This thesis develops a framework for simultaneous decentralized estimation and control, where each agent exchanges information with its neighbors to improve the estimate quality and influence the local control decisions. In this thesis, worst-case estimators are the primary focus. With motion capabilities, each agent is able to move to improve estimates of the environment. The control law design is based on a centralized scheme, in which each agent has perfect global knowledge, trying to satisfy a group objective. Average consensus estimators, instrumental in various decentralized estimation and control implementations, allow the agents to accurately estimate the group information needed for this controller. Stability and convergence conditions for several average consensus protocols are presented. This framework is applied to the problem of cooperative target localization. We validate our approach through computer simulations. A thorough analysis of these different decentralized estimation techniques looks at the effects of average consensus estimators on the systems performance and characterizes the performance differences under various target models.
展开▼
