On the design of a neural network autolander

Abstract

A research program directed at the development of an autolander for NASA's X-33 prototype reusable launch vehicle is described. The autolander is based on a new linear quadratic adaptive critic algorithm. It is implemented by an array of Functional Link neural networks and is trained by a modified Levenberg–Marquardt method. A full stability theory is developed for the new adaptive critic algorithm. Simulation results are presented for the linear–quadratic case. Copyright © 1999 John Wiley & Sons, Ltd.