Expert hierarchical adaptive controller for robotic systems

Abstract

An expert hybrid control scheme which does not use explicit parameter estimation, is presented. It adjusts its structure in order to provide the 'best' control for a broad range of plant conditions. The coarsest control, provided by a multiple-model adaptive control procedure, provides the initial level of control by forcing the plant's response to exist within a predetermined allowable range of specifications. This controller is also used to determine the adaptation coefficients and the initial values of the adaptation gains for the high-level controller. The finest control is determined by a model reference adaptive controller that causes the plant to follow a desired reference model. The heuristics used to determine switching between the two controllers is orchestrated by an expert system that bases its decisions on plant responses and controller parameter values. >