Understanding a battle-space is fundamental to winning a war, and a Distributed Common Operational Picture (DCOP) is only the beginning of the required understanding. In this project we will produce a system called Artificial Intelligence Monitoring for Battle Space Understanding (AIM4BSU) to monitor the DCOP. It will provide watch standers with customized alerts and prioritized lists of targets. To do this, we allow watch standers to express their goals and priorities by selecting from a list of objectives for war games involving units in a DCOP. AIM4BSU will use reinforcement learning from adversarial self-play to train AI agents to play these games. In real time, it will monitor moves the Artificial Intelligence (AI) agent for the selected game would make in the situation represented by the real-world DCOP. Watch standers will be alerted when the agent would make a significant move and techniques from Explainable AI will determine which targets and tactical situations prompted the move.
Benefit: AIM4BSU will allow watch standers to effectively monitor greater volumes of DCOP data than is possible manually. The alerts and prioritized target lists which AIM4BSU produces will provide a contribution to situational awareness which utilizes the proven ability of AI agents to play games at a level of skill which matches of exceeds human experts. Thus, this project will improve the fleets situational awareness. Moreover, our self-play training of AI agents will provide a powerful technique for automating the identification of threatening tactical situations without the need for training data.
Keywords: Situational Awareness, Situational Awareness, AI game playing, Autonomous Monitoring
