Worst-Case Identification under Binary-Valued Observations

Abstract

This chapter focuses on the identification of systems where the disturbances are formulated in a deterministic framework as unknown but bounded. Different from the previous chapters, here the identification error is measured by the radius of the set that the unknown parameters belong to, which is a worst-case measure of the parameter uncertainties. By considering several different combinations of the disturbances and unmodeled dynamics, a number of fundamental issues are studied in detail: When only binaryvalued observations are available, how accurately can one identify the parameters of the system? How fast can one reduce uncertainty on model parameters? What are the optimal inputs for fast identification? What is the impact of unmodeled dynamics and disturbances on identification accuracy and time complexity?