The ladder, from orbit to action
Physical AI at OrbiVigil is one continuous chain, and every rung is measurable:
- Orbit. Satellites measure the territory continuously and cheaply, but at the resolution and cadence that orbital mechanics allow.
- Model. Those measurements maintain the territory model — geology, water, ground motion, fire, weather, official risk — each layer with its own age and confidence.
- Task. Where the model is uncertain and the stakes are high, that is a place worth sending something. The output is a task with a stated reason: not "inspect the network", but "look here first, because this is what changed".
- Asset. A drone or a ground robot, operated by whoever holds the authorisation for that airspace or site, carries the task out.
- Return. What it observes goes back into the model as a new measurement with its own provenance — so the next task is better aimed than the last.
The consequence is the point: a robotic asset is both a consumer and a source. It consumes the model to know where to go, and it feeds the model what it saw. A fleet that only consumes is a camera on a stick; a fleet that also feeds back is an instrument that improves the thing that aims it.
Why tasking is the valuable half
Flying is a solved, competitive and regulated business with excellent operators in every country we work in. Knowing where to fly is not solved. A province with roughly 1 600 km of provincial road and a shrinking maintenance budget cannot inspect the network; anything that says look at these four kilometres first is worth money, because the alternative is driving all of it. Satellite measurement narrows a network to a shortlist; the asset confirms the shortlist. Neither half is useful alone.
Graduated autonomy, and who holds the authorisation
Autonomy here is a ladder, not a switch, and each rung is a deliberate step: a human decides and a human flies; the system proposes the target and a human approves it; the system proposes and a supervised mission executes within a pre-approved envelope. Each step up is taken only where the evidence trail from the step below supports it.
That ladder is bounded by law, not by ambition. In the European Union, unmanned aircraft operations fall under Regulation (EU) 2019/947 and its open, specific and certified categories, with the operator holding the registration and the operational authorisation; automated traffic services in designated airspace fall under the U-space framework of Regulation (EU) 2021/664 and its companions. OrbiVigil sits on the intelligence side of that line: we supply the reason a flight is worth making and the record of what it found. The operator remains the operator, and the authority remains the authority.
What runs today, and what the architecture is built for
Today. The measurement half runs in production: millimetre-per-year ground motion under every provincial road across a 91-municipality institutional pilot, per-point five-year displacement histories, post-intervention verification of completed works, and mapped burnt-area perimeters and burn severity after fire. Those are exactly the outputs a targeted inspection is aimed by, and they are served through the same machine-readable contract an AI agent reads.
Built for, not yet serving. The asset adapter — the interface through which a drone or ground robot receives a task and returns its observation — is on the roadmap. It is not live, it is said to be planned everywhere it appears on this site, and no capability on this page depends on it.
What it is worth to a buyer
For a road or utility owner, targeted inspection converts a fixed budget into coverage of the segments that changed. For an insurer, the same chain produces underwriting-grade evidence: exposure measured before an event and verified after it, with provenance attached at every step. For a fire or civil-protection service, it is a reason to commit an aircraft, written down before the aircraft is committed.
Related: the territory model that aims the task; agentic AI — the contract the asset would act on.