Robust Autonomous Flight Control of an UAV by Use of Neural Networks

This paper describes a method to develop robust flight control systems for UAVs. It was difficult to develop flight control systems, because the helicopter dynamics is nonlinear. Moreover the flight environment is not fixed because of the atmospheric changes, such as the wind. The wind affects the attitude or velocities of the UAV, but the wind speed or direction is hard to predict, so the wind is usually categorized into stochastic uncertainties. An efficient method to design robust controllers by training neural networks is proposed in this paper. Neural networks trained by the proposed method are robust against stochastic uncertainties. In this paper, the small unmanned helicopter is focused on, and numerical results of altitude control are shown to demonstrate the effectiveness of our approach.