Imagine you’re part of a search and rescue team, racing against time to find someone lost in the wilderness. It’s a daunting task, but researchers in Scotland have developed a groundbreaking AI system that might just make your job a little easier. This innovative tool, crafted at the University of Glasgow, predicts where missing individuals might be in outdoor settings by analyzing real-world data from past incidents.
The brain behind this project is Jan-Hendrik Ewers, a PhD candidate at the University of Glasgow’s James Watt School of Engineering. He and his team studied historical data on how people behave when they get lost. From this, they created ‘simulated agents’—think of them as virtual characters—whose actions are driven by algorithms. These agents mimic different psychological states, aiming for goals like finding water or a path.
Ewers, who has hands-on experience in the rural Highlands, knows just how crucial search and rescue work is. He points out that these teams often operate with limited funding and rely heavily on volunteers. So, any tool that can help them save lives is a welcome addition.
The AI system uses these simulated agents to make informed decisions based on the environment and historical data about where missing people are usually found. Initially, Ewers wanted to use machine learning to predict the locations of lost hikers, but the lack of data made this difficult. Instead, the team turned to existing studies to identify patterns in how missing persons behave.
To test their model, the researchers deployed AI agents in a digital simulation of the Isle of Arran. The results were promising and closely matched real-world outcomes. David Anderson, a co-author of the study, noted the flexibility of this psychological modeling approach, stating it could potentially be applied to any landscape.
Looking ahead, the team is excited about integrating their model with drone technology to further enhance search and rescue missions. The results so far suggest that this approach could have global applications, pending more development and validation. Their research has been published in IEEE Access.
