SBIR-STTR Award

RT&L Focus Area: Artificial Intelligence & Machine Learning - First introduced in 1972, the Fairchild A-10 Thunderbolt II has seen nearly five decades of success. To keep the platform mission ready, the 309th Software Engineering Group (309SWEG), designs, builds, tests, and deploys software upgrades to its multi-function display and pilot vehicle interface (MFD/PVI). The potential for error in their work and the harm it could inflict translates to lengthy test and evaluation (T&E) processes. Those processes add years to software rollout, stagnating both A-10 and more broadly Air Force modernization efforts while our adversaries march forward. Critically, an update to a single line of code for the MFD/PVI can take up-to 2 years to push into operations due to T&E alone. With the 309SWEG, CrowdAI seeks to reduce the time to complete visual inspection of MFD/PVI using computer vision, a form of artificial intelligence, to evaluate that interface during T&E. CrowdAI can address this critical challenge to aircraft modernization by leveraging artificial intelligence (AI). Using computer vision, a subfield of AI, CrowdAI software will evaluate the MFD/PVI, just as a pilot would see them, allowing 309SWEG System Test Engineers to reconcile the flight computer commands to the MFD/PVI with what is visually displayed. But, unlike a human, who tires under cognitive load and prolonged tasks, AI can process video longer and even faster than it can be collected. AI reduces the labor required for manual review of video, freeing personnel to higher-cognitive tasks. Long term, similar capabilities could be applied to other combat aircraft, not just the A-10, as well integrate at all different stages of testing, not just development, but acceptance, and operational. Under Phase I award, CrowdAI will provide a product demonstration to the 309SWEG, Air Force Test Center (AFTC) and other stakeholders for a potential Phase II. CrowdAI will work with 309SWEG to identify customers and stakeholders key to transitioning the solution into the enterprise, presumably within Platform One. CrowdAI will evaluate anticipated project data for suitability with an AI solution that would process MFD/PVI data. To evaluate this data, Air Force Research Lab has agreed to host it in its Global Unified Environment, a CUI-accredited cloud compute environment, where CrowdAI software is already deployed. CrowdAI machine learning models solve specific use-cases, exploiting virtually any form of imagery collected by ground, aerial, and satellite platforms—including full motion video and still imagery sensors. These models can evaluate greater quantities of imagery than an analyst and in a fraction of the time—and AI never tires, reducing the probability of error. For this SBIR Phase I, CrowdAI will work with 309SWEG to assess mission data and requirements in preparations for curating training sets and to train CV models in a Phase II.

As adversaries to the United States improve their combat aircraft and air defense systems, the Department of the Air Force (DAF) must continuously improve its fleet. An aircraft, when it enters into service, can operate for decades before being decommissioned. During this lifespan, the avionics, flight computers, and other systems receive upgrades to keep aircraft flight ready and mission capable. Software changes, even innocuous updates, take on new import in military systems, where the implications of poorly executed code can have catastrophic results for the pilot, crew, and mission. To mitigate these risks, the 309th Software Engineering Group (309th SWEG), which is responsible for developing “cradle-to-grave” software and hardware support systems, conducts thorough testing before pushing new software to operating aircraft. This process includes recording and reviewing hundreds of hours of video from inside a cockpit testbed to observe the software in operation in a relevant environment. This entire process can add months and sometimes years to mission critical upgrades for aircraft, such as the A-10, F-16, F-22, F-35. Working with 309th SWEG, CrowdAI proposes to leverage human-machine teaming to reduce not only the analytic burden on Airmen to perform this task, but to reduce the time to completion while delivering as good or better results. Computer vision (CV) is a field of artificial intelligence (AI) that enables computers to derive meaningful information from digital scenes, such as video— and to take actions or make recommendations based on that information. The proposed project would implement a CV solution to assist Airmen reviewing video of the cockpit testbed during software testing. CrowdAI is an industry leader in an advanced CV technique called image segmentation, which analyzes video at the pixel level. This capability is critical given the complexity of the “heads-up display”, which conveys many pieces of information, projected onto a dynamic background. CrowdAI was the first company to apply image segmentation to video, earning it the top place in the DAF’s HyperSpace Challenge in 2018 as well as selection by the DOD’s Joint Artificial Intelligence Center as its first CV vendor. For this proposed work, CrowdAI would develop new CV models for use by the 309th SWEG. The project would include gathering relevant video data, labeling relevant objects in that data for the model to learn, and training multiple computer vision models. These models will then be transferred from the development environment to an operations environment not only to explore cross-domain approaches and technical requirements, but to evaluate model performance on production data. AFRL will provide compute resources during this phase. 309th SWEG is the project end-user and transition partner for integrating into the Advanced Battle Management System (Party Bus) for anticipated long-term sustainment and which CrowdAI is already on contract (FA8612-21-D-0117).