Introduction

Air traffic control is a complex task requiring real-time planning under uncertainty, predicting potential conflicts and issuing commands to aircraft pilots to ensure safety. This project investigates machine learning methods that can be applied to this domain and could in future help air traffic controllers in effective decision-making. Using simulations similar to training scenarios for actual air traffic controllers, the project builds an open experimentation platform to evaluate possible machine learning approaches to this task, and explores algorithms to 'play' the simulation in the role of air traffic controllers.

Header image: Inside Stansted air traffic control tower, courtesy of NATS Press Office.

Explaining the science

Air traffic control is a complex task requiring a large number of safety-critical decisions in real time. The primary goal of air traffic control is to specify the trajectory of each aircraft in order to maintain a safe distance between them at all times, avoiding so-called 'loss of separation', while getting them to their destinations.

In addition to maintaining safe operation, the quality of an aircraft routing is judged against a number of secondary metrics, notably: fuel efficiency, the number and frequency of control instructions, environmental impact and orderly handover between sectors. This process requires looking out into the future, predicting potential conflicts and issuing instructions to the aircraft to reroute, typically by giving a target heading, or routing, or height, or a change of speed.

There are many sources of uncertainty inherent to the problem: from aircraft mass and individual pilot behaviour, to local and regional weather conditions. This makes air traffic control an interesting problem for machine learning research: any decision-making system applied in this area has to account for these uncertainties while optimising for the other objectives. The project will explore a range of possible approaches for this task.

UK airspace graphic
UK airspace as it appears today. Image courtesy of NATS Press Office.

Project aims

Flight movements above the UK are at a historic high and are forecast to increase by a further 50% over the next 15 years. Accordingly, the advancement of air traffic control is one of the key future challenges to providing robust and resilient transport infrastructure for the UK. The project is a collaboration between the Turing and NATS, formerly National Air Traffic Services, which provides air traffic navigation to aircraft flying through UK controlled airspace. 

The main aim of the project is to investigate how machine learning can be usefully applied in this area. Specifically, the project will build automated agents to play the role of an air traffic controller in a training simulation, whose performance will be evaluated according to the same criteria used to judge their human counterparts.

A key part of the project is to develop a research-focused open source simulation platform, along with a user-friendly interface for evaluating different machine learning algorithms for real-time decision-making, from optimisation approaches to reinforcement learning, in a complex and uncertain environment.

Applications

The main aim of the project is to help NATS, the UK's air traffic control service provider, explore the possibilities and limitations of applying machine learning in en-route air traffic control. The results of this project will guide future work in this area, with the potential to improve the capacity and safety of the UK's airspace. By building an automated system that would play an air traffic controller in a simulation the work will explore possibilities and gain insights that will lead to new and better tools and airspace design for air traffic controllers.

It will also lay the foundations of effective collaboration to investigate how machine learning may benefit a broad range of air traffic challenges, including airport operations and facilitating seamless passenger journeys.

Organisers

Researchers and collaborators

Contact info

[email protected]

Funders