which results in improved generalization properties and significantly faster training times for the task of controlling the motion of a two-link robotic manipulator over a desired trajectory. They use a control system configuration consisting of a conventional feedback controller and a neural network configured as a feedforward controller. The authors compare the performance of the structured neural network (SNN) to a standard FFNN and also to the cerebellar model articulation controller (CMAC). Through computer simulations, they establish that SNN gives excellent results, outperforming both FFNN and CMAC.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>