Avalanche control and ski-patrol operations.
Control work happens before the lifts open, in the bowls and couloirs the resort’s own network cannot see into — and casualty response happens in the same terrain, in weather that grounds helicopters. This mission is about using what the mountain already has: the top lift station as a relay, the patrol drone as eyes, and a network that treats the ridge as geometry to route around rather than a reason to fail.
Ski patrols, avalanche control teams and mountain rescue services working the terrain behind the ridgeline.
Blast reconnaissance and casualty response happen in bowls the resort network cannot see into, in weather that keeps helicopters on the ground.
A relay node at the top lift station bridges the shadowed bowls; casualty coordinates ride duplicated command-class traffic; video queues across gaps.
The drone’s companion computer, a small node at the lift station and the patrol room’s workstation. One Linux binary, one configuration file.
The mission
Forty centimetres of new snow overnight, wind out of the northwest, and the resort opens in three hours. The control team is on the ridge at dawn: charges into the loaded slopes, then reconnaissance of every blast result before a single guest rides the lift. A patrol drone drops into the bowls to read the debris and the crowns while the patrol room below tracks the picture. Later, mid-season and mid-morning, the same infrastructure earns its keep the harder way — a skier is caught in side-country, and the first accurate fix on the burial site comes from the aircraft that can be over the bowl in two minutes, in cloud that keeps every helicopter at base.
What breaks
The resort’s radio and cellular coverage was built for the front side — the pistes, the lift lines, the restaurants. The work happens on the other side of the ridge, and VHF-and-up radio does not bend around rock: the moment the drone drops into the bowl, the direct link to the base is not degraded but gone. The bowls are, precisely, the terrain the network cannot see — that is why they are dangerous and why they are patrolled.
The classical answers fit badly. Flying high enough to hold the link wastes the camera on the one morning it matters most. A dedicated repeater network for the back bowls is a capital project the mountain department does not own. And in a burial, the cost of a dead zone is measured in minutes of oxygen: a coordinate that arrives when the drone climbs back out is a coordinate that arrived late.
The architecture on this mission
The load-bearing observation is that the mountain already has the relay site. The top lift station sits on the ridgeline with power, a mast and a maintained building — it sees the base area on one side and down into the bowls on the other. A small node there turns the drone’s impossible long link into two short hops, each its own measured, encrypted session, with routes recomputed continuously as the aircraft moves. A second aircraft can do the same job over terrain no lift reaches; a lost or swapped relay is detected and routed around in roughly a second at default settings.
Traffic is classed by what it costs to lose. A casualty coordinate is a few hundred bytes that must arrive; it rides command-class on a reserved floor and, when measured loss climbs, adaptive redundancy duplicates it across every path that exists. Reconnaissance video is elastic — it sheds quality when a hop narrows and, in the bowls even the relay cannot see into, it queues on the aircraft store-carry-forward and delivers in priority order on the climb-out. The result is late imagery instead of no imagery, and coordinates that were never at risk.
Every leg is encrypted with keys rotated automatically every two minutes, and every leg is measured continuously — several probes per second, logged. Over a season those logs become a coverage map of the mountain: which bowls the lift-station relay holds, which need the second aircraft, and where a permanent node would pay for itself.
How the control morning unfolds
- Dawn, before first lift. The control team rides up in the dark. The node at the top lift station has been on all winter; the drone powers up at the ridge, authenticates by key and joins. The patrol room sees both hops’ measured state before the first charge is placed.
- Blast reconnaissance. The drone drops into the first bowl and the direct link to the base dies behind the ridge — as it does every morning. Traffic re-forms through the lift-station relay without a switchover; the crown and the debris field stream to the patrol room while the team is still on the ridge.
- The deep couloir. One pocket sits below a rock band even the relay cannot see into. The video queues on the aircraft, the drone finishes the pass and climbs, and the queued footage delivers in priority order on reconnection. The avalanche forecaster reviews it four minutes after it was shot, not never.
- A casualty. Mid-morning, a skier is caught in side-country. The drone is over the bowl in minutes, in cloud that grounds the helicopter, and the burial-site fix rides duplicated command-class traffic through the relay to the patrol room and the descending team — a few hundred bytes that arrive regardless of what the video is doing.
- Weather closes. Wind and cloud sit on the ridge through the afternoon. The links thin; video sheds quality; the drive channel and positions hold their reserved floors. The aircraft keeps working weather that stopped everything else that flies.
- The debrief. The journal holds a per-hop, per-minute record of the morning — which bowls held, where the shadow really starts, when each blast result was delivered. Over the season that record decides where the second relay goes.
What each mechanism contributes
- Mesh relaying — the lift station and any second aircraft are authenticated relay nodes; two short hops replace one blocked long one, routes recompute as the drone moves, and a swapped relay heals in about a second. Shipped.
- Class priorities — casualty coordinates and the drive channel rank above video; the moment a hop narrows, the stream sheds quality but control never contends. Shipped, opt-in.
- Adaptive redundancy — the coordinates that matter are duplicated across every available path as measured conditions worsen, and duplication scales back when they recover. Shipped.
- Store-carry-forward — footage from the bowls no relay can see into queues on the aircraft and delivers on reconnection, priority first. Shipped, opt-in.
- Continuous measurement — a season-long coverage map of the mountain accumulates for free, deciding where the next relay node earns its place. Shipped.
The honest boundary: Atlas contributes the network — the avalanche judgement, the blast decisions, the aircraft’s cold-weather endurance and the rescue itself are the patrol’s. A bowl no relay can see into stays dark until a relay is placed where it can see in; that is geometry, and the software’s honest contribution there is queued delivery and a measured map of exactly where the shadow begins.
What a pilot should prove
- Continuous command and telemetry for the drone across named bowls where the direct link to the base measurably fails.
- Delivery of blast-reconnaissance footage queued through a shadow zone — complete, in priority order, timestamped by the journal.
- A casualty drill: burial coordinates delivered through the relay during the worst measured interval of the week, against a stopwatch.
- A per-morning link record from the built-in measurement, good enough to choose the second relay site from.
One resort, one drone, one week of control mornings. The evaluation format covers the structure.