Modified Gaussian process regression based adaptive control for quadrotors

Abstract Robustness is crucial for flight vehicle control, which is directly related to control performance and mission security. Adaptive control methods are effective for handling perturbations, but most rely on expert knowledge in choosing parametric estimator structures. These uncertainties in quadrotor dynamics own complex representation form, so the structure of the estimator is difficult to determine. Our work integrates modified Gaussian process regression (GPR) with a quaternion-based command filtered backstepping framework, such that quadrotors subjected to perturbations can rapidly and accurately track the desired trajectory. GPR-based control architecture applies Bayesian nonparametric representation to estimate the distribution of perturbations; the nonparametric estimation method requires no prior knowledge about uncertainties and can inherently manage measurement noises. The zero-mean prior of GPR is modified by using the error integral, which compensates for the estimation bias of constant and/or low-frequency disturbances. Furthermore, to estimate the rapidly changing perturbations, a time-dependent form regarding state point selection is applied to increase the influence of recent measurements. Simulation results demonstrate the effectiveness of these modifications and the superiority of the proposed control system.