The AI Cowboys

· By Roger Wong Won

DARPA Triage Challenge Finals: What the Results Mean for Physical AI

The DARPA Triage Challenge Finals ended November 13, 2026 at Disaster City in College Station, Texas, with RoboScout winning the Systems competition and Disruptors the Data competition. Here is what DARPA confirmed and our read on what it means for physical AI and mission data teams.

A quadruped robot and a small fixed-wing drone in a white lab, the kind of ground and air platforms used in autonomous triage

On November 13, 2026, the DARPA Triage Challenge wrapped its Finals event at Disaster City in College Station, Texas. DARPA announced the results on November 16: RoboScout won the Systems competition and $1.5 million, and Disruptors won the Data competition and $1 million. Teams brought autonomous systems and human-machine teams into simulated catastrophes, including a tornado hitting an open-air market, with smoke, heavy darkness and complex obstacle lanes.

We build physical AI and mission data systems, and this is one of the clearest public tests of that kind of work we have seen this year. Here is what DARPA confirmed, and our read on what it means for teams building autonomy that has to work when conditions are at their worst.

What DARPA announced

The challenge exists because mass casualty incidents produce far more victims than medics. DARPA describes the goal as autonomous systems, such as drones and ground robots, that can locate casualties, identify signs of injury and send critical data to medical responders so limited care reaches the right patients first. Teams built algorithms, unmanned aerial vehicles and unmanned ground vehicles designed to work alongside the human medics who give hands-on care.

The Finals results, as published by DARPA:

Systems, autonomous platforms (scored on most lives saved on the human-machine teaming lane): RoboScout took the $1.5 million grand prize and DART was runner-up at $500,000.

Systems, technical capabilities (overall performance on component testing gates): UAS-DTU took first place at $300,000 and Guardian Angel was runner-up at $150,000. DARPA marks both as self-funded teams.

Gate awards of $12,500 each: DART won find and locate, rapid triage and trauma assessment, and RoboScout won accurate vitals.

Data competition (prediction of life-saving interventions and resource allocation, scored on first look, continuous triage and resource allocation): Disruptors took the $1 million grand prize and OCA-SENTINEL was runner-up at $300,000.

DARPA Director Stephen Winchell said the event showed that "human machine teams outperform" in large, complex environments, and noted that different teams were strongest at different parts of the job: some at identifying victims, others at triage, lifesaving or evacuation.

Our read: the scoring is the story

The grand prize was not awarded for the most autonomous robot. It went for the most lives saved on a human-machine teaming lane. That is the right metric, and it is the one most physical AI pitches skip. A system that finds a casualty but cannot hand a medic a trustworthy, readable assessment in seconds has not helped. We wrote about why the body and the environment change the engineering in embodied AI vs physical AI, and why the deployment cost lives in the field rather than in the model in what physical AI costs.

Winchell's point that teams excelled at different stages is the second lesson. Locating, assessing vitals, predicting interventions and allocating resources are separate problems with separate failure modes. In our experience, programs do better when they evaluate each stage on its own test, the way DARPA's gates did, instead of judging a whole system on one demo.

What the environment implies (our interpretation)

DARPA's published scenarios include smoke, partial and full obstructions, multi-story indoor spaces, extreme lighting and tree lines. DARPA's announcement does not detail each team's hardware or compute, so the following is our interpretation, not a reported finding. Conditions like these are where connectivity is unreliable and sensors degrade, which pushes perception and decision support onto the platform itself. That is the case for edge, on-prem and air-gapped AI, for power-efficient perception such as neuromorphic computing, and for testing noise robustness claims before anything reaches a responder.

The Data competition matters just as much. Predicting who needs a life-saving intervention, and then allocating scarce resources, is a mission data problem: labeled, validated data, honest evaluation, and a decision trail a medical director can review. That is the same discipline we bring to government and defense AI work.

Why it matters in Texas

The Finals ran at Disaster City in College Station, a short drive from our lab at UTSA in San Antonio. For Texas teams working in defense medicine, emergency response and robotics, a public, prize-scored benchmark run in-state is a concrete reference point for what human-machine triage can do today.

An open door: TRIAGE-X

DARPA also pointed to an open SBIR topic, TRIAGE-X, which seeks autonomous medical triage systems that can recognize chemical toxidromes and physically deliver the corresponding antidote or life-saving intervention. DARPA lists the topic as open until November 25, 2026. Check the official solicitation for eligibility and requirements before planning around it.

Talk to us

If you are scoping autonomy, edge perception or mission data for medical, disaster response or defense use, we can help you define the test that matters before you build. Request a briefing or email contact_us@theaicowboys.com.

Common questions

Who won the DARPA Triage Challenge Finals?

According to DARPA, RoboScout won the Systems competition grand prize of $1.5 million, with DART as runner-up at $500,000. Disruptors won the Data competition grand prize of $1 million, with OCA-SENTINEL as runner-up at $300,000. UAS-DTU and Guardian Angel placed first and second in the technical capabilities category.

Where and when were the Finals held?

DARPA says the Finals concluded on November 13, 2026 at Disaster City in College Station, Texas, and announced the winners on November 16, 2026.

Is there still a way to work with DARPA on autonomous triage?

DARPA pointed to an open SBIR topic, TRIAGE-X, for autonomous medical triage systems that recognize chemical toxidromes and deliver the matching antidote or intervention. DARPA lists it as open until November 25, 2026. Check the official solicitation for eligibility.

Further reading

See also