Distributed networked sensing and control systems: Robust estimation and real-time control

There is a growing interest in distributed networked sensing and control systems, such as wireless sensor networks, networked control systems, distributed control systems, multiagent systems, and heterogeneous sensor networks. A distributed networked sensing and control system consists of a number of autonomous agents. Information among these agents is usually shared by wireless communication. However, each agent is resource-constrained, e.g., it may have limited processing power, storage capacity, and communication bandwidth. These constraints create measurement inconsistency and communication unreliability and they are the major obstacles in realizing an autonomous distributed networked sensing and control system which is capable of real-time situation understanding and control.; In this thesis, we take an advantage of spatio-temporal correlation among the neighboring agents and develop robust real-time algorithms for situation understanding and control. We also design and implement a real-time situation understanding and control system using wireless sensor networks.; For this purpose, we use the mathematical frameworks of multi-target tracking and pursuit evasion games. Multi-target tracking is a general framework which can be used to describe many estimation and inference problems appearing in distributed networked sensing and control systems. The pursuit evasion game is a mathematical framework for many challenging control problems and it can be viewed as the worst-case control problem.; The thesis starts with the simplest distributed estimation problem in a sensor network. After showing that the method cannot be applied to more general multi-target tracking problems, we develop a general Bayesian framework for multi-target tracking problems. The Bayesian framework allows a method which is robust against inconsistency in measurements and missing measurements due to communication unreliability. Since the exact computation of Bayesian estimates is a time-consuming task, we develop an approximate method, called Markov chain Monte Carlo data association, to efficiently solve the data association problems appearing in multi-target tracking problems.; Markov chain Monte Carlo data association is also used to improve the robustness of the multi-target tracking methodology by compactly managing identities of multiple objects using the identity-mass-flow framework. We then develop a real-time hierarchical control system with multiple layers of data fusion to solve the multi-target tracking and pursuit evasion games using a distributed networked sensing and control system. This thesis presents the first demonstration of multi-target tracking using a wireless sensor network without relying on classification.; We also present a general framework for modeling a distributed networked control system consisting of multiple agents communicating over a lossy communication channel. We describe exact and approximate filtering methods to estimate states of a distributed networked control system. In addition, we describe how to find a communication control which stabilizes a distributed networked control system.