Towards a Neural Control Artificial Pilot

In this paper we will address our efforts to design a neural control system that can control a simulated aircraft, which ultimately should lead to a realistic artificial pilot. The system we designed consists of a flight plan module, the actual neural controller module, and a graphic user interface. The goal of the flight plan module is to manage the global control of the whole system. For the neural controller we chose to use a Forward Modeling and Inverse Controller. The Jordan Network was used to construct the pre-trained identifier and the online learning controller. Our first experiments showed that improvements were necessary to make the aircraft fly more smoothly. Although the aircraft “wobbles” a bit at the start of a new flight procedure, the controller is able to adapt to changing circumstances during flight.