Drones learn to search forests for lost hikers

190216 forestdrones 1
With the new drones, missing persons can be found and rescued quickly in forests and mountain areas. – UZH/USI/SUPSI

Imagine: on a bushwalk in the wilderness, you fall and break your leg. Your personal locator beacon is nowhere to be found, you've run out of food and it's getting dark. Far from any mobile phone reception, your only option is to wait for help … which, in the middle of nowhere on a rarely used track, could take days.

But after a few hours you hear a buzzing noise. Suddenly, around the bend flies a small drone. And as you wave it down, it sets off a location beacon, alerting emergency workers to your exact position.

This rescue scenario is one step closer. A team of Swiss and Italian scientists taught drones to recognise and follow forest tracks on their own. They were even able to detect faint man-made trails better than people could.

Autonomous drones are already used at high altitudes, but they struggle in the complex environment of, say, a forest. This is because it takes an enormous amount of processing power to see and recognise objects or, in the case of these drones, paths.

“Interpreting an image taken in a complex environment such as a forest is incredibly difficult for a computer," says lead author Alessandro Giusti from the Dalle Molle Institute for Artificial Intelligence in Italy. "Sometimes even humans struggle to find the trail!"

So Giusti and his colleagues added two small cameras – similar to those found in smartphones – and wrote powerful artificial-intelligence algorithms to interpret images.

190216 forestdrones 3
To ‘train’ the algorithms of the drones, the team hiked several hours along different trails in the Swiss Alps. – UZH/USI/SUPSI

They employed what's called a "deep neural network". As a brain learns from experience, the computer algorithm learns to solve complex tasks from training examples.

To get enough data to "train" their algorithm, the team spent hours hiking in the Swiss Alps and took more than 20,000 photos of tracks using the cameras attached to a helmet.

When they plugged that data into the deep neural network and presented it with a new, previously unseen image, it was able to find the correct direction 85% of the time. This is better than humans, who guessed correctly 82% of the time.

190216 forestdrones 2
The newly developed software is based on an adaptive network. – UZH/USI/SUPSI

The researchers admit much work is needed before fleets of AI drones are released into forested areas to search for missing people. "But small flying robots are incredibly versatile, and the field is advancing at an unseen pace," says Dalle Molle Institute for Artificial Intelligence chief Luca Maria Gambardella in Switzerland, who was also involved with the work.

"One day robots will work side by side with human rescuers to make our lives safer."

The work was published in IEEE Robotics and Automation Letters.

Please login to favourite this article.