Learning for Air Traffic Management: guidelines for future AI systems

Abstract

Abstract The SESAR-funded Modern ATM via Human / Automation Learning Optimisation (MAHALO) project recently completed two years of technical work exploring the human performance impacts of AI and Machine Learning (ML), as applied to enroute ATC conflict detection and resolution (CD&R). It first developed a hybrid ML CD&R capability, along with a realtime simulation platform and experimental User Interface. After a series of development trials, the project culminated in a pair of field studies (i.e., human-in-the-loop trials) across two EU countries, with a total of 35 operational air traffic controllers. In each of these two field studies, controller behaviour was first captured in a pre-test phase, and used to train the ML system. Subsequent main experiment trials then experimentally manipulated within controllers both Conformance (as either a personalised-, group average-, or optimized model) and Transparency (as ether a baseline vector depiction, an enhanced graphical diagram, or a diagram-plus-text presentation). The proposed paper presents guidelines on the design and implementation of ML systems in Air Traffic Control, derived from the results and lesson learned from the Simulations, as well as the qualitative feedback received from the controllers themselves.