Self-supervised Machine Learning Based Approach to Orbit Modelling Applied to Space Traffic Management

Aug 3, 2026

Upload by Hazel

This paper presents a self-supervised machine learning methodology that improves space traffic management by leveraging a pre-trained orbit model. Inspired by BERT-like language models, the work introduces ORBERT to learn meaningful representations from large volumes of unlabeled orbit data. A proof-of-concept evaluates all-vs-all conjunction screening as a time-series classification task, showing that pretraining on unlabeled ephemerides boosts performance, especially when labeled data is scarce.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Motivation for ML in Space Traffic Management

Inspiration from Self-supervised Learning in NLP

ORBERT Approach for Orbit Representation Learning

Downstream Task: Conjunction Screening

arXiv:2312.06854v1 [[physics. space-ph](physics. space-ph)] 11 Dec 2023 Self-supervised Machine Learning Based Approach to Orbit Modelling Applied to Space Traffic Management Emma Stevensona , Victor Rodriguez-Fernandeza , Hodei Urrutxuab , Vincent Morandc , David Camachoa a School of Computer Systems Engineering, Universidad Polit´ecnica de Madrid, Calle de Alan Turing, 28038 Madrid, Spain, Email: {emma.stevenson, victor.rfernandez, [david.camacho](david.camacho}@upm.es)[}](david.camacho}@upm.es)[@upm.es](david.camacho}@upm.es)[ ](david.camacho}@upm.es)b European Institute for Aviation Training and Accreditation, Universidad Rey Juan Carlos, Camino del Molino 5, 28942 Fuenlabrada, Spain, Email: [hodei.urrutxua@urjc.es](hodei.urrutxua@urjc.es) cCNES, 18 avenue Edouard Belin 31400 Toulouse France, Email: [vincent.morand@cnes.fr](vincent.morand@cnes.fr) Abstract This paper presents a novel methodology for improving the performance of machine learning based space traffic management tasks through the use of a pre-trained orbit model. Taking inspiration from BERT-like self-supervised language models in the field of natural language processing, we introduce ORBERT, and demonstrate the ability of such a model to leverage large quantities of readily available orbit data to learn meaningful representations that can be used to aid in downstream tasks. As a proof of concept of this approach we consider the task of all vs. all conjunction screening, phrased here as a machine learning time series classification task. We show that leveraging unlabelled orbit data leads to improved performance, and that the proposed approach can be particularly beneficial for tasks where the availability of labelled data is limited. keywords: self-supervised learning, transfer learning, machine learning, orbit modelling, orbit prediction, conjunction assessment 1 Introduction Ensuring the safety and sustainability of space operations in the New Space era is an ever-increasing challenge for Space Traffic Management (STM) [9] . In the face of rising space traffic, large constellations, and a growing space debris population, STM activities such as collision avoidance are critical for preserving both current day space assets, and the future usability of the space environment. To address the challenges posed by the scale and complexity of these activities, one emerging approach is the exploitation of recent advancements in the fields of machine learning (ML) [14, 13 , 12] . Advancements in this field are far reaching, with breakthroughs in a variety of different domains benefiting the space sector, from image-based computer vision to text-based Natural Language Processing (NLP) . Current applications range from vision-inspired tasks such as space object characterisation [5] and satellite pose estimation [6], to the use of NLP techniques for aiding in early space mission design [1] . However, the success of ML in many of these tasks relies on large, labelled datasets, the availability of which can be a limiting factor in their performance. Until recently, the achievements of NLP in leveraging vast quantities of unlabelled data have been largely ignored outside of the text realm. In this work, we take inspiration from these techniques, constructing an orbit model that is able to leverage large quantities of readily available orbital data, which can be built upon to perform better STM. In much the same way that different STM tasks rely on our ability to accurately model orbits, different NLP tasks, such as next word prediction or sentiment analysis, rely on an underlying common understanding of how to model language. The latest breakthroughs in this field can be attributed to the use of self-supervised learning (SSL), whereby high performance underlying language models such as Google’s BERT [2] can be pre-trained on extensive datasets by predicting masked words from text, before being fine-tuned to the objectives and data of spe

FAQs

What is ORBERT and what problem does it address?

ORBERT is a pre-trained orbit model built with a self-supervised approach to learn orbit representations. It addresses the limitation of labeled data in space traffic management tasks by using abundant unlabeled orbit data.

How does the proposed self-supervised training work for orbit ephemerides?

The model masks sections of orbit ephemeris and learns by reconstructing the missing parts. This time-series reconstruction objective enables the learning of useful orbit representations.

Which space traffic management task is used to demonstrate the approach?

The paper demonstrates the method on all-vs-all conjunction screening, formulated as a time series classification task for assessing pairwise conjunctions.

Aug 3, 2026

Upload by Hazel