An application of artificial neural networks in adaptive helicopter hover training of novice student pilots is introduced. It was hypothesized that novices can be trained to fly a helicopter system automatically if the helicopter system adapts to the learning curve of the student. Adaptive neuro-controllers, together with a critic model, are used to implement the system. Two different techniques are based on this approach. In one, the helicopter system actively enforces optimality by augmenting the novice's control inputs by amounts necessary to satisfy desired performance criteria. The other uses relaxed performance criteria that are not initially optimal, but appraoch optimality in a graded fashion, based on the learning curve of the student. For both techniques, a simulated student helper adapted to the student pilot's learning curve, providing the amount of augmentation necessary to satisfy system optimality criteria.
Abstract