How AI-Powered Drones Could Improve Search and Rescue

Over a decade ago, one of the first ways public safety agencies began using drones was to look for missing people.

And that’s because the value of drones in search and rescue (SAR) is clear—a hiker goes missing, so you put a drone in the air to see if you can spot them.

Credit: DJI

But this can be like looking for a needle in a haystack. If the person is missing in a large wilderness area, where do you start your search? And what’s the most efficient way to search an area by drone anyway?

Time is often the key factor in search and rescue. For people missing without water in hot climates or in cold places where hypothermia could set in, the difference of a few hours could be life or death.

At the moment, drone searches are carried out using a combination of intuition and search theory. (Search theory involves using statistics to make a map showing the most to least probable places where a person could be, then creating a search path that starts at the point of highest probability before going to the next highest, and so on.)

And this approach has been pretty effective—way more effective than searching on the ground.

But could there be a better way?

How AI-Powered Drones Could Find People More Quickly

Researcher Jan-Hendrik Ewers at the University of Glasgow, in Scotland, has used AI and machine learning to show that yes, there is absolutely a better way to plan drone searches.

Think about it: When a person uses search theory to create a path for their drone search, it’s going to be pretty thrown together.

That’s because they’re already out in the field, the clock is ticking, and they just need to start moving so they can find the missing person.

Missing hikers wave at a drone to alert rescuers of their presence

To improve on this, Ewers trained an AI model with data from thousands of search and rescue efforts that have taken place all around the world. He’s fed the model data such as the nature of the incident—was the person hiking, or did they have dementia—as well as their age and where they were found.

Using all this information, the model made an algorithm that can take the details of a new incident and create a probability map showing where it’s most likely someone might be found. Then, it  creates a drone search path that has the highest likelihood of finding the missing person.

Is It Actually Better?

To test the model, Ewers had it go head-to-head with two common search patterns for drones—a “lawn mower” back and forth path, and a path created by a less sophisticated algorithm—in a series of virtual tests.

Credit: Skydio

The results were clear: The AI model made the best search path by far, both for creating a path that would require the least distance flown and for creating a path with the highest likelihood of finding the person.

• Lawn mower path—found the person 8% of the time
• Less sophisticated algorithm—found the person 12% of the time
• New AI algorithm—found the person 19% of the time

While the tests show the model can work, it’s not ready for prime time just yet.

Ewers says there is still more testing needed, and that the model needs to be trained on a lot more data. Also, the model won’t work if you don’t have enough data on hand, which means not all SAR scenarios will be able to use it.

We have this problem in search and rescue where the training data is extremely sparse, and we know from machine learning that we want a lot of high-quality data. If an algorithm doesn’t perform better than a human, you are potentially risking someone’s life.

– Jan-Hendrik Ewers, Researcher at the University of Glasgow

Even when the model is perfected and in use, it will still be people who play the crucial role in determining which approach is the best one for finding a missing person. In some instances, the algorithm could make mistakes and suggest focusing search efforts on an unlikely area, or even an impossible area.

When they’re ready, AI-powered drones will be a powerful new tool in the SAR toolbox. But ultimately it will be up to the intuition and experience of the drone pilot and their colleagues to choose the best approach.