Port and harbour operations.

The mission

A morning at a container terminal: a drone runs the hull inspection of a berthed vessel, then a perimeter patrol along the fence line and the water side. The operations centre wants the live feed throughout — the seaward side of the hull, the lanes between the stacks, the corners the fixed cameras cannot see. The flight path is dictated by what needs inspecting. It is emphatically not dictated by which lanes happen to have line of sight to the ground station this particular morning — because that changes by the hour.

What breaks

Steel does not attenuate a radio link; it kills it. A drone descending between two stack rows, or working the far side of a berthed hull, is cut off from any single ground antenna as completely as if it had flown behind a mountain. What makes a terminal harder than a mountain is that the shadows move: cranes swing tens of metres of steel through the air, stacks are built and torn down every shift, and a coverage survey from Tuesday is wrong by Thursday. Cellular inside the port is shadowed by the same geometry and shared with the whole terminal’s traffic.

The classical answers each fall short. More transmitter power does not bend a signal around a stack. A fixed repeater network is designed for one stack plan and degrades as the yard diverges from it. And accepting the dropouts means the feed dies precisely in the lanes and on the hull faces that most need inspecting — that is where the steel is.

The architecture on this mission

Every participant runs the same software node: the operations centre, the aircraft, fixed nodes mounted on crane superstructures, and the patrol vehicles moving through the yard. Together they form one encrypted mesh over the port’s existing IP radios. The cranes are the load-bearing placement — the highest steel in the terminal holds line of sight into lanes no ground antenna can see, and a node bolted to a crane cabinet turns the biggest shadow-maker into the best vantage point. Patrol vehicles do the same job at ground level, mobile: whichever one is near a shadowed lane becomes the way out of it.

The aircraft itself bonds two paths — its direct radio link and cellular — and the mesh adds the relay geometries on top, up to eight hops if the yard demands it. Every link is probed several times a second and every route re-scored continuously, so as a crane swings or the drone drops behind a hull, traffic moves to the geometry that currently works before the failing path is gone. Relays forward traffic they cannot read; a node on a contractor’s vehicle is a courier, not a reader. On top of the bond, classes hold the order of sacrifice: the command channel keeps its reserved floor through the worst lane, and video sheds quality before continuity.

How the working day unfolds

1. Shift start. The crane nodes have been up for months; the patrol vehicles join the mesh on ignition. The drone powers on at the operations centre and the dashboard shows the morning’s geometry — every link, measured, not assumed.
2. The hull run. The drone works the seaward side of the berthed vessel, fully shadowed from the operations centre by tens of thousands of tonnes of ship. The feed rides through the quay crane’s node — two short hops over the hull instead of one impossible path through it.
3. Mid-shift restacking. A stack row grows by three containers and yesterday’s clean path is gone. Nothing is reconfigured: the continuous re-scoring has already shifted the route through a patrol vehicle two lanes over.
4. The worst lane. Deep between the stacks, direct link and cellular are both shadowed. The feed threads vehicle-to-crane, video sheds resolution under the narrower path, and the command channel — a few kilobits on a reserved floor — does not flinch.
5. Recovery. The drone climbs out of the lane; the direct path returns and traffic migrates back to it within the second. The pilot has flown the whole sequence without touching a network setting.
6. The debrief. The journal holds a measured shadow map of the terminal as it actually was today — which lanes went dark, which relays carried them. That map decides where the next fixed node earns its mounting bracket.

What each mechanism contributes

• Mesh relaying with per-hop encryption — crane nodes and patrol vehicles as authenticated relays that cannot read the traffic they carry; routes recomputed as the steel moves. Shipped.
• Multi-link bonding on the aircraft — direct radio and cellular as one tunnel; whichever survives the current geometry carries. Shipped.
• Continuous measurement and re-scoring — every path probed several times a second; degradation detected and traffic reweighted in under a second. Shipped.
• Class floors and ordered degradation — command reserved, video elastic, through the narrowest relay path. Shipped, opt-in.
• The event journal — a measured shadow map of the terminal per shift, for placing permanent nodes where they pay. Shipped.

The honest boundary: Atlas cannot see through steel. A lane into which no node — crane, vehicle or fixed — currently holds a geometry stays dark until one does, and choosing where relays live is the terminal’s planning decision, informed by the journal but made by people. The aircraft, the airside safety rules and the flight approvals are yours.

What a pilot should prove

• Feed continuity through a full hull inspection, including the seaward side, verified from the built-in journal.
• Relay handover without operator action as cranes move and vehicles reposition — measured, not anecdotal.
• Command-channel continuity in the worst lane of the yard, with video demonstrably shedding quality before continuity.
• A shadow map of the terminal from one working week, good enough to justify each permanent node placement.

One terminal, one instrumented drone, one working week. The evaluation format covers the structure.