Learn More

Triage, originating from Napoleon-era battlefield casualty sorting, is now guided by standardized decision trees, but strict protocol adherence remains challenging in mass casualty incidents (MCIs). DARPA's three-year challenge requires teams to use non-contact sensors on autonomous systems to detect physiological signals and identify casualties in real time without connectivity.

The challenge simulates real MCIs with darkness, dust, fog, explosions, and flashing lights, with casualties buried under rubble or hard to find. Scoring includes casualty detection rate, injury classification accuracy, and whether critical casualties are identified before the life-saving intervention window closes.

The University of Pennsylvania's PRONTO team, combining trauma surgery, robotics, and computer vision experts, used drones for rapid scene scanning and ground robots for stable imaging and vital sign capture in Phase 1 (2024). Data was fed into Meta's SAM 2 for image segmentation, supporting point or bounding box prompts.

In Phase 2 (September 27 to October 4, 2025), the team improved algorithms using Phase 1 data, building multiple parallel injury classification pipelines. DINO extracts visual features without labeled data via a custom deep neural network; Grounding DINO detects injury-related features in each image region using text prompts like 'wound?' or 'blood?'.

The system also includes body pose estimation and wound-to-bone alignment algorithms, outputting heart rate, respiratory rate, consciousness, and wound/amputation status, displayed on a mobile interface with location, description, and clinical features to help first responders prioritize resources.

There is currently little clear evidence which triage technique saves the most lives in specific scenarios; DARPA's challenge is building evaluation infrastructure through large-scale data collection. Phase 2 already incorporates real-time MCI features, with increasing difficulty each year. In Phase 3, teams will leverage experience from the first two phases to explore the latest Meta SAM and DINO models for triage.

PRONTO team lead Professor Eric Eaton notes that the team's trauma surgeons deal with such issues daily on the front lines, combining cutting-edge machine learning and robotics to develop technology that can truly save lives.