Data-Driven Learning Extended State Observers for Nonlinear Systems: Design, Analysis and Hardware-in-Loop Simulations

Abstract

This paper considers the disturbance/uncertainty estimation of first-order nonlinear system subject to fully unknown internal dynamic, external disturbance, and unknown control input gain. Compared with existing extended state observer (ESO) where priori knowledge of model parameter such as nominal input gain should be known as a priori, reducedand full-order data-driven learning ESOs are developed for estimating the lumped disturbance and control input gain. A salient feature of the proposed data-driven learning ESOs is that the unknown input gain and lumped disturbance can be estimated synchronously, in the meantime, the estimation convergence is guaranteed benefiting from the data-driven approach. Hardware-in-loop simulation is carried out to substantiate the performance of the proposed data-driven learning ESO for surge speed tracking of a robotic marine vehicle without knowing the model parameter in advance.