Adaptation of a Human Resource Model by the Use of Machine Learning Methods as Part of a Military Helicopter Pilot Associate System

Abstract

This article describes an approach how to improve a knowledge-based pilots’ associate system in the domain of military helicopter missions by the use of machine learning methods. To prevent the pilot from being overtaxed the associate system estimates the pilots’ residual mental capacity and thereby the current subjective workload . This estimation enables the associate system to selectively direct automation induced dialogues, e.g. hints, warnings, alerts, suggestions to the perceptual modality, which can be assumed to provide spare resources. Therefore, we developed task related models of mental resource demands for the military helicopter flying domain. To eliminate subjective influences from these models as far as possible, laboratory experiments have been conducted to better match the predicted resource conflicts within distinct task situations with the objectively measured pilots’ performance. Based on these experiments, we applied machine learning methods (i.e. genetic algorithms) to adapt the underlying human resource model to the measured human performance. By using suchlike models the associate system is enabled to cooperate with the pilot by resource adaptive information exchange. This article focuses on a specific aspect of the overall associate system related trials. We provide a detailed description of the conducted experiments used for adaptation of the resource model and the application of the machine learning technique for the model optimization as well as detailed results of the overall evaluation of thee associate system’s adaptive capabilities in a relevant mission context obtained in simulator experiments.