Drones For Search And Rescue

This post is also available in: עברית (Hebrew)

All around the world, search and rescue team respond to many thousands of calls annually.They search for lost travellers, hikers, and holidaymakers in mountain areas, marshes, bogs, coasts, and out at sea. These operations can be resource and time consuming, taking many hours and the combined effort of scores of trained professionals and volunteers. Well, search and rescue is about to become a bit more hi-tech and a lot more efficient.

Unmanned aerial vehicles (UAVs) can play a vital role in such operations and supplement the work performed by rescue teams. UAVs are small, relatively inexpensive, and can be deployed in large numbers. This makes them ideally suited for this type of mission. Researchers from the Dalle Molle Institute for Artificial Intelligence and the University of Zurich are about to realise this dream. They have developed artificial intelligence (AI) algorithms that enables small quadcopters to autonomously navigate around forests, to recognise and follow trails and paths.

“While drones flying at high altitudes are already being used commercially, drones cannot yet fly autonomously in complex environments, such as dense forests. In these environments, any little error may result in a crash, and robots need a powerful brain in order to make sense of the complex world around them,” says Prof. Davide Scaramuzza from the University of Zurich.

Through a pair of small cameras, like the ones in most smartphones, the UAV observes its environment. AI interprets the images, specifically seeking man made paths in them. When it finds one, the UAV is steered in the appropriate direction.

To make accomplishing this hard task possible (after all, trails can be hard to find even for a human), the researchers set Deep Neural Network (DNN) algorithms at it. DNNs are part of a branch of AI called machine learning. They learn to solve problems by solving “training example,” similar to how brains learn. To give their AI enough example, the researchers hiked for hours along several different trains in the Swiss Alps, taking more than 20,000 photos along the way with cameras attached to a helmet.

The DNN worked. When presented with a new, untested trail it was able to identify the correct direction in 85 percent of cases. That’s better than humans, who only choose correctly 82 percent of the time. There is still work to do before this solution becomes a reality. Technological challenges need to be overcome. But soon enough, this work could end up saving lives.