Neuro-dynamic programming for adaptive control of Bayesian networks for global awareness

In order to be efficient and robust, future systems for global awareness applications must accommodate new data sources and alter their computational structure automatically. We demonstrate the utility of neuro-dynamic programming (NDP) for addressing such challenges in a real-time application. We describe our implementation of a simple information fusion system that uses NDP to allocate resources to Bayesian networks. The Bayesian networks use information from distributed sensors and databases to infer identity of vehicle clusters.