In this paper, neural architectures for controlling non-linear plants with
parameter variation are proposed. In the first part of the document, the co
ncept of specialized learning over an operation region is considered in ord
er to identify the inverse dynamics of a given plant. Some aspects concerni
ng discretization and invertibility of continuous-time plants are also addr
essed. In the second part of this work, a control architecture which combin
es the former approach of inverse identification through specialised learni
ng with a multiple model scheme is presented. Finally, simulation results a
re discussed, evaluating the performance of the proposed schemes; specifica
lly, the presented controllers are applied to the simulation of the control
of a robot arm. Copyright (C) 1999 John Wiley & Sons, Ltd.