AI drone to track down missing hikers
Locating missing persons in rough terrain can take a long time, even with drones from the air. Artificial intelligence can improve the search.
(Image: Mark Poprocki/Shutterstock.com)
A team of engineers at the University of Glasgow has developed a drone that can track down lost hikers using artificial intelligence (AI). To train the AI, the scientists relied on, among other things, data that combines the paths of lost hikers and people found by search teams.
Drones are very well suited to search tasks. Missing persons can be tracked down more quickly from the air thanks to the better overview. Larger areas can also be searched in less time. In the study "Deep Reinforcement Learning for Time-Critical Wilderness Search And Rescue Using Drones", which has now been published as a preprint on Arxiv, scientists from the University of Glasgow wanted to find out to what extent an AI can improve the location of missing persons in the rugged terrain of the Scottish Highlands.
Combined training data
The scientists created an AI model that they trained using various types of data. This included global geodata on routes taken by missing persons and routes taken by search parties. The location where the missing persons were found was also considered. The researchers also considered the age of the people, the reason for the hike, whether they were traveling alone or in a group, or whether they were on foot or on horseback. The scientists combined this data with the geographical conditions on site, such as rivers, streams, roads, mountains and open terrain. They also added specific data on the geography of Scotland. They trained the model using deep reinforcement learning. Furthermore, they ran a fictitious search several million times until the AI was able to spit out the most likely paths that a lost hiker would take.
The scientists implemented the AI in a drone to get it to first fly the most likely paths that a lost hiker would take.
The researchers tested the system in comparison to conventional search methods, such as flying evenly over an area. Using the AI method, lost hikers were found 19 percent more often. Conventional methods only deliver a result of between eight and twelve percent.
The scientists believe that the system has not yet reached its performance limit. They therefore want to use more data to train the AI. They hope that this will make it possible to find lost hikers more quickly and save lives.
(olb)