Gradient Descent Algorithm Based Backstepping NN Control of a Perturbed Quad-Rotor UAV: Theory and Experiments

This paper aims at using neural network (NN) based control method to solve the flight control problem of a perturbed quad-rotor unmanned aerial vehicle (UAV). By combining backstepping technique with NNs, the proposed control method can deal with the position and attitude tracking control problems of a perturbed real UAV. The radial basis function NNs with gradient descent algorithm are first introduced to deal with the UAV system's unknown complex nonlinearities caused by the external disturbances. By resorting to Lyapunov stability criteria, it is obtained that the tracking errors of the quad-rotor UAV are practically stable. Compared with cascade proportional-integral-derivative control method, the frequently-used control method for real UAVs, the proposed control scheme can achieve higher position and attitude tracking performance. Finally, experimental results on a real quad-rotor UAV are demonstrated to verify the effectiveness of the proposed control method.