SOCOM seeks ‘self-service’ synthetic data generation platform to boost drones’ computer vision
U.S. Special Operations Command and the SOFWERX innovation hub are launching a new initiative to support SOCOM’s pursuit of new capabilities to train computer vision models for unmanned systems.
American special operations forces use various types of drones in the air, maritime and ground domains to support intelligence, surveillance, reconnaissance and other missions. Computer vision can boost those efforts by using AI to detect, identify, classify, and track objects and people of interest. However, there are challenges associated with the technology.
“Training high-performance CV models requires large volumes of labeled, operationally relevant [elctro-optical/infrared] imagery. Manual data collection is constrained by cost, access, and the inability to capture rare events, contested environments, or classified threat signatures. Synthetic data generation addresses this gap by producing labeled datasets that simulate operational variability without real-world acquisition constraints,” officials wrote in a special notice posted June 2 on a government contracting website.
SOCOM’s program executive office for SOF digital applications wants that type of tech for the Unmanned Systems Autonomy and Interoperability (UxSAI) program. More specifically, the organization is looking to acquire a new synthetic data generation platform that can produce EO/IR training imagery for algorithm development within the program’s existing machine learning operations (MLOps) pipeline. Notably, SOCOM wants a “self-service” platform that its personnel can operate “organically” without vendor assistance.
Officials are planning to host a series of assessments and downselects in the coming months to identify and examine potential solutions from industry, as SOCOM looks to enter into contracts or other agreements with vendors.
The deadline for providers to submit their whitepapers for initial review by the command is June 29.
Following an initial downselect, an assessment event is slated for mid-July where proposers will be expected to demonstrate their capabilities.
In September, after another downselect for the test-and-evaluation phase, program officials will “run models and associated data through the process of T&E in their associated pipeline,” according to the notice.
In October, SOCOM plans to pick one vendor and negotiate an award to develop and deliver a synthetic data generation tool for the UxSAI program, per the notice.
