Nonlinear neuro-predictive control of a miniature coaxial helicopter

In this paper the design of a nonlinear predictive controller based on a neural network model for a coaxial miniature helicopter is presented. The first step is the development of neural network models for all four movements of the helicopter. Recurrent dynamic networks are trained to predict accurately the behaviour of the helicopter. Further, this neural models are used for prediction in order to design nonlinear predictive controllers for yaw, altitude, pitch and roll control. When computing the optimal control signals, contraints are considered using a quadratic programming optimization technique.