Quadcopter Attitude Control using Nonlinear MPC and RBF Neural Networks

Quadcopter attitude control is considered a challenging task, due to the inherent nonlinearities and various factors that affect system behavior. In this work, we develop a data-driven nonlinear model predictive control (MPC) framework that can successfully address these challenges. To achieve this, the proposed scheme incorporates radial basis function (RBF) neural network models for predicting the quadcopter orientation dynamics; in contrast to first principles-based predictive models, which are prone to modeling uncertainties, the data-driven nature of RBF models helps them to capture phenomena like aerodynamic effects, thus, leading to more accurate quadcopter control. The resulting controller is evaluated on a simulated quadcopter, while a comparison to a nonlinear MPC scheme using first principles-based predictive models validates the superiority of the proposed approach.