Output-feedback robust control of systems with uncertain dynamics via data-driven policy learning

In this article, we develop two online learning techniques to address the output-feedback robust control problem of systems with uncertain dynamics. For this purpose, an equivalence is first constructed between the robust control problem of the uncertain systems and the optimal control problem of the nominal systems. Then, an output algebraic Riccati equation (OARE) is constructed using its state-feedback control counterpart, which can be adopted to realize the online learning. To obtain the online solution of the OARE, an online policy learning (PL) algorithm based on the state reconstruction (SR) is first presented, where the unknown system internal states can be reconstructed via using the input-output information. To further relax the requirement on the system internal states, a novel online PL method is designed, where only the system output information is required, thus the observer or SR is removed in this online PL algorithm. Simulations are provided to test the developed online learning methods.

Automated summary

In this article, two online learning techniques are developed to address the output-feedback robust control problem of systems with uncertain dynamics. An equivalence is constructed between the robust control problem of uncertain systems and the optimal control problem of nominal systems. An output algebraic Riccati equation (OARE) is constructed using its state-feedback control counterpart, and online learning is realized through this equation. An online policy learning algorithm based on state reconstruction is presented to obtain the online solution of the OARE, and a novel online PL method is designed to relax the requirement on system internal states.