Quadcopter Trajectory Tracking Control Based on Flatness Model Predictive Control and Neural Network

Abstract

In this paper, a novel control architecture is proposed in which FMPC couples feedback from model predictive control with feedforward linearization. The proposed approach has the computational advantage of only requiring a convex quadratic program to be solved instead of a nonlinear program. Feedforward linearization aims to overcome the robustness issues of feedback linearization, which may be the result of parametric model uncertainty leading to inexact pole-zero cancellation. A DenseNet was trained to learn the inverse dynamics of the system, and it was used to adjust the desired path input for FMPC. Through experiments using quadcopter, we also demonstrated improved trajectory tracking performance compared to that of the PD, FMPC, and FMPC+DNN approaches. The root mean square (RMS) error was used to evaluate the performance of the above four methods. The results demonstrate that the proposed method is effective.