A Probabilistic Digital Twin of UK En Route Airspace for Training and Evaluating AI Agents for Air Traffic Control

Nick Pepper, Adam Keane, Amy Hodgkin, Dewi Gould, Edward Henderson, Lynge Lauritsen and Christos Vlahos, George De Ath and Richard Everson, Richard Cannon, Alvaro Sierra Castro , John Korna, Ben Carvell and Marc Thomas · 2026

Read A Probabilistic Digital Twin of UK En Route Airspace for Training and Evaluating AI Agents for Air Traffic Control by Nick Pepper, Adam Keane, Amy Hodgkin, Dewi Gould, Edward Henderson, Lynge Lauritsen and Christos Vlahos, George De Ath and Richard Everson, Richard Cannon, Alvaro Sierra Castro , John Korna, Ben Carvell and Marc Thomas (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-03113.txt.

Introduction

Nick Pepper and thirteen co-authors present a probabilistic digital twin of the United Kingdom's en route airspace, built for the London Area Control Centre. The twin is meant as a safe virtual environment where AI agents for air traffic control can be trained and evaluated, including with human controllers in the loop. It merges historical and live operational data with a physics-informed machine learning model of aircraft performance.

The twin reproduces real traffic scenarios while honestly reflecting the uncertainty that controllers face, and it runs far faster than real time with a standard Python gym interface. A structured assurance case, built on a trustworthy and ethical assurance framework, supplies quantitative evidence for the twin's accuracy and fidelity. The authors present it as the first probabilistic digital twin of operational en route airspace.

Catalog record

CREATORNick Pepper, Adam Keane, Amy Hodgkin, Dewi Gould, Edward Henderson, Lynge Lauritsen and Christos Vlahos, George De Ath and Richard Everson, Richard Cannon, Alvaro Sierra Castro , John Korna, Ben Carvell and Marc Thomas
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-03113

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Probabilistic Digital Twin of UK En Route Airspace for Training and Evaluating AI Agents for Air Traffic Control Nick Pepper, Adam Keane, Amy Hodgkin, Dewi Gould, Edward Henderson, Lynge Lauritsen and Christos Vlahos, George De Ath and Richard Everson, Richard Cannon, Alvaro Sierra Castro , John Korna, Ben Carvell and Marc Thomas Abstract This paper presents the first probabilistic Digital Twin of operational en route airspace, developed for the London Area Control Centre. The Digital Twin is intended to support the development and rigorous human-in-the-loop evaluation of AI agents for Air Traffic Control (ATC), providing a virtual representation of real-world airspace that enables safe exploration of higher levels of ATC automation. This paper makes three significant contributions: firstly, we demonstrate how historical and live operational data may be combined with a probabilistic, physics-informed machine learning model of aircraft performance to reproduce real-world traffic scenarios, while accurately reflecting the level of uncertainty inherent in ATC. Secondly, we develop a structured assurance case, following the Trustworthy and Ethical Assurance framework, to provide quantitative evidence for the Digital Twin’s accuracy and fidelity. This is crucial to building trust in this novel technology within this safety-critical domain. Thirdly, we describe how the Digital Twin forms a unified environment for agent testing and evaluation. This includes fast-time execution (up to x200 real-time), a standardised Python-based “gym” interface that supports a range of AI agent designs, and a suite of quantitative metrics for assessing performance. Crucially, the framework facilitates competency-based assessment of AI agents by qualified Air Traffic Control Officers through a Human Machine Interface. We also outline further applications and future extensions of the Digital Twin architecture. 1 Introduction G rowing demand for air travel, with flight numbers rising steadily each year [ 16 ] , is placing significant pressure on Air Traffic Management (ATM) systems to increase capacity while maintaining rigorous safety standards. This challenge is intensified by the introduction of new airspace users such as drones and electric Vertical Take-Off and Landing (eVTOL) aircraft, as well as by governmental requirements to improve airspace efficiency and reduce emissions (see, e.g., [ 9 ] ). In response, major ATM modernisation programmes [ 19 , 53 ] aim to improve workload forecasting, trajectory prediction accuracy, strategic planning, route optimisation, and overall system resilience [ 54 ] . The scale and urgency of these challenges mean that incremental enhancements to traditional tools and processes are unlikely to be sufficient, and a step change in automation for tactical air traffic control will be required. Within this context, there is growing interest in AI-based decision-support technologies for Air Traffic Control (ATC). Throughout this paper, we use the term “AI agent” to denote any AI-based decision-support technology for tactical ATC, irrespective of its envisioned level of autonomy. Such agents may range from tools that highlight potential conflicts and propose resolutions to more advanced systems with greater degrees of control authority [ 11 , 18 ] . ATC is a safety-critical task, requiring controllers to balance efficiency, orderliness, and other factors in a highly context-specific manner, while also maintaining rigorous safety standards. In consequence, AI agents require extensive and rigorous validation before operational deployment. Automated methods alone, whether formal methods [ 41 ] or the recent validation-through-scenario-generation paradigm that has emerged in autonomous driving [ 6 , 10 ] are insufficient, as they cannot capture the situational judgements made by trained controllers. Human-in-the-Loop (HITL) assessment [ 8 ] is therefore essential, enabling qualified controllers to interrogate agent behaviour in realistic conditions and against appropriate competency frameworks. HITL evaluation must faithfully represent the uncertainty inherent in ATM operations. At a system level, uncertainty arises from schedule deviations [ 51 , 56 , 52 , 60 ] , convective weather [ 50 , 38 ] , and unexpected events such as loss of communication, military activity, or hijacking [ 15 ] . At the aircraft level, uncertainty in mass [ 57 ] , operator procedures [ 25 ] , pilot intent, and local atmospheric conditions [ 28 ] contributes to substantial variability between predicted and observed trajectories [ 47 , 63 ] . Deterministic simulations therefore provide an incomplete basis for evaluating AI agent behaviour under realistic operational conditions, particularly when the agents themselves must reason under uncertainty. A simulation environment suitable for AI agent development and HITL assessment must therefore satisfy several requirements: • A high-fidelity digital representation of airspace that captures sectorisation, route structure, procedures, and other factors that form the operational context for an air traffic sample. • The capacity for probabilistic simulation of aircraft performance and traffic evolution, in order to represent both aleatoric variability and the level of epistemic uncertainty inherent in ATM operations. • An interface for AI agents to perform tactical control under uncertainty , enabling them to perceive, decide, and act within the simulated environment. • Quantitative metrics to assess AI agents, together with a Human Machine Interface (HMI) that enables competency-based assessment by human instructors. • An interface with the live or recorded real-world operational environment , allowing AI agent behaviours to be assessed on real-world traffic samples and scenarios. • The facility to run significantly faster than real-time as an enabler for AI agents that employ Machine Learning methods such as Reinforcement Learning. These techniques require many thousands of hours of simulation in order to train effective policies. Existing ATC simulators and open source systems offer valuable capabilities, but do not simultaneously provide probabilistic aircraft performance modelling, integration with operational data streams, and interfaces designed for AI agent control and HITL assessment [ 26 , 36 , 2 , 14 , 13 , 40 , 58 , 1 , 46 ] . These limitations motivate the development of a probabilistic Digital Twin. Digital Twins provide high-fidelity, data-driven virtual representations of physical systems and have seen growing adoption in safety-critical domains such as healthcare [ 30 ] , energy [ 62 ] and transportation [ 61 , 34 , 33 ] . Probabilistic Digital Twins, which explicitly model the uncertainties of the system, have been explored in aerospace for structural and component modelling including wings [ 35 ] , unmanned aerial vehicles [ 29 ] , and rotorcraft systems [ 37 ] . At the airspace level, 55 consider a Digital Twin for future airspace concepts focused on uncrewed and urban air mobility operations below 400 ft. To the authors’ knowledge, however, there is currently no probabilistic Digital Twin of an operational en route airspace designed specifically to support AI agent development, benchmarking, and HITL evaluation. This work makes the following contributions: • A probabilistic Digital Twin of operational en route airspace , integrating multiple operational data sources with a probabilistic, physics-informed aircraft performance model. • A quantitative validation of the Digital Twin , demonstrating trajectory accuracy, uncertainty fidelity, and replication of human ATCO training exercises. • A unified framework for AI agent development, training and evaluation , including automated metrics and interfaces for AI agent control of aircraft and HITL assessments. This paper introduces a probabilistic Digital Twin for the London Area Control Centre (LACC) environment. The Digital Twin integrates real-world surveillance, flight plans, sectorisation, coordination procedures, and meteorological data with a probabilistic, physics-informed machine learning (PIML) model of aircraft performance to

Read the complete text