Sensor Scheduling for Energy-Efficient Target Tracking in Sensor Networks

Abstract

In this paper we study the problem of tracking an object moving randomly through a network of wireless sensors. Our objective is to devise strategies for scheduling the sensors to optimize the tradeoff between tracking performance and energy consumption. We cast the scheduling problem as a Partially Observable Markov Decision Process (POMDP), where the control actions correspond to the set of sensors to activate at each time step. Using a bottom-up approach, we consider different sensing, motion and cost models with increasing levels of difficulty. At the first level, the sensing regions of the different sensors do not overlap and the target is only observed within the sensing range of an active sensor. Then, we consider sensors with overlapping sensing range such that the tracking error, and hence the actions of the different sensors, are tightly coupled. Finally, we consider scenarios wherein the target locations and sensors' observations assume values on continuous spaces. Exact solutions are generally intractable even for the simplest models due to the dimensionality of the information and action spaces. Hence, we devise approximate solution techniques, and in some cases derive lower bounds on the optimal tradeoff curves. The generated scheduling policies, albeit suboptimal, often provide close-to-optimal energy-tracking tradeoffs.

A unified framework for energy efficient physical activity tracking

“…, f(y k |e n , u k−1 )). The optimization problem formulated in (9) can be solved using the finite-horizon DP equations given in Theorem 2 in terms of the predicted belief state.…”

“…Finally, the "perfect sensing of state" assumption, i.e. perfect sensing of the unknown phenomenon of interest is achieved if transmission is successful, that is usually adopted [8], [9], [12], fails also in our setting.…”

A Weiss-Weinstein lower bound based sensing strategy for active state tracking

“…Thus, the decision making process, guided by appropriately defined performance objectives, exploits the knowledge of previous measurements to adapt the sensing strategy. Active state tracking problems arise in various applications: sensor scheduling for target detection and tracking [1], health care [2], generalized search [3], [4] and waveform selection for radar imaging.…”

“…The active state tracking problem 1 has been previously studied for both time-invariant [4], [8] and time-varying cases [1], [2], [9]. For the latter, most prior work assumes discrete observations [2], [9], which is not realistic for the types of problems we are interested in.…”

“…In contrast, our focus is on time-varying systems with Gaussian observation vectors. Our work also considers a more general signal model than [1], which assumes w Gaussian samples from w sensors under independence assumptions, since we account for multiple correlated measurements from different sensors. We underscore that even though the approaches in [1] are fairly general, they are still tied to the target tracking application.…”