Design of neural network and backstepping based adaptive flight controller for multi-effector UAV

Research of modern flight control methods for unmanned aerial vehicles (UAVs), which lower the cost and risk associated with the design of actual physical flight systems, has become research hotspot in recent years. However, design of control laws for UAV is complicated and challengeable due to UAV uncertainties, nonlinearity and coupling. This paper proposes a new hybrid controller design scheme, which consists of a backstepping controller and a neural network compensator. The backstepping controller realizes linearization and decoupling of the highly nonlinear and tightly coupled UAV model. For cancelling out uncertainties such as unmodeled dynamics and external disturbances, the neural network compensator is designed to enhance the robustness of flight system. Pseudoinverse method is applied to establish the mapping between moments and multiple control surfaces. Numerical simulation shows that the UAV equipped the hybrid controller has good maneuverability, strong self-learning ability of compensating the unmodeled dynamics and enough robust stability against constraints of actuators.