In mass casualty situations, quickly identifying and prioritizing injuries can be the difference between life and death. Traditional triage systems rely heavily on trained human responders, but in complex or hazardous environments, human access may be delayed or limited. Robotic systems offer a promising solution to this problem by enabling remote, scalable, and fast health assessments. The recently launched DARPA Triage Challenge aims to accelerate progress toward autonomous triage robots that can evaluate the physiological state of human casualties using non-contact sensing methods.

This thesis presents our contributions developed as part of the DARPA Triage Challenge, combining methods for physiological sensing with autonomous robotic navigation. We present two key algorithms for casualty assessment: (a) a lightweight algorithm for estimating respiration rate using monocular RGB video, and (b) a geometry-based technique for 3D human pose estimation, applied to the detection of possible amputations. On the navigation side, we design an autonomous system that enables a ground robot to move around a casualty and collect multi-view observations critical for robust physiological inference. Finally, we show how these methods work together within a deployed robotic platform, helping move closer to real-time, autonomous casualty assessment in the field.