MAHALO: Modern ATM via Human/Automation Learning Optimisation

MAHALO investigates the relationship and impact between the conformance and transparency of a conflict detection and resolution automation support system on air traffic controller understanding, acceptance, trust, and performance. MAHALO intends to answer a simple but profound question: should we develop automation that solve problems like the individual human (conformal), or should we develop automation that explain itself to the human (transparency). The project has developed two conflict detection and resolution support system based on machine learning (ML) methods. One of the models builds on Supervised Learning methods where the objective is to generate solution advisories that match a groups of air traffic controllers’ preferred solutions. The other model builds on Reinforcement Learning methods where the objective is to generate optimized solution advisories. In addition, solution advisories that match individual controllers’ solution preferences have been derived through conventional data analysis methods. The system has been empirically tested in Italy and Sweden with 36 air traffic controllers in human-in-the-loop simulations. The simulations have varied solution conformance (individual, group, optimal) and system transparency (control, domain transparency, solution transparency) to study controllers’ trust and acceptance of solutions and system understanding and performance.

MAHALO is an exploratory research project within the Horizon 2020 research and innovation program, funded by the SESAR Joint Undertaking.