On Distributed Processing for Underwater Cooperative Localization

Due to the limited bandwidth of underwater communication links, underwater cooperative localization usually adopts a distributed processing architecture. Members of the team estimate positions using their local sensor data, and fuse the information communicated by other members for cooperation. It is common practice to naively assume independency during information fusion between cooperative members. The assumption is not always valid. This results in overconfidence in estimation as a result of the double-counting of the common information. While this problem is recognized by many researchers, there has been no explicit study on the dangers of naive filtering in presence of inter-dependency. In this paper, we derive an optimal fusion for distributed cooperative localization in a multi-sensor tracking application, and evaluate its gap with respect to the central filtering. For the naive filtering, we examine one-step and asymptotic performance and demonstrate the existence of safe and dangerous regions of operation.