THE APPLICATION OF NEURAL NETWORKS TO DRONE CONTROL

Abstract

The FAA Sponsored a six months research program to investigate the application of neural networks to controlling aircraft. It was found that properly configured neural networks offer powerful new computationally robust methods to generate command vectors corresponding to collision free routes. Methods using neural networks which capture the expertise employed by controllers in resolving conflicts were formed. This paper shows that many of the neural network techniques applied to ATC can also be applied to drone control. Two different networks are presented: a multi-layer feed-forward network using back-propagation and a method using a potential field where a gradient measure is employed to maintain the aircraft separation in real time.