Data-Driven Pilot Behavior Modeling Applied to an Aircraft Offset Landing Task

This paper shows studies for the development of a mathematical model that adequately represents a pilot behavior in the specific task of offset landing, using data-driven modeling techniques. Flight test data was used for the identification procedure. Considerations on the pilot's cognitive process and mathematical modeling possibilities were discussed to select the most appropriate inputs and outputs for the model. This data was used to identify the model using artificial neural network techniques. The models obtained were validated against the identification data and different data not used in the training process to evaluate the quality of the models. Conclusions include the difficulties of showing the generalization capabilities of those non-linear models and further studies.