On Fusion of Multiple Objectives for UAV Search & Track Path Optimization

Due to the significant advancement of the unmanned vehicle (UV) technologies in recent years, a great deal of research effort has been devoted to the problem of path optimization (planning and dynamic replanning) of a single or multiple UVs in uncertain and possibly hostile environments. While various UV mission scenarios have been considered in the literature, this paper is focused on UAV surveillance missions which typically include search (detection and localization) of new targets and possibly tracking of detected targets. The techniques considered, however, can be easily applied to other types of missions as well. Most of the literature on autonomous UAV surveillance deals with search oriented systems, e.g., [8], [7], [4], [15], [14]. Multiple-UAV tracking has been addressed in [9], [12], and tracking combined with detection has been dealt with in [10], [11]. In all of its variations an S&T mission includes multiple objectives, often conflicting to each other. At a high level these objectives can be grouped into several different types including, but not limited to, target detection, target tracking (classification, recognition), UAV survivability, UAV cooperation, UAV efficiency, and possibly others [7]. Quantifying various objectives and defining a fused scalar mission objective function to be optimized during a mission is a crucial issue in the design of S&T systems. Commonly, search-only systems use mission objective functions made up of, most often probabilitybased, gain/loss functions–e.g., cumulative detection probability, survival probability, etc. [8], [7], [4], [15], [14]. The tracking oriented systems of [9], [12] use information gain based mission objectives, in terms of the Fisher information matrix (FIM) of the tracking filters, and [10], [11] further include the detection objective measured also in terms of FIM. This makes it possible to use standard estimation fusion techniques [1] to fuse the detection and estimation objectives into a scalar objective. However, expressing all objectives through FIMs is difficult to extend to more complicated practical scenarios, e.g., to include efficiency (UAV flight regime cost) or other objectives. Achieving the mission goal is inherently a multiobjective optimization (MOO) problem and in this paper the problem of designing a mission objective function is treated as such–within the framework of the MOO methodology. There are two issues associated with the MOO formulation. First, due to the conflict among the individual objectives the solution in general is not unique. There is a set of optimal points (referred to as Pareto front) such that, loosely speaking, each optimal point corresponds to a certain trade-off among the values of the objective functions. A decision has to be made as to which Pareto optimal point provides the “best trade-off” among all the alternatives. The second issue is implementational–solving an MOO problem by the known computational methods is usually associated with solving a great number of single nonlinear