UMD's rescue bots team up to triage disaster victims

In a mass-casualty disaster, the cruelest bottleneck is almost always math: too many injured people and far too few medics to check them all in time.

To solve that gap, the University of Maryland’s RoboScout team is deploying an autonomous tag team. Drones scout from above, robot dogs triage on the ground, and human medics get vital seconds back.

The project is part of DARPA's three-year Triage Challenge. Following large-scale field tests in Perry, Georgia last month, the team showed off their autonomous rescue stack on Oct. 16 at the Maryland Robotics Center.

Drones map the site, robot dogs take vitals

The workflow starts in the air. Autonomous drones, co-designed by project engineer Josh Gauss, fly into the disaster zone first to spot bodies and injured victims, even in newly added night scenarios.

Once survivors are located, the drones beam real-time coordinates down to ground medics and an autonomous ground scout: a Boston Dynamics Spot dog named Apollo.

Apollo carries an onboard AI computer. In the demonstration, the quadruped walked up to a mannequin, announced it was there to help, and scanned for vital signs.

From there, the robot dog uses an onboard large language model to hold a conversation with the patient, gathering symptom details before beaming a full assessment straight to medics' phones.

It also uses raw acoustics. If a patient is critically hurt, the robot physically shouts out to nearby medics. As project lead Derek Paley noted, emergency workers usually have their hands and eyes full, but their ears are open.

After triage, the robot stays in place as a sentry monitor, alerting doctors if vitals worsen. In the simulations, which mix mannequins and human actors, the robot maintains a strict two-meter distance for safety.

Why it matters

Triage under pressure is brutal, and every minute spent taking pulses on walking wounded is a minute lost on someone bleeding out.

What makes RoboScout smart is that it does not try to turn a robot into a surgeon yet. Instead, it turns the robot into an automated sensor and communications relay. Offloading visual scanning and triage interviews to AI frees human doctors to do what they actually specialize in: saving lives.

There are clear hurdles ahead. Keeping a strict two-meter buffer shows that close physical interaction with fragile victims is still too risky. Real-world rubble, comms blackouts, and screaming crowds will test onboard language models far harder than a controlled drill.

Still, pairing aerial reconnaissance with ground-based conversational AI turns the chaos of disaster response into structured data. Paley expects these systems to be active first responders within a few years, and judging by this demo, the playbook is already taking shape.