Carnegie Mellon develops AI drone swarms for rescue and battlefield triage
Carnegie Mellon is developing AI-guided drone swarms to help responders triage victims and map dangerous sites. The work targets natural disasters, mass casualty events and battlefields, where emergency teams often face unstable terrain, limited visibility and major safety risks. The aim is to send coordinated groups of drones into hazardous areas before more people are placed in danger.
One effort is focused on autonomous medical assessment from the air as part of the DARPA Triage Challenge. Researchers are building AI systems that allow drones to use cameras to scan injured people from about three meters away and estimate vital signs including respiratory rate, heart rate and body temperature. The swarm approach is designed to capture the same victim from multiple angles, while smaller robots can enter tight spaces and larger drones can carry heavier equipment such as cameras.
The medical models can detect subtle changes in facial color to estimate heart rate, but moving the technology outside the lab has been difficult. Rain, shadows, darkness, smoke and faces partly hidden by rubble, mud or blood can degrade performance in ways that controlled laboratory tests do not capture. Researchers spent about two years developing more robust AI methods, and the current system remains an advanced proof of concept that is still several years from deployment.
A second line of work is aimed at search and rescue in caves, earthquake rubble and other places where human access is slow or dangerous. The goal is to miniaturize drone teams so responders could release multiple vehicles into a cave or disaster zone, have them search quickly, build a shared map and return safely with locally stored data. The need is acute because survival rates for people trapped in natural disasters can fall sharply after 72 hours, while extreme sites often lack internet, cloud links or reliable communications with operators.
The teams are also testing physical capabilities that could make small drones more useful without adding major payload. One approach allows a drone with a downward-facing camera to flip over and capture images of a ceiling or overhead structure. If the systems mature, drone swarms could scan flooded towns, find survivors waiting on rooftops, fly at speeds up to 100 kilometers an hour and give responders faster information with less risk to human crews.