Ingo Waldmann, Spaceflux; Marco Rocchetto, Spaceflux; Kerri Mertz, Spaceflux; Joshua Williamson, Spaceflux; Fiorenzo Stoppa, Spaceflux
https://doi.org/10.64861/NMDT5653
Keywords: Pattern of Life, Space Domain Awareness, Space Situational Awareness, Anomaly Detection, Physics-Informed Machine Learning, Orbital Behaviour Analysis, Satellite Characterisation, Proactive Threat Detection, Autonomous Sensor Tasking
Abstract:
Space Domain Awareness is undergoing a fundamental shift. Legacy systems detect and track objects. The next generation must anticipate intent. This paper presents Spaceflux’s Pattern of Life (PoL) analytical framework, Sentinel, which uses physics-informed machine learning to establish behavioural baselines for resident space objects and detect anomalies that precede hostile or anomalous manoeuvres across Low Earth Orbit, Medium Earth Orbit, and Geostationary Orbit.
Current space surveillance systems excel at collision avoidance but struggle to identify the subtle behavioural precursors that distinguish a routine station-keeping burn from the opening move of a coordinated proximity operation. These precursors, including minor photometric changes, missed burns, and gradual positional drift, are individually ambiguous but collectively diagnostic. Detecting them requires both the right data and the right analytical framework.
Spaceflux operates a global optical sensor network across five continents, providing continuous multi-regime observations that feed our Sentinel PoL. Sentinel is build from the ground up to fuse multiple phenomenologies and sensor networks By embedding physical constraints derived from orbital mechanics directly into the learning process, our models learn to distinguish genuinely anomalous behaviour from the expected variability of each orbital regime. This physics-informed approach significantly reduces false positive rates while preserving sensitivity to real threats. Critically, the same framework generalises across LEO, MEO, and GEO without regime-specific retraining, enabling unified multi-orbit intelligence from a single operational system.
Speed is a defining performance characteristic. The system processes observations to actionable alerts in under 15 minutes. In contested orbital environments where manoeuvre windows can open and close within a single pass, this latency represents the difference between strategic warning and a post-event forensic analysis. Critically, once an event is detected, Sentinel automatically tasks our or an associated sensor network to obtain follow-on observations. This enables real-time analysis as the event unfolds.
We present results demonstrating greater than 90% anomaly-detection accuracy across our operational catalogue, together with validated cross-regime performance for objects spanning all three orbital domains. We discuss how the physics-informed architecture achieves this generalisation without sacrificing speed, and we examine the implications for sovereign space surveillance capability as orbital congestion and adversarial activity continue to grow.
Date of Conference: September 15-18, 2026
Track: AI for SDA Applications