Reinforcement Learning and Disturbance Observer Based Optimal Control for Uncertain Systems

This paper investigates the robust adaptive optimal control problem for linear systems in the presence of matching uncertainties. A nominal controller utilizing Q-Iearning algorithm is designed for the nominal linear system without uncertainties. Introducing a disturbance observer serves the purpose of actively estimating uncertainties, enabling proactive compensation for matching uncertainties. The analysis of the closed-loop system's stability under the robust adaptive optimal controller is conducted, presenting sufficient conditions to ensure its stability. Finally, the control algorithm is validated using a two-wheeled mobile robot as the experimental platform.