A Neural ODE Approach to Aircraft Flight Dynamics Modelling

Gabriel Jarry 1 , Ramon Dalmau 1 , Xavier Olive 2 , Philippe Very 1 1 EUROCONTROL Aviation Sustainability Unit (ASU) Brétigny-sur-Orge, France 2 ONERA – DTIS Université de Toulouse Toulouse, France · 2025

Read A Neural ODE Approach to Aircraft Flight Dynamics Modelling by Gabriel Jarry 1 , Ramon Dalmau 1 , Xavier Olive 2 , Philippe Very 1 1 EUROCONTROL Aviation Sustainability Unit (ASU) Brétigny-sur-Orge, France 2 ONERA – DTIS Université de Toulouse Toulouse, France (2025) 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-2509-23307.txt.

Introduction

Flight trajectory forecasts feed air traffic control, airline planning, and emissions checks. Gabriel Jarry and 3 co-authors model flight dynamics with neural ordinary differential equations. Their NODE-FDM trains on quick access recorder data from real flights. Analytic kinematic relations mix with learned components. The hybrid aims for high fidelity without a full physics build.

Against a BADA-based state-of-the-art method, recorded paths were matched more closely. The descent phase showed the clearest gains. Altitude, speed, and mass dynamics all improved. Limits remain: physical constraints are light and recorder data are scarce. Lateral dynamics are slated for future work.

Catalog record

CREATORGabriel Jarry 1 , Ramon Dalmau 1 , Xavier Olive 2 , Philippe Very 1 1 EUROCONTROL Aviation Sustainability Unit (ASU) Brétigny-sur-Orge, France 2 ONERA – DTIS Université de Toulouse Toulouse, France
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-23307

Browse the full library · About the Tower of Babel

Full text

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

A Neural ODE Approach to Aircraft Flight Dynamics Modelling Gabriel Jarry 1 , Ramon Dalmau 1 , Xavier Olive 2 , Philippe Very 1 1 EUROCONTROL Aviation Sustainability Unit (ASU) Brétigny-sur-Orge, France 2 ONERA – DTIS Université de Toulouse Toulouse, France Abstract Accurate aircraft trajectory prediction is critical for air traffic management, airline operations, and environmental assessment. This paper introduces NODE-FDM, a Neural Ordinary Differential Equations-based Flight Dynamics Model trained on Quick Access Recorder (QAR) data. By combining analytical kinematic relations with data-driven components, NODE-FDM achieves a more accurate reproduction of recorded trajectories than state-of-the-art models such as a BADA-based trajectory generation methodology (BADA4 performance model combined with trajectory control routines), particularly in the descent phase of the flight. The analysis demonstrates marked improvements across altitude, speed, and mass dynamics. Despite current limitations, including limited physical constraints and the limited availability of QAR data, the results demonstrate the potential of physics-informed neural ordinary differential equations as a high-fidelity, data-driven approach to aircraft performance modelling. Future work will extend the framework to incorporate a full modelling of the lateral dynamics of the aircraft. † † aftertitle: I Introduction Accurate aircraft trajectory prediction is essential to air traffic management (ATM), airline operations, and aviation environmental assessment. Precise trajectory models are indispensable for conflict detection and resolution, sector capacity planning, and trajectory-based operations (TBO). Beyond operational safety and efficiency, high-fidelity predictions enable robust estimates of fuel consumption, emissions, and noise, thereby supporting the development of sustainable aviation strategies. The most widely adopted reference for aircraft performance modelling is the Base of Aircraft Data (BADA) [ 1 , 2 ] , developed by EUROCONTROL. BADA provides a standardized set of equations and performance parameters to represent aircraft dynamics under nominal conditions. Its analytical formulation offers transparency and robustness, which explains its longstanding use in both operational and research contexts. Nevertheless, such models inevitably rely on simplifying assumptions and manufacturer-provided data, which can lead to divergences between simulated trajectories and those actually flown, particularly in speed profiles and mass dynamics as captured by Automatic Dependent Surveillance-Broadcast (ADS-B) measurements. Alternative open-source frameworks, such as OpenAP [ 3 ] , have emerged to complement BADA by providing flexible, data-driven tools for trajectory modelling and validation. This study introduces NODE-FDM, a novel data-driven Flight Dynamics Model that integrates large-scale Quick Access Recorder (QAR) data with physics-informed neural architectures. We focus first on the vertical and speed dimensions of the trajectory, including altitude profiles, flight path angle, and true airspeed. This paper does not address lateral trajectory management (e.g., route choice and turn dynamics), which we leave for future investigations. We apply the model to the Airbus A320-214 aircraft. The proposed model combines analytical kinematic relations, regression-based propulsion and aircraft angle modelling, and a Neural Ordinary Differential Equations (ODE) formulation to predict the temporal evolution of altitude and speed with high fidelity. We evaluate its predictive performance against a BADA-based trajectory generation methodology (BADA4 performance model combined with trajectory control routines, hereafter referred to as BADA), focusing on trajectory reconstruction accuracy and suitability for environmental impact assessments. These assessments include developing CO 2 and non-CO 2 emissions inventories, which often require robust trajectory predictions even when only partial datasets or pre-flight planning information are available. The remainder of this paper is organized as follows. Section II reviews the literature on aircraft trajectory prediction, covering model-based, data-driven, and hybrid physics-informed approaches, and discusses operational use-cases and open challenges. Section III describes the QAR datasets and preprocessing steps used to train the model. Section IV details the NODE-FDM methodology, including its modular architecture, state representation, neural and analytical layers, and training objectives. Section V presents quantitative results, benchmark comparisons with BADA, and analysis of trajectory reconstruction and fuel consumption estimation. Section VI discusses the limitations of the proposed approach, and Section VII concludes with a summary of findings and directions for future work. II Literature review Trajectory prediction and generation underpin a wide range of ATM and airline operations, from conflict detection and sequencing to environmental assessment and strategic planning. Two complementary paradigms dominate: model-based approaches grounded in aircraft performance and point-mass dynamics, and data-driven methods that learn spatio-temporal regularities from large-scale surveillance and weather datasets. Physics-based point-mass formulations and performance databases remain the historical backbone of operational trajectory prediction. The EUROCONTROL BADA family provides aircraft-type parameters and procedures for 4D trajectory computation under forecast conditions [ 1 , 2 ] , while OpenAP extends these capabilities to the research community with an open-source framework [ 3 , 4 ] . Beyond canonical models, the literature formalizes trajectory representation, compression, and optimization using splines, principal component analysis, and connections to optimal control and wavefront propagation under constraints [ 5 ] . Recent efforts by Poll and Schumann refine performance characteristics and cruise fuel modelling from aerodynamic theory and empirical data [ 6 , 7 ] . These approaches provide interpretability and controllability, but rely on accurate state and intent estimates and may under-represent operational variability. Data-driven methods have achieved notable gains in 4D trajectory prediction using supervised learning with surveillance and meteorological features. Early work by Ayhan and Samet demonstrated machine-learning-based prediction of arrival times and trajectories using historical and weather data [ 8 ] . Sequence models then became prevalent: Liu and Hansen proposed a seq2seq long short-term memory (LSTM) [ 9 ] , Pang et al. combined convolutional encoders of convective weather “cubes” with recurrent decoders [ 10 ] , and Ma and Tian introduced a convolutional neural networks (CNN)–LSTM on ADS-B data [ 11 ] . Online frameworks fuse point-mass physics, BADA parameters, and ADS-B conformance to update intent and improve estimated time of arrival precision [ 12 ] . Generative models address multi-modality and error accumulation: generative adversarial network (GAN)-based forecasters produce full-sequence 4D trajectories [ 13 ] , variational autoencoders (VAE)/GAN generators improve realism in sparse regimes [ 14 ] , privacy-preserving LSTM–GANs synthesize flows while mitigating re-identification risk [ 15 ] , and imitation-learning methods such as TrajGAIL learn realistic route choices and flow-level statistics [ 16 ] . Data-driven models excel at capturing variability and airline-specific practices, though generalization off-distribution and physical guarantees remain challenging [ 17 ] . Hybrid strategies combine the adaptability of data-driven approaches with the consistency of physics-based models, either by calibrating physics-based predictors using data or by embedding known dynamics into neural architectures. Calibration work has improved climb predictions by inferring aircraft-specific parameters from radar data [ 18 ] . Physics-informed learning enforces equations of motion and aerodynamic limits via loss functions or architectural constraints. Neural ODEs formalize these approaches as continuous-time dynamical systems, trained efficiently with adjoint methods and evaluated using ODE solvers [ 19 , 20 ] . Aerospace applications include learning residual dynamics under icing [ 21 ] , hypersonic glide vehicle prediction [ 22 ] , and differentiable predictive control for unmanned aerial vehicles [ 23 ] . Practical challenges remain regarding stiffness, training cost, stability, and hybrid dynamics with discrete modes [ 24 ] . Beyond prediction, ATM applications also require the generation of realistic trajectories and flows with quantified realism and flyability. Statistical density models, such as Gaussian mixture models and vine copulas, combined with dimensionality reduction, reproduce flow statistics and go-around patterns

Read the complete text