| This dissertation is a culmination of research in several parallel, yet related avenues dealing with statistical estimation in the context of dynamic state tracking. The main contribution of the thesis is to provide a novel scalable framework for tracking of a high-dimensional target state using a distributed and cooperative network of agents. This framework is incorporated into a specific application consisting of a multiple target tracking (MTT) environment, with each agent employing a number of mobile received-signal-strength (RSS) sensors capable of collecting measurements revealing localized information regarding the target state environment. The dimensionality of this problem, which is the dominating factor determining feasibility of any proposed solution, is effectively managed by dynamically assigning each agent a partition of the full target space to track. Agents are thus able to individually focus on a small piece of the full estimation problem and rely on inter-agent communication to compensate for state subspaces outside their estimation scope which may adversely influence their own measurements. The framework mentioned is coined Multi-Agent Systems for Cooperative Tracking or MASCOT.;Specific implementation details and associated challenges relating to this framework and the specific application considered are presented. Namely, inter-agent communication and the required information fusion necessary for RSS sensors is extensively investigated. Additionally, optimal RSS sensor placement within the described multi-target environment is addressed, greatly facilitating enhanced performance of the main subspace partitioning algorithm. The MASCOT framework has been fully implemented in the MATLAB modeling language; computer simulation results are presented, demonstrating algorithm performance and its superiority to earlier methods of handling such a problem. Finally, specific analysis of performance measures dealing with general statistical estimation (Bayesian estimation particularly) are presented as well as some novel results regarding a connection linking Frequentist/Bayesian estimation paradigms. |
