Online Neural-Network Learning and Model Predictive Control Applied to a Tilt-Rotor Unmanned Aerial Vehicle
Gregorio Carughi, Guillaume Ducard, Christopher Harald Onder · 2022
This paper presents an online neural-network (NN) learning algorithm applied to a Model Predictive Control (MPC) controller for a propeller-tilting hybrid vertical take-off and landing (VTOL) unmanned aerial vehicle (UAV). Neural networks are trained online to learn unknown dynamics and disturbances acting on the system to reduce the error between the actual dynamics of the system and the nominal dynamics considered in the MPC algorithm. Online learning is based on two theorems which provide Lyapunov proofs and guarantees of convergence for the tracking error between the desired and the real dynamics. The results presented in this work show that the proposed online-learning algorithm yields significant improvements in the tracking of the desired trajectory, even in the case of a fault in one of the actuators. Simulations prove that this controller is suitable for a complex system such as the tilt-rotor VTOL UAV.