U.S. Special Operations Command has posted a competitive assessment seeking a self-service synthetic data generation platform to train the computer vision models powering its unmanned systems, according to a DefenseScoop report published June 9, 2026. The effort sits under the Unmanned Systems Autonomy and Interoperability program — known as UxSAI — which is managed by the Program Executive Office for SOF Digital Applications. Whitepapers from interested vendors are due June 29, and SOCOM expects to select a vendor and open contract negotiations in October.
What UxSAI Is and Why It Needs Synthetic Data
UxSAI is SOCOM's program to build autonomous capability and interoperability across its portfolio of unmanned systems — covering air, ground, and maritime platforms. The program's computer vision models allow those platforms to detect, classify, and track objects in the field without continuous human input. That capability depends entirely on the quality and volume of imagery used to train the models.
SOCOM has identified a clear problem with relying on real-world imagery alone. As the command stated in its solicitation materials, "Training high-performance CV models requires large volumes of labeled, operationally relevant [EO/IR] imagery" — and the constraints of real-world collection make synthetic generation necessary. Collecting actual electro-optical and infrared imagery at the scale, diversity, and operational specificity required for high-performance model training is logistically difficult, operationally sensitive, and time-consuming. Field collection in denied or contested environments is frequently impossible. Labeling that imagery — annotating every frame so that a model can learn what it is seeing — adds further time and cost.
Synthetic data generation addresses those constraints by producing artificial imagery that mimics real-world conditions. Rather than collecting footage of an actual vehicle in an actual desert at night, a synthetic platform renders a photorealistic simulation of that scene and generates labeled outputs automatically. The resulting datasets can be produced at scale, varied across terrain types, lighting conditions, weather, and sensor perspectives, and tailored to the specific operational environments SOCOM anticipates. When done well, models trained on high-fidelity synthetic EO/IR data perform comparably to those trained on real collected imagery — without the collection burden.
What the Platform Must Do
SOCOM's requirement is specifically for a self-service platform — one that UxSAI program personnel can operate without ongoing support from the vendor. That constraint reflects a deliberate policy choice: the command wants to generate synthetic datasets on its own schedule, modify parameters to match new mission requirements, and iterate on its training pipelines without waiting for vendor assistance or paying for professional services engagements every time conditions change.
The platform must produce labeled EO/IR imagery — meaning it outputs not just rendered scenes but also the associated annotation data (bounding boxes, segmentation masks, object class labels, and similar metadata) that machine learning engineers need to feed model training pipelines. The imagery must be operationally relevant to the specific unmanned platforms and environments UxSAI supports, spanning air, ground, and maritime domains. That breadth means the platform needs to handle a wide range of sensor types, platform perspectives, and scene compositions.
No ceiling dollar value for the resulting contract has been disclosed. The competitive assessment is being managed through SOFWERX, the innovation hub that serves as SOCOM's primary mechanism for rapid technology engagement with industry.
Assessment Timeline
The process runs across four stages. Whitepapers are due June 29, 2026 — vendors submit written descriptions of their platforms and relevant technical approach. A demonstration event is scheduled for mid-July, at which shortlisted vendors will show their platforms in action. SOCOM will then conduct a downselect in September to identify the vendor or vendors advancing to test and evaluation. Vendor selection and the start of contract negotiation are expected in October 2026.
The assessment structure is typical of SOFWERX-managed technology evaluations: it is faster than a traditional FAR Part 15 competitive acquisition and is designed to let the government evaluate working systems rather than written proposals alone. SOFWERX assessments are conducted under Other Transaction Authority or related streamlined mechanisms, which allows SOCOM to move from whitepaper to contract in months rather than years.
What It Means for Contractors
This opportunity is aimed squarely at companies operating in the synthetic data, simulation, and AI training infrastructure space — particularly those with experience generating EO/IR imagery for defense applications. That includes established defense AI vendors as well as commercial synthetic data companies that have begun building government-sector offerings. Small businesses with a track record in computer vision tooling or simulation environments should also review the SOFWERX posting, as SOCOM has historically used these assessments to engage non-traditional vendors.
The self-service requirement is the most important technical filter. Vendors whose platforms require professional services engagements for configuration, dataset generation, or pipeline changes are unlikely to satisfy the stated requirement regardless of how capable the underlying technology is. Whitepapers should address the self-service architecture directly — explaining how UxSAI personnel would operate the platform, what training or onboarding is required, and how the system handles new scenario configurations without vendor involvement.
Whitepapers should also address EO/IR fidelity specifically. SOCOM is not looking for generic imagery generation — it needs outputs that are operationally realistic for the infrared and electro-optical sensors mounted on its unmanned systems. Vendors should describe how their platform models sensor characteristics, atmospheric conditions, and target signatures in the relevant wavebands, and how labeled outputs are formatted for integration into standard computer vision training pipelines.
For firms unfamiliar with the SOFWERX pathway, the process works as follows: SOFWERX serves as the contracting and coordination interface between industry and SOCOM program offices. Once a vendor is selected through the assessment, the actual contract is negotiated directly with the program office — in this case, PEO SOF Digital Applications. The whitepaper stage is non-binding, but vendors selected for the July demonstration will be expected to show a working platform, not a prototype or concept. Companies should register on the SOFWERX event page and submit whitepapers through that portal before the June 29 deadline.
