Technology
SDR-017 ·

One stack, from photon to decision.

Four capability cores, CUSTOS, PRAETOR, IGNIS and ASTRUM, built as one system: sensing feeds estimation, estimation feeds autonomy, autonomy answers to one operator.

THE FOUR CORES

Control loopState estimate · EKF · 400 Hz

CUSTOS

IN DEVELOPMENT

Autonomy core for uncrewed platforms. Edge perception, on-board reasoning, and resilient flight across degraded and denied links. The guardian process every airframe runs.

CLASS
Autonomy / flight
DOMAIN
Air · Coastal
LINK
Degraded-tolerant
Ground stationOne operator · many aircraft

PRAETOR

IN DEVELOPMENT

Ground control layer. Fleet tasking, telemetry aggregation, and human-on-the-loop authority from a single ground station: one operator, many aircraft.

CLASS
Ground control
SPAN
Multi-vehicle
AUTHORITY
Human-on-the-loop
Concept schematicHybrid-electric · zero-emission drive

IGNIS

CONCEPT

Clean-propulsion architecture for aerospace and maritime. Hybrid-electric and zero-emission drive engineered for endurance over noise.

CLASS
Propulsion
CYCLE
Hybrid · ZE
TARGET
Endurance
Pushbroom swathSwath · low-light · on-edge

ASTRUM

RESEARCH

Sensing layer. Multi-spectral Earth observation built around SPAD arrays with in-house calibration pipelines, tuned for low-light edge inference: photon-starved scenes read on board, at dusk, in smoke.

CLASS
Sensing / EO-IR
ARRAY
SPAD · multi-spectral
INFERENCE
Low-light · on-edge
Fig. 05b · GNSS deniedSimulatedThe CUSTOS estimate in LOG-014, June 2026: coasting a 20-second GNSS blackout.
Simulation, LOG-014, June 2026. Levels are the logged peaks, not a trace. Innovation gating drops one-off outliers; it does not protect against spoofing.

THE AUTONOMY LOOP

PHOTON IN
  1. 01SENSEASTRUM

    ASTRUM multi-spectral sensing: SPAD arrays with in-house calibration, tuned for photon-starved scenes: dusk, smoke, moonless sectors. Inference is designed to run on the aircraft, not in the cloud.

  2. 02ESTIMATECUSTOS

    A navigation filter that earns its trust: optical flow, barometer and magnetometer carry the estimate through GNSS blackouts, and innovation gating drops one-off outliers. It does not protect against spoofing.

  3. 03DECIDECUSTOS

    CUSTOS mission autonomy: patrol, detect, divert, orbit, resume. Contacts are confirmed and geolocated without a human in the loop; authority stays with the operator.

  4. 04ACTCUSTOSIGNIS

    A cascade flight controller that flies on the estimate, not on hope. The same control law runs in the engineering simulator and is written to run unchanged on the flight computer, and IGNIS is what turns its mixer output into thrust.

DECISION OUT

VALIDATED IN SIMULATION, WIRED INTO CI

Every performance figure on this site traces to the engineering simulation suite in the programme repository, a 6-DOF physics model of the NIGHTJAR flying the same autonomy stack. The suite is wired into CI to run on every change to the drone code, across seeded randomised conditions, and to fail the build if the aircraft stops meeting its numbers. CI is paused at present, so the figures here come from running the suite directly.

IGNITION → ALERTTarget under 90 s on patrol. Simulator so far: 0.4–17 s to alert, 16–87 s to a confirmed fix over the six-sortie fire season
GNSS BLACKOUT1.20 m peak error over 20 s coasting on the listed sensors (2.59 m worst of six runs; 1.01 m with an optical-flow sensor not yet fitted)
GEOLOCATION< 0.31 m CEP from a simulated autonomous orbit
GATEWired into CI for drone-code changes; CI paused at present
CAPABILITY CORES4
MISSION AIRFRAMES7
ESTABLISHED2026
NIGHTJAR ALL-UP MASS1.33 kgTarget
· · ·