A neural autopilot for bank-to-turn control

A neural autopilot has been developed that compares favourably with a standard gain scheduled autopilot. Whilst there is no distinct improvement shown in the set of step functions, the frequency responses of the neural autopilot are much more consistent throughout the flight range. One advantage of the neural autopilot over that of the gain scheduled autopilot is that there is no set control law for the neural autopilot to follow to obtain the four controller gains. Once the cost weighting parameter has been set in the LQ optimal control case, there is only one set of possible control gains to match this value. The benefit of the neural control law is that these gains are allowed to vary independently of one another