Robust Adaptive Dynamic Programming

Abstract

This chapter introduces a new concept of robust adaptive dynamic programming (RADP), a natural extension of ADP to uncertain dynamic systems. It presents an online learning strategy for the design of robust adaptive suboptimal controllers that globally asymptotically stabilize the system. The chapter introduces the robust redesign technique to achieve RADP for nonlinear systems. To begin with, it considers the nonlinear system with dynamic uncertainties. The RADP methodologies can be viewed as natural extensions of ADP to dynamically perturbed uncertain systems. The RADP framework decomposes the uncertain environment into two parts: the reduced‐order system (ideal environment) with known system order and fully accessible state, and the dynamic uncertainties, with unknown system order and dynamics, interacting with the ideal environment. The presence of dynamic uncertainty gives rise to interconnected systems for which the controller design and robustness analysis become technically challenging.