Partially distributed optimization for mobile sensor path-planning

Abstract

Mobile sensor platforms may provide low-cost, versatile means for obtaining high resolution information from spatially distributed processes. The aim of this paper is to efficiently calculate receding horizon-type motion plans that optimize information retrival by mobile sensors from distributed processes. Formulating the problem in a rather general framework as minimization of entropy, gives a huge, non-convex, in general intractable optimization problem for path planning. Based on this formulation, using approximation and decomposition strategies, we propose a new, more computationally tractable framework. In the case of multiple mobile sensors, the framework allows a partially distributed setup where each mobile sensor optimize its path using an economic-type Model Predictive Controller, where the “economic” (in the sense of information) objective is constructed based on simulation of covariances for the distributed process, performed by a central entity.