A neural network development environment for adaptive inverse control

Presents a neurocontroller development environment using the ideas of adaptive inverse control. The goal of the paper is to describe the environment utilized and show a very simple example. A PC running a commercial simulator communicates in real-time with a /spl mu/controller to learn how to control the power supply of a motor using the speed as input. This /spl mu/controller runs a feedforward neural network online, but the weights are adapted periodically off-line by the simulator and downloaded once the weights stabilize. The user interacts with the user interface of the commercial package to define the topology, to test the adequacy of the control topology and to set the control parameters.