Wildlife monitoring and anti-poaching patrols.

The mission

A reserve of several thousand square kilometres, a handful of ranger teams, and two kinds of eyes: patrol drones flying dawn and dusk sweeps, and fixed camera traps at the waterholes and crossing points. The operations room needs one thing above all else — to know about an incursion while there is still time to act on it. A vehicle can only be vectored to a crossing point if the sighting arrives in minutes; the ecological record — the imagery, the counts, the collar data — matters too, but it matters tomorrow. The reserve’s network was built for neither: a cellular patch near the gate, a repeater on one hill, and silence over most of the map.

What breaks

The physics is sparse coverage over enormous area. The drone that spots a vehicle where no vehicle should be is, almost by definition, deep in the dead ground — incursions avoid the covered places for the same reason the rangers struggle to reach them. VHF voice crosses some of the distance but carries no images and no positions a system can act on; cellular exists only in patches; and a satellite terminal on every vehicle is a budget the administration does not have.

The classical failure mode is that everything waits for the drone to come home. The detection, the imagery and the telemetry land together when the aircraft lands — hours after the moment they could have changed. Treating all of it as one stream means the urgent kilobyte is held hostage by the routine gigabyte.

The architecture on this mission

Every vehicle, drone and camera-trap gateway runs the same node, forming one encrypted mesh across the reserve — no controller, no cloud, nothing that stops working when the one repeater does. A vehicle on a ridgeline with one bar of cellular becomes, for that hour, the reserve’s uplink: the drone in the valley reaches HQ through it, two short hops replacing an impossible long one. Whatever coverage exists anywhere on the patrol is bonded and shared; where none exists, the mesh still moves traffic between the platforms that are in reach of each other.

The class system is what turns geography into a triage rather than a blockade. A detection — position, class, a thumbnail — is a few kilobytes of high-class traffic, and a few kilobytes get through on almost anything: a thin relay chain, a single bar, a brief line-of-sight contact. Full imagery is bulk: it queues on the aircraft, store-carry-forward, and drains on the return leg as the drone re-enters coverage, or at the charging point, in priority order. The interception is never waiting behind the archive.

The fixed sensors work the same way in slow motion. Camera-trap clusters queue their take for days and drain opportunistically to whatever passes — the dawn drone sweep, the vehicle on the weekly loop — so the sites need no uplink of their own and no visit to collect cards. And because every link is measured continuously, a coverage map of the reserve accumulates as a by-product of patrolling it, showing exactly where a mast or a relay would pay for itself.

How the patrol day unfolds

1. Dawn. Two drones launch on the waterhole sweep; three vehicles depart on their loops. Nodes joined the mesh at power-up — the operations room sees every platform and every link’s measured state on one screen.
2. The sweep. The eastern drone passes two camera-trap clusters and drains their queues in the seconds of overflight — four days of images, riding home in the aircraft. No visit, no cards.
3. The detection. Deep in the dead ground, the drone flags a vehicle track where none belongs. The detection — position, heading, thumbnail — leaves immediately as high-class traffic, hops through the vehicle on the ridge, and is on the operations screen in seconds.
4. The interception. The nearest team is vectored to the crossing point. Their vehicle’s voice and data ride the same mesh; as they drive out of the cellular patch, the route shifts to relay hops without anyone touching anything.
5. The return legs. Drones and vehicles converge on coverage through the afternoon, and the day’s bulk — imagery, camera-trap archives, collar data — drains in priority order as contact allows. By evening the ecological record is complete without a single manual transfer.
6. The debrief. The journal shows the day: when the detection left the aircraft, which hops carried it, how long the interception window really was — and the accumulated coverage map argues, with data, where the next relay mast belongs.

What each mechanism contributes

• Traffic-class triage — detections as high-class kilobytes that get through on the thinnest contact; imagery as bulk that never blocks them. Shipped, opt-in.
• Self-forming mesh — vehicles, drones and gateways relaying for each other across the dead ground, up to 8 hops, encrypted per hop. Shipped.
• Store-carry-forward — camera-trap queues drained by whatever passes; the day’s bulk delivered on the return leg, in priority order. Shipped.
• Bonding of whatever coverage exists — the ridge vehicle’s one bar of cellular becomes the reserve’s shared uplink for the hour it exists. Shipped.
• Continuous measurement — a coverage map of the reserve accumulating free with every patrol. Shipped.

The honest boundary: Atlas moves the detection; it does not make the arrest. Patrol doctrine, sensor placement, the legal chain around evidence and the response on the ground are the administration’s — and a reserve’s worth of dead ground still needs platforms passing through it for opportunistic delivery to have something to ride.

What a pilot should prove

• Detection latency from the dead ground: flag to operations screen, timestamped end to end, through at least one relay hop.
• Priority under load: a detection delivered promptly while gigabytes of imagery queue behind it on the same thin links.
• Camera-trap drain: sites emptied by overflight alone across a multi-week period, with delivery confirmed by the journal.
• The coverage map: a measured picture of one patrol sector, good enough to site the next relay mast from.

One sector, two drones, three vehicles, two weeks. The evaluation format covers the structure.