Framework of Reprogrammable Agents for Changing EW Threats (REACT)

ABSTRACT: Electronic Warfare (EW) continues to evolve as a critical operational threat, leading to information loss and corruption. EW adversaries rapidly change their behaviors, which are responsive to the electronic protections (EPs) used by defenders. Furthermore, adversaries employ wartime reserve modes (WARMs) that adapt behaviors to differ extensively from previously established approaches. Current adversary models used for training are not dynamic or reactive enough to provide realistic electronic attacks (EAs) and become obsolete overnight as WARMs surface. We propose to design and demonstrate the feasibility of a Framework of Reprogrammable Agents for Changing EW Threats (REACT), supporting the rapid adaptation, construction, validation, and deployment of adversary EW behavior models. REACT will focus on three key components. First, the Adversary Model Suite will provide dynamic and adaptable models that capture the behaviors describing how adversaries use EAs, incorporating details underlying that behavior (e.g., physics and EP interactions) as needed. Second, the Model Reprogramming Toolkit provides tools to domain experts with novice-level modeling experience to construct behavior models of emerging WARMs. These include user-focused model editing tools and automated guidance tools to drive editing. Third, the Agent-based Deployment System integrates adversary models into high-fidelity simulation environments to provide robust EW training. BENEFIT: We expect the full-scope REACT framework, and particularly the Model Reprogramming Toolkit, to have immediate and tangible benefits to military defenders in electronic warfare (EW) environments, providing dynamic and reprogrammable EW adversary models that produce more realistic behaviors for training purposes. While REACT is designed to support adversary construction for EW training, the underlying model reprogramming technology can be generalized to provide dynamic and reprogrammable adversaries for a wide range of training settings, including (but not limited to): cyber warfare, air-to-air combat, anti-submarine warfare, and counterinsurgency. Commercially, we will explore licensing REACT to generate robust and reprogrammable adversaries for the gaming industry, and we will incorporate REACT capabilities into our graphical agent development environment, AgentWorks, providing a model management system, more intuitive modeling tools, and a range of guidance tools for model editing.