Dynamic sensor tasking in heterogeneous, mobile sensor networks

Abstract

Modern sensor environments often attempt to combine several sensors into a single sensor network. The nodes of this network are generally heterogeneous and may vary with respect to sensor complexity, sensor operational modes, power costs and other salient features. Optimization in this environment requires considering all possible sensor modalities and combinations. Additionally, in many cases there may be a time critical objective, requiring sensor plans to be developed and refined in real-time. This research will examine and expand on previous work in multi-sensor dynamic scheduling, focusing on the issue of near optimal sensor-scheduling for real-time detection in highly heterogeneous networks. First, the issue of minimum time inference is formulated as a constrained optimization problem. The principles of dynamic programming are applied to the problem. A network model is adopted in which a single "leader" node makes a sensor measurement. After the measurement is made, the leader node chooses a successor (or chooses to retain network leadership). This model leads to an index rule for leader/action selection under which the leader is the sensor node with maximum expected rate of information acquisition. In effect, the sensor and modality with the maximum ratio of expected entropic decrease to measurement time is shown to be an optimal choice for leader.

(cont.) The model is then generalized to include networks with simultaneously active sensors. In this case the corresponding optimization problem becomes prohibitively difficult to solve, and so a game theoretic approach is adopted in order to balance the preferences of the several sensors in the network. A novel algorithm for multiplayer coordination is developed that uses iterative partial utility revelation to achieve bounded Pareto inefficiency of the solution.