Autonomous Distributed Systems

We will develop an autonomous distributed Bayesian information fusion architecture for surveillance and situation monitoring based on data acquisition from sensor networks. This architecture employs a powerful new Bayesian lambda-calculus modeling languagefor probabilistic representation and inference developed at the University of New Mexico. The representation provided by the lambda-calculus supports an extremely compact expression of probabilistic data and information, a critical feature in the limitedbandwidth environment of sensor networks. In addition, the accompanying algorithm for data fusion exploits locality to reduce computational and communication costs. This fusion architecture operates at the level of information and knowledge processingrather than pure signal processing. It uses Bayesian models to encode prior knowledge about situations, objects, and events, and their manifestations in observable data. The models support context-sensitive data interpretation, emulating the skill andbehavior of trained observers who know, in a general sense, what to look for, what to attend to in particular types of environments. In addition this prior knowledge supports dynamic reconfiguration of sensor hardware and signal processing algorithms,thereby optimizing the value and reliability of information provided by available sensor resources. These capabilities in distributed representation, data interpretation, and decision support can potentially generate important advances in AutonomousCommand, Communication, and Control infrastructures, with significant benefits for intelligence and surveillance systems such as DADS and MicroDADS. In addition, this technology will provide commercial opportunities in autonomous oceanographic monitoringand management of commercially important fishing resources. We also foresee wide-reaching applications in autonomous surveillance of process plants, urban traffic control, environmental monitoring, and autonomous space exploration.