Learning uncertainty parameters for assistance in conflict resolution

Helping Air Traffic Controllers (ATCOs) to solve conflicts is challenging because ATCOs only have a partial control on pilots reaction time and trajectory change, and cannot estimate very precisely the aircraft speed. A previous research [1] proposed a method to estimate ATCOs’ uncertainty margins during their deconfliction task. It was shown that, given a predefined uncertainty model, it is possible to learn uncertainty parameters on two-aircraft exercises resolved by an automatic solver. In this article, we collect new data on a more realistic simulator showing a Singaporean En-Route sector and estimate individual and collective uncertainties. These uncertainties are then used in the automatic solver and the resolutions are compared to the actual maneuvers given by the ATCOs. Results on 6 ATCOs who performed several hours of control show that common uncertainties could be estimated with an error of the same range as individual uncertainties. When these uncertainties are used in the automatic solver the solutions are conform to the ATCOs decisions 77 per cent of the time which is 15 percent higher than without considering uncertainties.