Learning air traffic controller strategies with demonstration-based and physiological feedback

Abstract

In this research, we demonstrate an Artificial Intelligence framework that is able to learn conflict resolution strategies from human Air Traffic Controllers and then employ such knowledge in developing conflict resolution advisories. The proposed framework is designed to assist reinforcement learning, using conflict resolution actions and brain signals. By involving human-in-the-loop in the training, the Artificial Intelligence framework is expected to generate conflict resolution advisories with high acceptability. Our preliminary results have shown the ability of our framework in learning Air Traffic Controllers' strategy and providing human-like resolutions.