ARETE ASSOCIATES — Department of Defense STTR Phase I: N23A-T013

Description

Areté and its teaming partner the University of Arizona (UofA) will develop a software tool that transforms sensor and metadata from a given sensor system into realistic synthetic data as if it were collected by a different sensor system. The exponential rise in available data from a multitude of sensor systems has driven commercial and academic entities to achieve significant innovations in artificial intelligence (AI) and machine learning (ML) methods which are a cornerstone to modern ATR techniques. These innovations represent an opportunity for US Navy (USN) organizations to capitalize upon and advance in analogous albeit military-centric applications. However, the prevailing roadblock for the USN to use these innovations or develop their own is the lack of sufficient, operationally relevant truthed data to train and innovate advanced ML architectures for the USN community. Unmanned Underwater Vehicle (UUV) systems are subject to an even more challenging situation, as collecting a dataset of objects underwater is a cumbersome task. In underwater environments, the types and locations of interesting objects are typically unknown or represent rare events. Alternative data collection methods, such as manually placing objects on the seafloor and capturing images with sensors mounted on an autonomous UUV are time and cost-intensive, especially if a high variability in the dataset is desired. This may even lead to longer resource commitments, test runs, and mission times to ensure each sensor, and associated algorithms, have an opportunity to collect target information in a variety of environmental conditions. Areté's aim is to develop the Navy UUV Sensor Transformation Tool (STT) technology to provide the USN with a cost-effective, easy-use tool for generating realistic synthetic data. The STT application will make use of both existing real sensor data collections, which can be transformed into needed sensor data types or by creating synthetic scenarios, which can be transformed into needed sensor data types. This data creation capability will remove the bottleneck of a dearth of training data necessary for developing, training, testing, and validating advanced ATR algorithms for UUV applications.