The problem

When an avalanche buries someone, every minute counts.

People in avalanche terrain often carry a small radio beacon. If they are buried, rescuers sweep the slope with a receiver to find its signal. But searching a large, unstable slope on foot is slow, and it puts the rescuers at risk too.

A drone can cover the same ground much faster, from above, without putting anyone else in danger.

Our idea

Give the drone a radio receiver.

The beacon sends out a steady radio signal. The closer you are, the stronger you hear it. If a drone measures that strength at lots of places it knows the position of, a computer can work backwards to find the most likely source.

The beacon

A small transmitter that sends out a steady tone. The person at risk carries it. In our tests, a laptop with a radio plays this role.

The drone

Flies a planned search pattern on autopilot. It carries a radio receiver, a GPS, and a small computer that records everything it hears.

The software

Matches every signal reading to the place it was taken, then draws a probability map of where the beacon most likely is.

How a search works

Fly a grid, listen, and narrow it down.

An illustration of the idea (not real data): the drone flies back and forth over the search area, and the signal it hears (darker dots) gets stronger as it passes near the beacon. The software turns those readings into a map of where the beacon probably is.

1

Plan

Draw a "lawnmower" search grid over the area in the flight-planning software.

2

Fly

The drone follows the grid on autopilot, low and steady over the ground.

3

Listen

The receiver measures the beacon's signal about five times a second. The GPS notes where the drone was for each reading.

4

Estimate

The software tries thousands of possible beacon positions and scores how well each one explains the readings. The result is a probability map.

5

Zoom in

Fly a smaller, tighter pattern over the most likely area, and repeat for a sharper answer.

Meet the drone

A small quadcopter carrying a listening kit.

Close-up of the drone held by hand, showing the GPS on top and the electronics underneath — The GPS sits on top; the computer and radio receiver are mounted underneath.

The drone seen from above, on the grass — From above: four arms, one on-board listening kit.

AutopilotA Pixhawk flight controller running ArduCopter keeps the drone steady and follows the planned route.

GPSTells the drone, and our software, exactly where it is.

Radio receiverAn ADALM-Pluto software-defined radio with an antenna. It measures how strong the beacon's signal is.

On-board computerA Raspberry Pi. It records every signal reading with the drone's position. It only listens; the autopilot does the flying.

Ground stationA laptop running QGroundControl, used to plan the search grid and watch the flight.

The test day

Out in an open field.

We set up the beacon at a measured spot, planned the search in QGroundControl, flew it, and compared the software's answer with where the beacon really was. A tape measure marked out the exact positions.

QGroundControl showing a lawnmower search grid over the field — The search grid, planned in QGroundControl.

QGroundControl showing a smaller, tighter pattern — A smaller, tighter pattern for a closer look.

A laptop with a blue ADALM-Pluto radio next to a tape measure — The beacon side: a laptop with a radio, and a tape measure.

The wide field with the laptop in the foreground and the drone in the distance — The test field.

Side view of the drone showing the radio receiver mounted under the frame — The receiver (blue) hangs under the frame.

The drone on the grass before a flight — Ready for a flight.

Watch the flights with live graphs, and see all the photos →

What we found

It works end to end. Now we make it sharper.

11 mIn flight 005, one pass of the search grid placed the beacon about 11 metres from its true position.

A clear peakWhen the drone flew closest to the beacon in flight 005 (about 53 seconds in), the signal rose well above the background.

1.6 mIn flight 006 the drone came down 1.6 metres from the beacon. That was a shorter, higher flight, so its position estimate was much rougher (about 17 m off).

What we've learned. Real radio signals are messy. Readings jump between strong and weak even when the drone is hardly moving, so the software has to cope with a lot of noise. This is an early proof of concept from a small number of flights: the whole chain from planning to a map works, and the next job is making the estimate more precise.

Built with

ArduPilot / ArduCopterQGroundControlPixhawkRaspberry PiADALM-Pluto SDRPythonPlotlyLeaflet

An early prototype. All numbers on this site come from a small number of real test flights.