Picture the worst moment of a wildfire patrol. A drone is working a high-risk ridge at first light. Below it, a heat source is just starting to throw a column of smoke. That column rises straight into the drone's line of sight to the ground. The valley behind it has no signal. The radio band is thick with the chatter of an incident already running. This is the exact instant the aircraft has to make its hardest decisions — is that a fire, confirm it, call it in — and it is the exact instant the link to the ground is least likely to be there.
That is the problem Maestro is built around. The moment the work matters most is the moment the connection matters least. So the part of the system that thinks has to live on the aircraft, not on a laptop at the truck.
The Ground-Station Assumption Breaks
Most drone software treats the aircraft as a remote-controlled limb of a ground station. The brain sits on a laptop; the drone streams video and telemetry up and waits for commands down. That model is fine in a clear sky on a calm day with a strong radio link. It is exactly the model that fails in the conditions wildfire detection actually runs in.
Smoke is the obvious culprit, and it attacks the link in more than one way. A plume scatters and blocks the signal it rises through. It is also the very thing the drone is there to find — so the aircraft is, by design, flying toward the conditions that degrade its own connection. Add the rest of the terrain: a ridge between the drone and the operator, a cellular dead zone over open country, a band already crowded by the incident's own radios. The link does not fail at a convenient time. It fails at the worst one.
When the link degrades, a ground-brained drone has three poor options: loiter and wait, keep flying a stale plan it can no longer update, or turn around and come home. None of those finds the fire. They are the absence of intelligence dressed up as a fallback. The most demanding seconds of the flight arrive precisely when a ground-brained system has the least to offer.
Maestro inverts that. The autonomy runs on the aircraft. The drone keeps perceiving, deciding, and acting when the ground can no longer reach it, because the part of the system that does those things is already onboard.
Detect, Confirm, Alert — Onboard
The hardest calls in a detection mission are the ones that cannot wait for a round trip to a laptop. Maestro keeps them on the drone.
- Detect. Perception runs on the aircraft's own compute. What the camera sees becomes a candidate detection in milliseconds — not a video frame queued for a link that may not be there. The drone notices the early heat signature while it is still small.
- Confirm. A single frame is a guess; a confirmed detection is a decision. The aircraft reasons over what it is seeing onboard, and the fleet can bring a second drone to corroborate the call from another angle — turning one drone's hunch into a verified sighting without anyone on the ground having to stitch it together.
- Alert. The instant a detection is confirmed, the alert is raised — pushed the moment the link allows, and held onboard until it does. The operator gets a confirmed location, not a backlog of raw footage to review after the fact.
This is the difference between a drone that is remote-controlled and a drone that is autonomous. A remote-controlled drone is only as smart as its connection. An autonomous one carries its judgement with it — straight into the smoke.
What Each Drone Decides for Itself
Maestro splits the work in two. The ground side owns the mission: the high-risk zones you want watched, how the fleet divides them up, and the limits you set under your own authorizations. The aircraft owns the next few seconds: where it is, what it sees, what to do right now, and how to keep flying the plan when conditions change faster than a radio link can keep up with.
Concretely, the onboard autonomy owns:
- Local decision-making. Given the current mission intent and its live state, the aircraft decides the next action onboard — hold the search line, re-route around terrain, move in to confirm a heat source, or escalate — without waiting on the ground.
- Onboard perception. Detection and scene understanding run on the aircraft, so a faint plume on the horizon becomes a decision in the moment rather than a clip reviewed later.
- Degraded-comms behaviour. When the link to the ground drops, the drone does not freeze. It keeps executing the agreed mission intent, applies its onboard safety rules, and reconciles with the rest of the fleet the instant the link returns.
- The safety floor. Geofence, battery, and the autopilot's own failsafes remain the inviolable bottom layer. The onboard autonomy reasons above that floor; it never reasons around it.
Onboard Autonomy and Shared State, Together
Onboard autonomy on its own would just be a fleet of clever loners. What makes it more than that is that every aircraft works with the rest of the fleet. Each drone makes fast decisions on its own, then shares what it learns — where it is, what it found, what it is doing, how much battery it has left — so every drone and the operator work from one live picture of the search.
So the loop is simple: the ground side sets the mission, each aircraft executes it onboard at the speed the fire moves, and everyone keeps working from the same live picture. When the link to one drone drops, the rest of the fleet and the operator still hold the last state it shared; when it returns, the drone reconciles. A confirmed detection one aircraft makes behind the ridge is not lost — it surfaces the moment that drone is back in contact. The system degrades gracefully instead of going dark.
Runs on the Drones You Already Fly
Maestro is built to run on the drones you already fly. It deploys onto open MAVLink, PX4, and ArduPilot airframes with a companion computer you control — that companion is where the onboard autonomy lives. We are a system vendor for autonomous aerial operations, not a drone manufacturer. The job is to bring autonomy to your fleet, whatever mix of drones it is, and to do it without forcing you to buy new aircraft.
The focus is early detection. Maestro patrols the high-risk zones you define, under your own authorizations, and flags a heat source while it is still small enough to act on — the window where minutes matter most. It is built to spot ignition early, not to fly into an active fire.
The Bet Maestro Makes
Autonomy belongs onboard. Each drone keeps flying the mission even if the link drops, decides the next few seconds for itself, and turns what it sees into a confirmed alert without waiting on a radio. The link to the ground makes everything better — it is not what the drone depends on. That is the principle Maestro is built on, and it is the one that survives the place where wildfire detection actually happens: the ridge with no line of sight, the band crowded with traffic, and the smoke column that blinds the very link a ground-brained drone would be begging for.
Want to see how it fits together? Read about Maestro Fire and the platform underneath it, or launch the live demo to watch a fleet run a patrol.