Koopman Operator-Based Rapid Adjudication of Federated COA's (KOBRA)

This Phase II effort advances a composable, federated course-of-action (COA) planning and adjudication system built on Koopman operator theory. We will develop algorithms for ROM composition, uncertainty quantification, and robustness testing under noise, partial observability, and adversarial perturbations. Spectral analysis of the composed operator yields fast, accurate COA adjudication, eigenvalues characterize multi-domain stability, and eigenfunctions produce spatiotemporal mode evolution. For testbed, we integrate kinetic and non-kinetic simulations, AFSIM for kinetic engagements, an agent-based Medical Treatment Facility (MTF) model for casualty outcomes, and an urban insurgency agent-based model, into a unified data pipeline. Each subsystem is approximated via surrogate Koopman models, producing linear ROMs in a lifted observable space. We then “stitch” these into a block-structured federated Koopman operator, whose diagonal blocks capture subsystem dynamics and off-diagonals encode interdependencies.