A ROC-Based Approach for Developing Optimal Strategies in UUV Search Planning

This paper addresses the high-level search planning that is necessary to find hidden objects in difficult undersea environments where sensor performance varies geographically and the susceptibility to false alarms is high. We develop a game-theoretic approach for maximizing the information that is collected as a multiagent collaborative search is conducted over a region of interest. To accomplish this, we apply 1) a search channel formalism to the discrete search modeling paradigm to develop information measures as a function of searcher regional visitation; and 2) a receiver operator characteristic analysis to map sensor detection characteristics to search events. We formulate an area search game where player action directs where and how information is collected under a fixed cost constraint on search effort. We discuss the properties of the information measure developed for a repeated look over the area search channel and examine the impact of setting ROC operating points to their game equilibrium values. We describe a new algorithmic capability for solving the search allocation problem where searchers can collaborate in their efforts to resolve false alarm outcomes. Results are presented to show the dominance of this approach over alternative strategies allotted to the game and to demonstrate the capability of exploiting the known nonhomogeneity in the search environment.