Autonomy

DroneCommand™ makes the limit of autonomy explicit.

Autonomy in this platform assists perception, classification and planning. It does not decide, and it does not fly the aircraft. The boundary is enforced in the command layer, not in a policy document.

Sensor fusion visualisation around a fixed-wing unmanned aircraft, with an edge compute node connected by secure datalinks.

Autonomy assists perception, classification and planning. Flight control remains independent.

Authorisation model

Three tiers, configurable per operation.

Each tier defines what an agent may do without a human in the loop. Tiers are set per mission, per agent class and per operating area, and every transition is logged.

  1. TIER 1 · AUTONOMOUSLEVEL 1 / 3

    Routine actions

    Navigation, station-keeping, sensor sweeps, data relay and return-to-base run without operator input.

  2. TIER 2 · SUPERVISEDLEVEL 2 / 3

    Confirmed actions

    Task reassignment, area changes and classification acceptance are proposed by the system and executed only on operator confirmation.

  3. TIER 3 · HUMAN AUTHORITYLEVEL 3 / 3

    Irreversible actions

    Any action that cannot be undone requires explicit, logged human authorisation. No exceptions and no automatic escalation.

Exclusions

What we deliberately do not build.

No weapons integration.

The platform has no interface to an effector and none is planned.

No autonomous engagement.

Classification of an object as hostile is a human decision, always.

No electronic warfare.

No jamming, spoofing or emission control functionality is provided.

Procurement consequence

A system without engagement functions carries a narrower risk classification under the EU AI Act and a shorter path through internal review. Human oversight, logging and traceability are built into the command layer rather than added as compensating controls.

Auditability

If it happened, it is in the log.

Oversight is only credible if it is reconstructable after the fact.

Audit record coverage
Every authorisationWho authorised what, at which tier, at what time, against which agent.
Every task transitionAccepted, started, degraded, completed, aborted, with the triggering condition.
Every overrideManual intervention is recorded with the state that preceded and followed it.
ExportStructured records can be exported for internal review, oversight bodies or incident analysis.
Briefings under NDARequest a briefing