A Self-Supervised Framework for Space Object Behaviour Characterisation

doi:10.1111/exsy.70382

A Self-Supervised Framework for Space Object Behaviour Characterisation

2026-09-30

0

OA

AI

Diego Ramírez Rodríguez

DOI：10.1111/exsy.70382

Original

Save

View PDF

Abstract

En 中文

Foundation Models, which leverage large neural networks pre-trained on unlabelled data before fine-tuning for specific tasks, are increasingly being applied to specialised domains. Recent examples include ClimaX for climate and Clay for satellite Earth observation, but a Foundation Model for Space Object Behavioural Analysis (SOBA) has not yet been developed. As orbital populations grow, automated methods for characterising space object behaviour are crucial for space safety. Here, we present a self-supervised framework for SOBA, representing a first step towards a Foundation Model for Space Object Behavioural Analysis. The backbone of this framework is a Perceiver-Variational Autoencoder (VAE) architecture, pre-trained with self-supervised reconstruction and masked reconstruction on ~227,000 light curves from the MMT-9 observatory. The VAE enables anomaly detection, space object motion prediction and generation of synthetic light curves. We fine-tuned the model for anomaly detection and motion prediction using two independent light curve simulators (CASSANDRA and GRIAL respectively), using CAD models of boxwing, Sentinel-3, SMOS and Starlink platforms. Our pre-trained model achieved a reconstruction mean squared error of 0.0012, identifying potentially anomalous light curves through reconstruction difficulty. After fine-tuning, the model scored 85% accuracy (0.92 ROC AUC) on anomaly detection and 82% accuracy (0.95 ROC AUC) on motion mode prediction (e.g., sun-pointing, spin and tumbling). Analysis of high-confidence anomaly predictions on real data revealed distinct patterns including characteristic object profiles and satellite glinting. The motion prediction model successfully differentiated between various movement behaviours such as sun-pointing, spin and tumbling. Our work demonstrates how self-supervised learning can simultaneously enable anomaly detection, motion prediction and synthetic data generation from rich representations learned in pre-training. More broadly, our work supports space safety and sustainability through automated monitoring and simulation capabilities.

Keywords:

attitude prediction

generative AI

light curve anomaly detection

self-supervised learning

space object Behavioural analysis

space situational awareness (SSA)

A Self-Supervised Framework for Space Object Behaviour Characterisation

Abstract

AI Summary

Journal

Organization

Cited Papers