Gaussian Process Regression-based Disturbance Compensation Control for Urban Air Mobility

Abstract

This paper deals with a nonlinear attitude controller considering the disturbance rejection for urban air mobility (UAM). Using the disturbance observer-based control (DOBC) methodology, the proposed controller is constructed with a two-stage design procedure. The baseline control is established first by employing the time-scale separation approximation assumption and the feedback linearization approach, and then the Gaussian process regression (GPR) is augmented. Given the computational burden, the GP model is learned offline with a fixed-size training dataset. The GPR works as an adaptive law like the nonlinear disturbance observer. However, GPR can flexibly model the disturbance because the GPR describes the disturbance as a distribution over the functions. Furthermore, the control allocation method for the over-actuated system is presented to distribute the control command efficiently. Consequently, the proposed controller is validated with the numerical simulation under the various disturbance conditions such as model parameter uncertainties.