The Pentagon wants swarms of robots small enough to fit inside a combat medic's kit. Their job: drag a wounded soldier 10 meters to cover. Inject drugs. Wrap around a bleeding limb and tighten until the bleeding stops. And do it all without a human giving orders.
→ Non-compressible torso hemorrhage (NCTH) is the leading cause of potentially survivable combat death. No tourniquet can reach it.
→ Two parallel programs address the same gap: MASH (internal bleeding, robotic surgery) and Medical Swarm Robotics SBIR (extraction, hemorrhage control, drug delivery). The gap is the time between injury and a surgeon's hands.
→ Phase 1 starts summer 2026. Functional prototypes expected within 24 months.
A soldier is hit in the torso. The wound is internal: no external bleeding, no point where a tourniquet can go. A medic reaches him within minutes, but the bleeding is inside the abdomen. There is no equipment to find it, let alone stop it. The soldier dies before evacuation arrives.
This is not a hypothetical scenario. In the last two decades of US deployments, 87.3% of pre-hospital mortality occurred before a casualty reached a medical treatment facility. Among deaths judged potentially survivable, hemorrhage dominated: 90.9% of cases. And the single hardest category, non-compressible torso hemorrhage, has no field solution today.
As we wrote in July, AI-driven diagnostic software for hemorrhage triage received FDA clearance, a step forward for identifying the problem. The agency is now funding the interventional side: robots that can find the bleed and stop it.
The Problem that Tourniquets Cannot Reach
A tourniquet stops bleeding from a limb. That is a solved problem: the US military has fielded tens of thousands of them. But the torso is a different anatomical category. The aorta, the vena cava, the liver, the spleen — none can be compressed from the outside. A soldier with a torn liver bleeds into his own abdominal cavity, and there is nothing a medic in the field can do except watch and call for evacuation.
This is the gap the Medics Autonomously Stopping Hemorrhage (MASH) program, announced in September 2025 and now entering its active solicitation phase, is designed to close. The objective is explicit: develop autonomous robotic systems that can find and stop torso bleeding without a surgeon present, buying 48 hours or more until definitive care.
The technical challenge is what the agency calls a "DARPA-hard problem." The system must work through the terrain of organs and blood inside the torso using AI-driven sensor fusion, locate the exact bleeding point, and deploy an endovascular or minimally invasive tool to stop it, all with a medic providing physical assistance but no surgical decision-making.
"Imagine developing a GPS for the inside of the human body," MASH Program Manager Dr. Adam Willis said. The program is structured as a 36-month, two-phase effort: the first 24 months focus on integrating sensors with robotic platforms and building the autonomy software; the final 12 months aim for a live demonstration on perfused cadavers or animal models.
Small Robots, Swarm Intelligence, and a Smart Tourniquet
While MASH targets internal bleeding through a single robotic platform inserted via a trocar or vascular access point, a second DARPA program is pursuing the opposite end of the scale. The Medical Swarm Robotics for Extraction and Life-Saving Interventions SBIR topic (DPA26TZ01-NV001), published April 13, 2026 with a June 3 deadline, calls for small, self-deploying swarm robots that can reach a casualty in inaccessible terrain, assess wounds, and perform life-saving interventions.
The solicitation defines four required capabilities, of which a proposal must satisfy at least two, one from each category:
Extraction:
1. Movement of a casualty 10 meters onto a litter
2. Stabilization of a fractured limb through entanglement of rigid structures
Treatment:
1. Massive extremity or junctional hemorrhage control
2. Medication delivery through intramuscular injection or intraosseous needle placement
The "smart tourniquet" specification stands out. The SBIR envisions multiple robots self-arranging and reassembling into a shape that clamps around an injured limb with sufficient pressure to stop arterial blood flow, essentially a tourniquet that builds itself around the wound. The same swarm logic applies to splints: robots interlock to form rigid structures around a broken limb, protecting it during movement.
The Phase I budget for these efforts is $300,000 per award, with Direct-to-Phase-II slots available at up to $1.5 million, bypassing the standard 6-12 month feasibility gate. The SBIR is managed through the Biological Technologies Office, which has been consolidating its investment around a "Live Chain" framework, the sequence of interventions between the point of injury and arrival at surgical care.
NATO's Vigorous Warrior: The Autonomous Casualty Care Chain Goes Live
DARPA is not the only organization testing autonomous battlefield medicine. On June 12, 2026, the iMEDCAP consortium demonstrated a fully integrated autonomous casualty evacuation system at NATO's Vigorous Warrior 2026 exercise in Estonia, led by the Technical University of Munich with 24 partners from nine nations.
The demonstration showed a four-phase autonomous casualty care chain:
Phase 1 — Search and locate: A thermal-imaging reconnaissance drone autonomously finds the casualty. A second drone with millimeter-accurate radar confirms respiration.
Phase 2 — Ground extraction: A THeMIS unmanned ground vehicle, fitted with the iMEDCAP autonomy kit (dual LiDAR, IMU, satellite navigation), navigates to the casualty. The system dynamically re-plans routes when GNSS signals are jammed, a realistic condition in electronic-warfare-contested environments.
Phase 3 — In-transit treatment: The patient transport box contains a sensor suite (3D body scanner, CBRN sensors, diagnostic camera, blood detector, eye movement sensor) and an autonomous robotic arm capable of administering autoinjectors, relieving tension pneumothorax, and applying a smart tourniquet. All diagnostic data is relayed to a remote physician who authorizes invasive interventions.
Phase 4 — Aerial evacuation: The transport box is transferred to Avilus's Grille UAV, an electrically powered aircraft with 175 kg payload and 50 km range, deployable by two personnel in 15 minutes, for evacuation to a medical facility.
The Avilus Grille X4 completed its maiden flight on June 3, 2026, moving from prototype to pre-series production. The company has also partnered with South Korea's UI Helicopter to expand into Asian markets, and joined the RASCAP consortium (funded by the European Defence Agency) to lead unmanned flight testing for medical rescue missions.
What Changes When the Medic Is Not Making Decisions
MASH does the most controversial thing in military medicine: it removes the human from the clinical decision loop. The program's specification states that "medical decision making, including localizing the bleed, choosing treatment, and deciding whether to treat, must be managed by the system's autonomy." The medic can position hardware, change end effectors, and provide actuation force, but cannot make surgical judgments.
This creates a legal and ethical gap that existing frameworks were not designed to cover. Military medical ethics and the laws of armed conflict impose specific obligations around care of the wounded. When an autonomous system makes a clinical error, misclassifies a wound, applies treatment to the wrong site, or fails to recognize a contraindication, there is no settled answer for who bears responsibility. The manufacturer? The program manager? The medic who inserted the port?
The same question applies to the swarm robots: a misidentified junctional wound that receives a smart tourniquet when it should not is not a software bug, it is a clinical outcome with no human in the loop. It has not publicly addressed how this accountability gap will be resolved.
What Phase 1 Tells Us About the Timeline
Probability: 65% — MASH Phase 1 and the Medical Swarm Robotics SBIR are both structured as direct-to-phase efforts with aggressive timelines. Its track record on medical programs (the FSHARP artificial blood program, the GOLDEVAC prolonged field care system) suggests technical milestones will be met. The primary risk is regulatory: both programs require FDA acceptance plans in Phase II, and no autonomous surgical system has ever received FDA clearance for an unsupervised clinical decision.
The two programs operate on different scales but the same logic: the current battlefield medical system was designed for small-unit counterinsurgency, not large-scale peer engagements. Peer-on-peer conflict produces casualty volumes that overwhelm the evacuation chain. Autonomous medical interventions are not a technological luxury. They are a numerical necessity.
The total addressable market for autonomous battlefield medical robotics spans both military procurement (the US Army's last-mile medevac robot program alone is a multi-hundred-million-dollar pipeline) and civilian disaster response. The solicitation explicitly notes dual-use applications: "Collapsed buildings, fires, and hazardous chemicals can make reaching civilian casualties impossible outside of the means of robotic and autonomous systems."
Why This Is Not a Distant Concept
Robotic surgery is already routine in controlled hospital environments. The da Vinci system has performed over 10 million procedures worldwide. Swarm robotics has been demonstrated in warehouse logistics and precision agriculture. The novelty of the approach is the intersection: autonomous clinical decision-making in unstructured, contested environments, with no human supervisor available to override a bad call.
The 24-month timeline to functional prototypes means the first field demonstrations could occur by mid-2028. That is two years, within a single US presidential term and a single Pentagon budget cycle. The companies that win these SBIR awards will define the technical baseline for an entirely new category: autonomous trauma care.
The question is not whether the technology can work. Ultrasound-guided robotic needle placement already exceeds human accuracy in lab settings. Computer vision systems trained on tens of thousands of trauma CTs can identify internal bleeding sites faster than a radiologist. The question, as the solicitation implies, is whether the systems can be trusted to do it alone.
→ MASH and Medical Swarm SBIR award announcements (Phase I awards expected Q4 2026 – Q1 2027)
→ Avilus Grille X4 production milestones and iMEDCAP Phase 2 demonstrations
→ US Army last-mile medevac robot program integration with DARPA swarm platforms
→ FDA regulatory pathway decisions for autonomous surgical systems: any precedent-setting guidance will shape the entire category
→ Foreign equivalents: EU's iMEDCAP (€45M EDF funding), UK's Defence Innovation conflict wound care call (£1.5M), Japan's MoD autonomous casualty evacuation research program
DARPA is building robot medics. The question is not whether they work, but whether the military is ready to let them decide who lives.
