7.8.4 — Military AI Systems — maturity: live

Sensor Fusion Engines

Combining real-time data streams from radar, optical, RF, SIGINT and space-based sensors into a single, continuously updated operational picture using AI-driven fusion algorithms.

Auto-drafted reference page — content reviewed for shape, individual references not yet link-checked.

A modern battlespace produces more sensor data than any human team can synthesise. Radar tracks, EO imagery, SIGINT intercepts, AIS feeds, HUMINT reports and allied liaison feeds all arrive asynchronously, in different formats and on different classification nets. Without a machine-speed fusion engine sitting at the centre of that architecture, commanders are working from a picture that is already stale — and in contested environments, staleness kills.

Satellite constellations are the primary feeder layer for national-level fusion engines. A multi-domain LEO constellation — combining synthetic aperture radar, wide-area optical, and RF geolocation payloads — delivers continuous, independent sensor streams that no single ground or airborne asset can replicate. On-board edge inference strips each pass down to object detections and track hypotheses before downlink; the ground fusion engine then correlates those hypotheses across all sensors, resolves ambiguities using probabilistic data association, and emits a single fused track file within minutes of collection. The result is an order-of-magnitude improvement in latency compared with a human-mediated multi-INT workflow.

The operational outcome is a living common operating picture that updates automatically as new passes complete, flags track breaks and anomalies, and feeds directly into command-and-control systems without a data-entry step. Nations that operate this stack autonomously can fuse classified national intelligence with allied contributions on their own terms — sharing selectively, withholding what treaty or operational security demands, and never routing sensitive detections through a foreign commercial cloud.

What matters

Track correlation latency below five minutes from satellite overpass to fused COP entry is operationally achievable with on-board edge inference and a sovereign ground cluster.
Probabilistic data association algorithms (e.g. JPDA, MHT) require access to the raw sensor hypotheses — not the sanitised products a commercial vendor will release.
Export-control regimes (US ITAR, EU Dual-Use Regulation) routinely block the transfer of fusion software trained on classified sensor geometries to foreign militaries, including close allies.
A hostile actor who can degrade or spoof the fusion feed without the defender knowing has effectively blinded the entire command chain, making supply-chain integrity a first-order security requirement.

Sovereignty score: 9/10 — A sensor fusion engine that routes national intelligence through foreign algorithms or infrastructure is not a sovereign capability — it is a delegated one, revocable at the vendor's or adversary's discretion.

US ITAR and the EU Dual-Use Regulation prohibit export of fusion software trained on classified sensor parameters, meaning allies cannot legally receive — and adversaries will actively attempt to acquire — the core algorithmic stack.
Any commercial cloud or SaaS fusion vendor retains the contractual right to suspend service, impose data-retention obligations, or comply with foreign court orders — all of which are incompatible with wartime operational security.
A nation that does not control the fusion model weights cannot audit for adversarial poisoning, backdoors or deliberate track-suppression inserted during a supply-chain compromise.
Escalation control requires the ability to selectively share or withhold fused intelligence from allies and coalition partners; this is only possible when the fusion engine and its outputs reside on nationally governed infrastructure.

Reference architecture

Payload: Three complementary payload classes per orbital shell: (1) X-band SAR, 1m spotlight / 20km swath, on-board FPGA running CFAR and track-hypothesis generation; (2) RGB + SWIR EO imager, 0.5m GSD, 12km swath, on-board CNN for object detection at L0→L1; (3) RF survey payload, 100 MHz–18 GHz, 500m geolocation accuracy via TDOA across cross-linked satellites.
Bus class: ESPA-class microsat, 150–200 kg wet mass, 600–900 W end-of-life power, cold-gas propulsion for constellation maintenance; cubesat (16U) variant acceptable for the RF survey layer only.
Orbit: Sun-synchronous LEO at 480–550 km for SAR and EO shells (36-satellite walker, 6 planes × 6 sats, sub-60-minute global revisit); near-equatorial LEO at 550 km for RF geolocation shell (24-satellite walker optimised for mid-latitude theatre coverage).
Ground segment: 4-station national ground network (X-band for SAR/EO downlink, S-band TT&C); minimum two stations geographically separated for resilience; classified MPLS network connecting ground stations to the national fusion centre; SatNOGS on UHF/2.4 GHz for housekeeping backup only.
Software & data
Latency
Cost band
Lead time

References

NATO STO: Multi-Source Data Fusion for Military Applications (STO-TR-SET-250) — NATO Science and Technology Organisation report examining multi-sensor fusion architectures for military situational awareness, covering probabilistic association methods and latency benchmarks across radar, EO and SIGINT feeds.
ESA — Earth Observation for Security and Defence — ESA's security and defence programme overview describes how multi-payload satellite constellations feed national fusion centres and the importance of sovereign data access rights for member states.
DARPA — Mosaic Warfare and Multi-Domain Command and Control — DARPA's Mosaic Warfare programme explicitly frames sensor fusion engines as the enabling technology for disaggregated, resilient force employment, requiring machine-speed correlation across heterogeneous sensor nodes.
UNIDIR — Cyber and Space Security: Implications for Military AI — UNIDIR analysis highlights that AI-driven fusion engines processing classified multi-INT streams represent a critical national security asset where third-party cloud dependency creates unacceptable intelligence exposure.
IEEE Aerospace Conference — On-Board AI Inference for Satellite Multi-Sensor Fusion — Peer-reviewed proceedings document demonstrated on-board edge inference pipelines that reduce ground-to-fused-track latency from 45 minutes to under four minutes on LEO microsatellite platforms carrying combined SAR and optical payloads.