AI/ML Multi-Sensor/Multi-Target Localization

For Phase I of this effort, Crewless Marine will develop, test, and evaluate a multi-target, multi-sensor acoustic localization algorithm that combines Artificial Intelligence/Machine Learning with recent advances in Bayesian estimation techniques. The algorithm is an Artificial Acoustic Intelligence Agent (A2I Agent) that changes the current paradigm by fusing data from all acoustic sensors, performing measurement-to-track association, and autonomously ìdrilling downî back to each sensor. This approach will dramatically reduce operator workload while providing an ìexplainable AI (XAI)î displaying both a holistic tactical contact picture in geo-space and acoustic overlays for each track on sonar data identified by track. The algorithm supports all submarine acoustic sensors and phenomenology (e.g., broadband, narrowband, cross-correlation, DEMON) and is extensible to off-platform sensors. The A2I has five major components:N-State track estimation with three-dimensional Cartesian position, two-dimensional Cartesian velocity, and zero to M base frequencies (which can be converted to the traditional four-state bearing, range, course, and speed),multi-sensor, multi-target measurement to track association,array shape estimation that is initialized with standard techniques, but updated based on acoustic data once the track has measurements from a hull sensor,a layered machine-learning approach that uses zero-shot learning for new contacts and transitions to few-shot learning combined with deep learning techniques for the classification of known contact types, andan XAI engine that displays tracks in a geometric view with overlays on sonar operator displays.The algorithm will be developed using representative scenarios with a sensor suite including three hull-mounted sonars and one towed array. The algorithm will be developed, tested, and validated using three levels of data:measurement level simulation, with mid-fidelity time-series, that will simulate the outputs of the contact followers to create measurements of conical angle, bearing, bearing/elevation with frequency (representing narrow-band contact followers) and without (representing broadband contact followers),high-fidelity acoustic time-series modeling with detection algorithms representative of contact follower measurements (Phase I option), andlimited in-water data collection using Crewless Marine hardware to validate a simplified scenario with real-world data using contacts of opportunity. (Phase I option)Crewless Marine Acoustics is proposing to partner with two other small companies as sub-contractors, Aureum.ai and Applied Ocean Sciences on this effort.