DARPA Robotics Programs Currently in Active Funding

DARPA’s active robotics funding in 2026 is best understood through the failures it targets. Across six distinct programs and four offices, the portfolio addresses gaps between autonomous capability and dependable operation. For researchers and organizations, the practical entry point is to match a capability to one of those gaps, then identify the funding mechanism that fits.

What DARPA's active robotics portfolio reveals about where autonomous systems are failing

A robot trained in simulation faces a different problem from a drone that loses its operator link or a medical system operating without a surgeon. These distinct autonomy problems require different research methods, tests, and routes into use.

DARPA’s office structure reflects those distinctions. Biological Technologies (BTO), Tactical Technology (TTO), Microsystems Technology (MTO), and Defense Sciences divide the work across medical, platform, materials, and foundational questions. DARPA’s FY2026 budget provides the fiscal basis for this multi-front effort. The portfolio’s coherence comes from the operational failures it targets.

The sim-to-real gap TIAMAT is designed to close

TIAMAT addresses a foundational limitation of robot development. Increasingly sophisticated simulations have not reliably produced autonomy that transfers to field conditions. Rather than concentrating training in narrower, more detailed simulations, the program emphasizes breadth across lower-fidelity simulations. TIAMAT investigates whether exposure to that broader range can produce more robust transfer than closer reproduction of a limited environment.

The program’s phases make that question testable. Phase I worked on sim-to-sim transfer and ran through early 2026, while Phase II began in March 2026 and moved toward building hardware and testing it in the field. A final competition is scheduled for July 2027. That progression moves the assessment from performance across models to performance outside them.

Among TIAMAT’s 13 performers are teams led by the University of Central Florida and Johns Hopkins. Each performer has a different university or industry lead, enabling a parallel search. By letting both universities and industry take part, DARPA can try different ways to solve the same transfer problem and test whether they work with hardware and in the field.

Why REMA treats link failure as a design constraint rather than an edge case

REMA addresses a later point in the robot’s lifecycle. Even a drone with useful autonomous capabilities is limited in what it can do if it needs a constant link to its operator. That dependence also reinforces a personnel bottleneck. When operation depends on sustained human attention, the operator-to-drone ratio limits what a force can do with its available systems.

REMA therefore treats loss of communication as a design condition. Its proposed architecture combines a hardware adapter that detects drone type with an autonomy software layer that completes the mission when the link drops. Modularity is the central technical choice. The program separates the drone-autonomy adapter interface from mission-specific autonomy software, making them its two technical areas.

The schedule reflects the urgency of that problem. REMA uses an 18-month, single-phase structure, with development cycles compressing from three-month to one-month intervals as the program matures. TTO manages the work. Unlike TIAMAT’s focus on transferring trained behavior into the world, REMA targets the dependence that remains when a deployed platform cannot act without its communications link.

What SAFRON addresses when an LLM-commanded robot misunderstands a field instruction

SAFRON tackles how robots turn their understanding of language into physical action. Foundation models can accept natural language commands, but that ability does not mean they fail safely when instructions are ambiguous or incomplete. Physical execution changes the stakes of misunderstanding. An instruction that works in a controlled laboratory can become dangerous in an unpredictable environment on which the model was not trained.

The robot may receive the instruction successfully, yet still lack a safe interpretation of what it should do. As large language models become able to control robots, commands can bring risks of their own, even when the rest of the task is defined. SAFRON targets that frontier. It is listed as active, but specific awardee details are not publicly established at the time of writing, so its problem definition supports firmer conclusions than its implementation details.

How DARPA's physical-compute RFI challenges the assumption that robots think with chips

DARPA’s April 2026 physical-intelligence RFI, DARPA-SN-26-76, moves the inquiry beneath software to the architecture of computation. Central processors use power and add latency, and software optimization alone cannot remove these costs. The RFI asks whether a robot’s physical materials can encode computation, an architectural approach to autonomy.

The materials of interest show what that distinction means. Shape-memory polymers can change shape under stress, while piezoelectric materials turn pressure directly into electrical signals. Pneumatic systems have local feedback loops that work without sending information through a central node. These examples place part of the response in the physical system itself, rather than treating the body only as hardware that receives commands from a processor.

DARPA’s exclusions sharpen the research boundary further. Incremental hardware improvements, purely software-based AI approaches, and conventional CPU/GPU-centric architectures are explicitly outside the RFI’s interests. The mechanism is still informational, with no awards yet and first prototype contracts expected in early 2027, so it should not be read as an already funded development program.

The battlefield surgical capacity gap that two parallel programs are attacking from different angles

During mass-casualty events, people can die of injuries they could survive because no surgeon is available. DARPA is using both contract R&D and a prize competition because building prototype systems serves a different purpose from drawing new entrants into the market. MASH helps teams build a defined medical capability, while the DARPA Surgical Competition (DSC) tests what teams can do by having them compete.

BTO launched MASH, Medics Autonomously Stopping Hemorrhage, in September 2025 with a total budget of approximately $32.4 million. It targets the location and stabilization of severe abdominal bleeding through external, laparoscopic, endovascular, or hybrid approaches using machine autonomy. Phase 1, beginning in summer 2026, focuses on wound localization and artificial clot formation. By month 24, the system is expected to identify active bleeding autonomously. Phase 2 then optimizes the system and prepares it for potential battlefield deployment within a further 12 months, with the goal of sustaining bleeding control for 48 or more hours to allow evacuation.

The DSC offers a complementary route to progress. Its Component Gates assess Medical Knowledge, Physical Capability, and Rapid Learning before participants enter operational lanes. The Surgical Collaboration Lane evaluates safe assistance to a lead surgeon during damage-control procedures, while the Surgical Independence Lane evaluates autonomous assessment, planning, and execution without human intervention. Component Gate Prizes also reward individual capabilities. The Finals Competition Event is scheduled for March 2027.

Dr. Jeremy C. Pamplin states the intended relationship plainly. “We are not trying to replace surgeons; we are building them a highly skilled partner”. BTO hosts the competition for innovators in robotics, AI, and bioengineering, with registration and rules available through Central Florida Tech Grove and DARPA’s surgical page. DARPA says both programs can also help civilians, including by giving people in rural areas access to specialists and helping with disaster response. The capacity gap extends beyond the battlefield.

RSGS — the one program in the portfolio that has already left the ground

RSGS differs from the other programs in maturity. It progressed from concept through hardware testing to a successful launch on July 21, 2026. That makes it less an immediate funding opportunity than a demonstration of DARPA’s long-horizon investment model. Its underlying problem is the convention of treating satellites as disposable when fuel runs out or hardware fails.

DARPA built the robotic payload through the Naval Research Laboratory and Naval Center for Space Technology. SpaceLogistics, a Northrop Grumman subsidiary, integrated that suite with its Mission Robotic Vehicle and owns and operates the integrated spacecraft once it reaches GEO. Envisioned missions include orbit adjustments, inspections, mechanical corrections, and installation of additional payloads.

Launch marks a milestone while delivery of every envisioned service remains unproven. It nevertheless places robotic dexterity and autonomous proximity operations beyond the concept stage in this portfolio. For the satellite industry, the program frames maintenance and modification as service possibilities that could avoid replacing an entire spacecraft.

Earlier-stage and SBIR-track programs for organizations entering the portfolio at the concept stage

Organizations without a mature system have other entry routes. HyBRIDS, Hybridizing Biology and Robotics through Integration for Deployable Systems, began with an Advanced Research Concepts (ARC) solicitation issued in January 2025. It looks at platforms that combine cells, tissues, or organisms with synthetic systems. Its focus is on capabilities where biology can outperform fully engineered approaches, including resiliency, sensitivity, adaptability, and efficiency. This is earlier-stage work than MASH or TIAMAT.

The Medical Swarm Robotics for Extraction and Life-Saving Interventions STTR topic sought autonomous systems to help on the battlefield by moving casualties short distances and providing hemorrhage control, fracture stabilization, and medication delivery. TRIAGE-X, topic DPA26BZ06-DV028, seeks autonomous recognition of chemical toxidromes and delivery of antidotes or other life-saving interventions in mass-casualty events. Its November 25, 2026 deadline is imminent. This BTO opportunity is restricted to small businesses through SBIR.

The funding progression also differs from a major program. DARPA’s general SBIR/STTR structure provides modest Phase I seed awards and substantially larger Phase II awards, with topics released annually under DoW BAAs.

How the portfolio is structured for entry — by office, mechanism, and eligibility

BAA-track programs including TIAMAT, REMA, SAFRON, and MASH admit nonprofit organizations, universities, research institutions, and businesses of all sizes. A program manager first reviews an abstract or quad chart, then invites a full proposal. Awards typically range from hundreds of thousands to tens of millions, depending on scope. Organizations must also find a program that fits their work.

Office alignment narrows that search. BTO covers MASH, DSC, HyBRIDS, and TRIAGE-X, while TTO manages REMA and MTO houses the physical-compute inquiry. TIAMAT and SAFRON sit within the Defense Sciences or cross-office grouping. Mechanism then determines the route forward. The DSC uses open registration and component-gate qualification without requiring a prior DARPA relationship, while ARC supports exploratory concepts and SBIR/STTR restricts entry to small businesses.

Work that falls outside a named program has other routes to funding. Artificial Intelligence Exploration can evaluate proposals and issue awards in as little as 90 days for AI-robotics concepts within the I2O remit. The Young Faculty Award offers support over two years, with a possible Director’s Fellowship extension, to current tenure-track faculty regardless of years in position. It provides an academic entry point before a researcher is running a lab at scale.

The legitimate objection — funding the technology before the doctrine exists to use it

Retired Army General David Petraeus and nonprofit co-founder Isaac Flanagan question spending on autonomous systems before the military knows how to use them. They argue that much of the budget risks being wasted if investment precedes a clear understanding of how operators will integrate autonomous weapons into doctrine and logistics. Technical capability does not settle that question. A military must also know how to buy, train on, field, and maintain the systems it receives.

The objection applies unevenly across this portfolio. REMA and TIAMAT tackle limits tied to an operator-to-drone bottleneck that is already causing failures in the field, which makes a stronger case for work in the near term. The physical-compute RFI and HyBRIDS remain exploratory enough that demanding settled doctrine would be premature. DARPA’s program design offers a partial answer through transition partners, while the DSC seeks market entrants with their own commercialization paths. Neither mechanism removes the institutional risk, but both make the path from research to use a consideration in investment decisions without assuming technical success will resolve it.