This paper presents a systematic method to design robust multivariable controllers from experimental data so as to ensure that the controller is robust with respect to all plants which cannot be discounted based on the data (to within a specified statistical confidence). Estimation of a multivariable "plant set" rather than a point estimate produces uncertainty bounds in a form which can be incorporated systematically into robust control formulations. The present paper extends previous multivariable plant set estimation results to the structured uncertainty case and demonstrates the benefit to control synthesis of estimating uncertainty bounds in structured from. This benefit is illustrated by example on a bench-mark two-car problem.
展开▼