Navigation of Autonomous Cooperative Vehicles for Inference and Interactive Sensing

This paper addresses the problem of autonomously choosing navigation actions while searching for targets in littoral regions. The search amounts to performing detection and classification of measurements as they are collected by a moving sensor. Spatial grids are used to capture the detection and classification states, respectively, of the littoral region, and are used to indicate if objects exist and if so, indicate their class. The navigation actions are chosen by maximizing a recently proposed information-theoretic cost function that incorporates knowledge of the current detection and classification states captured in the spatial grids. Our prior work proposed a framework for performing this type of search with a single vehicle. Here, we propose a modified cost function appropriate for multiple cooperative vehicles capable of sharing information, and characterize the consequences as more vehicles are incorporated into the search. We examine the diminishing returns in performance as we increase the number of vehicles searching a fixed-size area.