Enabling Greener Air Traffic Control

Supervised by Samuel Cohen and Christoph Reisinger, and partnered with the National Air Traffic Service (NATS), the aim of this project is to harness stochastic control and data-scientific techniques to optimise - or at least improve upon - current methods for flight-path generation over Transatlantic flight corridors. Rather than exploring heavy modelling techniques, our project will aim to consider historical flight paths and what perturbations are optimal to improve performance, thus guaranteeing simple, useable instructions for pilots and air traffic controllers, much-needed algorithmic automation in the space, and theoretical guarantees on self-improvement over time.

More explicitly, initial ideas involve utilising deep learning techniques for learning feedback controls via historical flight path data for empirical risk minimisation. Whilst theories exist to guarantee mitigation of overlearning in underparametrised statistical learning regimes, the overparametrised regime within which this project falls requires much theoretical study to bring to application.

Such study involves how we may infer models for underlying data, regularise using entropy, and mean-field techniques for analysis of generalisation error bounds in order to give general (therefore robust) theoretical guarantees for learning techniques involved.

Following such theoretical work, NATS will provide data and work placement for the student. Study will then encompass practical implementation, which will involve ensuring robustness to real-world data problems (such as missing or unclean data), following industry-level coding practices, as well as ensuring efficiency; given the nature of air traffic control, flight path decisions need to be made quickly.

Another aspect of air traffic management is the potential safety risks involved, meaning that the workflow for generating, authenticating and finalising flight paths can be cumbersome, and it is perhaps this characteristic, combined with the scale of the problem, which has caused air traffic management to somewhat resist algorithmic automation. Any practical implementation will have to carefully consider how to adhere to these systems, which represent best ethical practices in the field.

This project broadly falls within the EPSRC research area concerning Statistics & Applied Probability, and ultimately aims to reduce environmental, economic and operational costs. NATS have emphasised that air traffic is clearly projected to increase, so it is imperative that methods are developed now to galvanise against the potentially serious impacts of these costs.