Bayesian variable resolution modeling for autonomous satellite cluster data fusion

We report on work on a new approach to counterspace situation awareness, based on variable resolution modeling using distributed relational Bayesian networks. Variable-resolution modeling provides the proven effectiveness of Bayesian data fusion in an efficient, scalable, distributed architecture. The core representational element is the Bayesian network (BN), a graphical representation for probabilistic modeling of large, highly structured problems. A recent breakthrough, relational Bayesian networks, rests on an underlying entity-relationship data model, providing a principled foundation for constructing probabilistic situation awareness. We are developing and evaluating a distributed architecture for scalable, distributed data fusion for counterspace situation awareness. The capabilities include: (1) detection, association, tracking, and assessment of entities, relationships, and overall situations; (2) multiple-hypothesis relationship and situation modeling; (3) distributed situation modeling, robust to single-point of failure in both processing nodes and communication links; and (4) automated management of node processing and storage utilization, as well as link bandwidth utilization.