Skip to content
Capability transparency

What is demonstrated, what is only simulated, and what is vision.

It is one of our three principles, so it gets its own page. Here is the detail behind every claim on the home page: the full ladder, what exactly the published figure measures, and why the bench is not the product.

Capability ladder

L1 today. L2 and L3 ahead.

We are explicit about what is operational, what is validated only in simulation and what is vision.

  1. L1
    Demonstrated onboard

    Edge ISR perception

    Onboard detection and geolocation of people, with compact output tolerant to a degraded or denied link.

  2. L2
    Validated in SITL simulation (Jul 2026) · blocked on a real platform by agreement with the manufacturer

    Perception-driven autonomy

    Both halves travel together: it is validated in simulation and it is blocked on a real platform, and the block is commercial, not technical. What we ship is passive by decision: it reads telemetry and emits events, and does not act on the flight until an agreement with the platform manufacturer exists.

  3. L3
    Vision · 18-24 months

    GPS-denied navigation

    Robust navigation in GPS-denied environments, a necessary condition for contested missions.

The caveats, in full

What sits behind each sentence.

"The module's own output" is not decoration

The quality figures are measured on the boxes the module emits over the CBOR contract, which is exactly what an integrator receives over the link. The figure measured on the detector's floating-point boxes, before quantisation, is different and better. We publish the worse of the two on purpose, because it is the one that describes the product.

Two populations, and they are not averaged

There is no single quality figure: there are two, each with its own validation set and object count. Averaging them would produce a number that corresponds to no real measurement. That is why both lines always travel together, each carrying its scope inside.

The bench is not the product

The NVIDIA Jetson Orin Nano Super is our development bench and the surface we demonstrate on. It is not the compute module of the flyable product: that one has not been chosen yet, and when it is, it will be measured on it and published with its surface declared, just like this one.

L2: both halves travel together

Perception-driven autonomy is validated in SITL simulation, and it is blocked on a real platform by agreement with the manufacturer. The block is commercial, not technical. Quoting only the first half inflates what we have; quoting only the second wastes real evidence. What we ship is passive by decision: it reads telemetry and emits events, and does not act on the flight.

Declared maturity: TRL 4

We have bench and simulation evidence. The integrated system does not fly yet, and we do not imply otherwise: no image and no claim on this site should be read as flight proof.

Sovereignty and export control

IP, models, data and mission layer developed and controlled in the EU, on COTS components from allied countries with no licence restriction (EAR NLR), with no excluded suppliers and no dependency on foreign cloud. No ITAR components in the stack. What is European is the software and the control of the IP, not the provenance of every chip.

The training and validation datasets are published under the CC BY 4.0 licence; their complete citations, with the mandatory statement of modifications, are in the legal notice.

Next step

Do you have a case where the computing has to happen on the device?

Let's talk. Institutional programmes, industrial integrators and allied partners.