Neuromorphic controllers

Abstract

Dcpartmsnt OfmEtriml Engineering"het This paper investigates the feasibility of utilizing neural Aa such networks for the design of learning controllers. controllers, the design of both regdators and trackers are considered. M o m r , the relationship of these learning controllera to conventional adaptive controllers are invwtiated. This enables one to develop a learning meth &logy fo r these networks which guarantees the stability of the tmining pmce# under mild conditions, and reqaires "alinformationabouttheplantdynamim. Introdu.c;finaThe design of learnin controllers for dynamic q"s is a my impmtant and chdnging research area [l-21. Despite many sophtticated control strategies, the pressing need for high precision and impxwed performance still raises the important question of how to design a generic control architecture for a givm srpstem which is capable of learning the mapping from the sgpte"r desired output to its corresponding reqaind input Realizing the learning