The newest member of search and rescue can’t track you by your footprints. It doesn’t know how to read the itinerary you left in your car. It can’t detect your scent or your body heat. Instead, it finds you by following the sound of your screams.
As you might have gathered, this new recruit is not a person: It’s a drone. Developed by researchers at the Fraunhofer FKIE lab in Germany, the robotic aircraft is designed to zero in on the sound of people in distress, pinpointing their location with an array of microphones, a built-in computer, and a sophisticated artificial intelligence program. If you think that sounds creepy, you’re not alone. Mashable has compared the drone to something out of Terminator; we’ll admit that to a lost hiker, it might not be as comforting as seeing a human. But, like the Terminator, you might end up having to come with it if you want to live.
“In a post-disaster situation [where] every minute is crucial to find victims, we decided to exploit the possibilities offered by drones to rapidly fly over extended, maybe hard-to-reach or dangerous areas,” Macarena Varela, a fellow with Fraunhofer FKIE who recently presented the project at the 180th Meeting of the Acoustical Society of America, said in an email.
When they’re operational, the drones will occupy a unique niche in the first-responder ecosystem. For SAR teams racing the clock to locate a lost or injured hiker, aircraft are a valuable tool, allowing them to quickly scan large areas for signs of life. But because few to no volunteer teams can afford to operate their own helicopters, rescuers are usually left to request assistance from law enforcement agencies, National Guard units, or private companies, resulting in delays and straining limited resources. Cheaper drones and ultralight aircraft help to make up some of the difference, but without the expensive and complex thermal imaging cameras and other sensors that the larger craft use, they’re limited to searching visually, making them ineffective at night or in terrain with dense tree cover.
Fraunhofer FKIE’s still-in-development project aims to sidestep that problem by capitalizing on a common behavior in hopelessly lost people. Namely: They scream for help. Starting with a commercial quadcopter (in this case a DJI Matrice M600), designers added an array of small microphones connected to a miniature computer that both processes and filters out the background noise from the wind and rotors and calculates the location of screams.
To make the drone feasible, the team from Fraunhofer FKIE had to solve several technical problems. One of the biggest was making the system light enough to fly: Varela says that she and colleagues Wulf-Dieter Wirth and Manfred Okum successfully created a microphone array that could pinpoint sounds in 2016, but the system, built with condenser microphones, was too heavy for human searchers carry on foot, let alone for a quadcopter. It wasn’t 2018 that Wirth suggested using microelectromechanical system microphones—ultralight etched-silicone mics often used in phones, laptops, and other electronic devices—that they were able to successfully modify their invention to the point where they could install it on a drone. To determine which direction a hiker’s screams are coming from, the drone relies on an audio processing technique called beamforming: Its microphones (the current iteration of the search-and-rescue drone uses 32) are arranged in a roughly spherical array called a crow’s nest, and it processes the signals from each microphone to maximize the array’s sensitivity to sounds in particular directions.
The project isn’t the first attempt to build autonomous drones for search and rescue. Researchers from Germany’s NEC Laboratories Europe are developing a drone that zeroes in on cell phone signals to locate lost people. In 2017, Lockheed Martin demoed a pair of autonomous helicopters capable of locating and retrieving a lost hiker in a simulation. (Similar aircraft are reportedly already in use by the United States military, a stark reminder that robots that find people are dual-use.)
While Fraunhofer FKIE’s drone is still in development, the initial trials have been promising. In tests conducted in open areas, the drone has been able to determine a person’s location within a few seconds. The team behind the drone—which Varela points out is more ethnically and nationally diverse than the field typically is, with colleagues from “almost every continent”—is currently pursuing it as a passion project using some funding from the institute. But with all the interest the drone has generated, they don’t rule out trying to bring it to market in the future.
Before the drone reaches that point, however, there’s more work to do. The researchers are still experimenting with machine learning to improve the drone’s ability to filter out noise and more accurately pinpoint lost people; Varela says that they plan to double the number of microphones in the drone’s array. It may be a while before we see the Searchinator in action.
“What has helped us is that our group is very positive and always willing to find solutions to challenges instead of focusing on them…If we encounter a challenge, e.g. something breaks, we will find [the] means to repair it or replace it, so that we can continue with our measurement,” she says. “In other words, a lot of engineering.”