The most dangerous moment in autonomous casualty evacuation is the handoff, the instant when control passes between the machine and the human operator. Full autonomy is manageable. Full manual control is manageable. The seam between them is where things break.
After several days at MHSRS 2026, I think that insight reaches well beyond unmanned vehicles. Military medical innovation keeps running into trouble at its handoffs: between technology and operator, between a diagnostic output and clinical judgment, between a funded prototype and an acquisition program, and between engineering intent and scalable production. The technology on display was impressive. The harder problem was building systems that people can trust, field, buy, and sustain under conditions that leave very little room for error.
MHSRS is the Department of War’s foremost scientific meeting on military medicine. Thousands of attendees moved through the Gaylord Palms across hundreds of presentations and posters, spanning haemorrhage control, blood products, traumatic brain injury, burn care, autonomous evacuation, and biomanufacturing. The scale of research funded through the US defence health enterprise is hard to appreciate until you are standing inside it.
The requirement collision
Start with what makes military medtech unlike its civilian cousin. A device here has to satisfy clinical, operational, human, and industrial requirements at the same time. It must meet strict regulatory and quality expectations, move from concept to field quickly, keep working in an austere and contested environment, and remain usable by non-experts under stress. Civilian products rarely have to carry all of those at once. In this market it is the starting line, and every handoff that follows is where one of those requirements tends to get dropped.
Handoff one: from automation to human control
COL Conrad Wilmoski of the US Army Medical Center of Excellence set the operational stakes. He framed casualty evacuation as a priority capability gap for large-scale combat operations, driven by the prospect of unsustainable losses, and worked through it with a DOTmLPF-P analysis, the military’s method for examining a problem across doctrine, organization, training, materiel, leadership and education, personnel, facilities, and policy rather than as a pure equipment buy. One distinction stuck with me. CASEVAC moves casualties on whatever non-medical platform is available and often without medical personnel aboard, while MEDEVAC uses dedicated, marked, staffed platforms. Much of what Ukraine has forced is the former, as contested airspace, shrinking evacuation capacity, and fewer medical formations forward push units to move casualties on whatever will roll or fly.
That is the setting for the talk that reframed the topic for me, from Maj. (Res.) Ariel Braverman of Ben-Gurion University and the IDF Medical Corps, on the human factors of unmanned medical evacuation. His argument was that an unmanned system is not one machine and one operator. It is a distributed cognitive system spanning frontline medics, remote vehicle operators, remote clinicians, and commanders, with the automation contributing monitoring, data fusion, decision support, and navigation. He laid automation out on a spectrum from teleoperation through supervised and conditional autonomy up to full autonomy, and located the risk at the handovers, when control and responsibility pass between person and machine.
He was direct about the variables that decide whether any of this survives contact with the field: trust, workload, situational awareness, and accountability. Remote operators lose direct perceptual access to the scene. They drift toward complacency in quiet stretches, then face sudden workload spikes when something fails. And the accountability question has no clean answer yet. He asked who is responsible for a patient’s condition during autonomous transport, and the room had no framework to offer.
The line I keep returning to, because it maps so closely to how we work at Inertia, was this: “Human-centered design is a safety requirement, not an option.” A device that non-experts have to operate while under fire, injured, and afraid cannot treat usability as polish added at the end. The clear behavioural cues, the simplified operation, the fail-safe handover, and the transparency that lets an operator calibrate trust are the product. Braverman went further and paired each human factor with a design response, matching workload to adaptive levels of automation and building operator competence through scenario-based training. That turns human factors from a talking point into an engineering specification.
That way of thinking is second nature in medical device development, where human factors and use-related risk are design inputs rather than finishing touches. It is far less common among teams building autonomy, where the platform gets the attention and the person using it comes later. Coming at these systems from a regulated-device background, that is the difference I notice first, and it is where a medtech lens reads them differently than a pure autonomy one does.
A second layer of autonomy
It is not only the vehicles. The most striking autonomy at MHSRS was not moving the casualty but managing them. Closed-loop systems now titrate inspired oxygen to a target saturation on their own, and autonomous ventilators adjust support continuously without a clinician at the bedside. These are aimed squarely at prolonged casualty care in austere settings, where oxygen is a scarce resource, evacuation is slow, and there may be no respiratory therapist forward. One closed-loop system reported making several hundred oxygenation adjustments in the time a clinician would make a handful.
That is a sharper version of the same handoff. When a machine makes hundreds of decisions an hour, the clinician’s job shifts from doing to supervising, and the risk moves with it: complacency in the quiet stretches, trust that is hard to calibrate, and no clean answer to who is accountable when the system runs unattended and something drifts. The clinical-autonomy teams are wrestling with this directly. It is the wider autonomy world, the one fixated on the platform, that still tends to wave the question off.
Handoff two: from objective signal to clinical action
Traumatic brain injury ran through the program as its own track, and it exposes a quieter handoff.
Concussion and blast injury still lean heavily on self-reported symptoms, and service members underreport their symptoms, often substantially. That pushes the field toward objective measures, including eye-tracking and pupillometry at or near the point of care, and toward better blast exposure monitoring. A comparison of three eye-tracking devices, published in the MHSRS proceedings, found the strongest performer reached an AUC of 0.95 on acute injury in training data, then fell to between 0.60 and 0.75 on a withheld validation set, and to chance on chronic cases. The signal is real and it is not yet stable.
Look at the shape of the problem. The technology can produce an objective signal. Whether a clinician trusts that signal enough to act on it, and to make a return-to-duty call, is a human factors question, the same one Braverman raised about autonomous transport. A signal that the user does not understand or believe does not change the decision. The handoff from data to action is where the value is either captured or lost.
Handoff three: from funded prototype to fielded capability
The next handoff is the one companies feel most sharply, and it was the subject of some unusually frank acquisition sessions.
Speakers kept returning to the “valley of death,” the gap where a technology has been developed but has no clear path into a program of record. One official described the risk of the government behaving as a poor business partner, asking industry to build something and then stepping back when it is time to buy. Another floated accepting a good-enough joint solution now over a perfect service-specific one later. This is the seam between a working prototype and a program that will actually procure, field, and sustain it, and a promising device can die in that gap for reasons that have nothing to do with whether it works.
Handoff four: from engineering intent to manufacturable scale
The handoff that decides whether any of this reaches a warfighter at volume was also the one MHSRS talked about least. Biomanufacturing had a real presence on the poster floor, but most of that work is about producing a biologic or a blood product, not about industrializing a device so that the thousandth unit performs like the first. This is a research meeting, so that emphasis is expected. It is still a gap worth naming, because it is where a great deal of promising technology quietly stalls.
A prototype proves that something can work once, in trained hands, under controlled conditions. It says very little about whether that performance survives tooling, tolerance stack-ups, supplier variation, process capability, and the test and inspection regime a fielded medical device demands. In a market where the device has to hold up in an austere environment and be operated by non-experts under stress, manufacturing variation is not only a cost problem. It is a readiness and safety problem. A unit that drifts out of specification at scale fails at exactly the moment the requirement collision is least forgiving.
And these outcomes are largely set early, by decisions about architecture, materials, tolerances, suppliers, and testability that are made long before anyone files them under manufacturing. By the time a program reaches transfer to manufacturing, the expensive choices are already locked. It is the same pattern that runs through every other handoff here. The risk is created early and discovered late.
For dual-use medtech companies, including Canadian ones
Canadian companies were well represented at MHSRS, and the technical quality was high. Deep Breathe, Thornhill Medical, MY01, Moonrise Medical, and Xpan were all there, a strong showing for a country our size. The structural challenge is not the technology. Canadian medtech and defence companies are, as a sector, stronger at research and prototypes than at industrialization, which makes the manufacturing-and-scale handoff the one that most often decides whether strong technology becomes a fielded product or stops at a successful demonstration.
Strong technology alone does not create a US defence-health opportunity either. The product has to be translated into an operational requirement, connected to the right acquisition pathway, and designed around the restrictions that affect foreign participation, manufacturing, data, and export-controlled technology. Funding and teaming are the questions companies ask about first. The Medical Technology Enterprise Consortium is one important route into the US military medical ecosystem, working across USAMRDC, the Defense Health Agency, and other government partners to move prototypes toward capability, but it is not a simple grant program. Each solicitation sets its own maturity, teaming, cost-sharing, regulatory, and contracting requirements, and its membership spans more than twenty countries, so the right structure is opportunity-specific. It may mean direct membership, a US partner, a prime-subcontractor relationship, or another route entirely. Reading the individual solicitation matters more than any general rule.
The part worth emphasizing is timing. None of this work can wait until the prototype is finished. By then some of the most consequential architecture, supply-chain, and evidence decisions may already be hard to reverse. In this market, as in complex hardware generally, the costly risks are created early and discovered late.
Where Inertia fits
MHSRS reinforced a belief that is central to how we work at Inertia. In complex, high-consequence hardware, the hardest failures are usually structural rather than purely technical.
Military medtech intensifies that. Clinical intent, human factors, autonomy, regulatory evidence, ruggedization, manufacturability, and acquisition strategy cannot be run as separate workstreams that eventually meet at the end. The critical decisions have to stay coherent as the product crosses from concept to verification, field evaluation, transfer, and production. Every one of those transitions is a handoff, and each is a place where intent can quietly leak out of the program.
We are not a military operator, and we should not pretend to be one. Our contribution is different. Designing for the person at the seam, treating human factors as a design discipline rather than a finishing touch, is what medical device development teaches, and it is a lens still uncommon among teams building autonomy. We bring that together with the manufacturing side, translating demanding operational and clinical requirements into systems that can be verified, built at scale, and used reliably, and keeping those requirements intact across the interfaces where they usually fragment.
The lasting lesson from MHSRS was about coherence. Military medical technology has to cross its handoffs without losing the intent that made it valuable in the first place.
Questions this raises
What is MHSRS?
MHSRS is the Department of War’s foremost scientific meeting on military medicine. Thousands of attendees move through hundreds of presentations and posters spanning haemorrhage control, blood products, traumatic brain injury, burn care, autonomous evacuation, and biomanufacturing.
What were the main themes at MHSRS 2026?
The technology on display was impressive. The harder problem was building systems that people can trust, field, buy, and sustain under conditions that leave very little room for error. Military medical innovation keeps running into trouble at its handoffs rather than in the technology itself.
What are the four handoffs in military medical innovation?
From automation to human control, when responsibility passes between operator and machine. From an objective signal to a clinical action, when a clinician has to trust a measurement enough to act on it. From a funded prototype to a fielded capability, the valley of death before a program of record. And from engineering intent to manufacturable scale, where the thousandth unit has to perform like the first.
If you are adapting a civilian medical technology for far-forward care, or moving a funded military-medical prototype toward something fieldable and manufacturable, I would be glad to compare notes on where transition risk is accumulating.
