SABM Agent to Maintain Situational Awareness (SAMSA)

Watch standers on AEGIS-equipped ships need air, surface, subsurface, and missile defense related Situational Awareness (SA), including an operational, environmental, and tactical Distributed COP (DCOP). The Navy would like an autonomous cooperative SA agent whose capabilities include detection, tracking, and ID of complex air, surface, and subsurface objects (including blue, red, white, and gray); engagement-quality tracks; sensor fusion; behavior analysis and prediction; dynamic information/threat prioritization (based on AOR, ship objectives, and/or overall mission objectives); recommendations to watch standers (e.g., how best to maintain SA and ship safety); user-customizability; and scalability (information overload is an enormous problem which complicates all the previous issues). Stottler Henke proposes SABM Agent to Maintain Situational Awareness (SAMSA). In Phase I, we will develop a SAMSA prototype. SAMSA utilizes several Artificial Intelligence (AI) techniques including Knowledge Engineering (KE), Fuzzy Logic (FL), Behavior Transition Networks (BTNs), Case-Based Reasoning (CBR), Deep Learning (DL), Convolutional Neural Networks (CNNs), Computer Vision (CV) techniques, Part-Based Models, and Template Matching. Knowledge Engineering refers to the process of extracting knowledge from subject matter experts, developing a Knowledge Representation (KR), and building a system based on formal and informal facts, rules, and frames. Scott Higgins, Robert Watkins, Anthony Gaskins, and Dan Shaffer, who collectively have tremendous expertise in surface, air, & subsurface warfare, ballistic missile defense, weapons systems, and more, at an own ship level and at a group/staff level, will provide subject matter expertise. These experts have served on various ships in several positions including Tactical Action Officer (TAO), Anti-Air Warfare Coordinator (AAWC), Anti-Submarine Warfare Coordinator (ASWC), and Commanding Officer (CO). Fuzzy Logic captures rules that reference qualitative, inexact, or “fuzzy” values such High/Medium/Low. Behavior Transition Networks create intelligent behaviors by dividing behavior hierarchically into tasks connected by transitions. Case-Based Reasoning (CBR) attempts to solve the current problem by retrieving a previously encountered similar problem and adapting that problem’s solution to the current situation. Deep Learning refers to Artificial Neural Networks (ANNs) with many hidden layers. Specifically, deep Convolution Neural Networks (CNNs), a type of ANN, are used extensively for Computer Vision applications. Part-based models decompose an object into a set of parts, modeling the local appearance of each part individually and then constraining the relative spatial configuration of the parts. Template matching uses robust features (e.g., histograms of oriented gradient (HOG)) to match new images against characteristic templates.