Rollout Algorithms for Wireless Sensor Network-Assisted Target Search

Abstract

We consider a mobile, autonomous searcher that aims to find the source of a broadcast message in a network of location-agnostic wireless sensor nodes. In certain types of networks, the hop count of the broadcast message, given the distance from the source node, is well approximated by a simple parametric distribution. The mobile searcher can interrogate a nearby node to obtain, with a given success probability, the hop count of the broadcast message. The search is modeled as an infinite horizon, undiscounted cost, and partially observable Markov decision process. A computationally efficient approximate online solution is obtained through policy rollout using a novel heuristic. Simulation results show that our rollout approach outperforms commonly used search methods based on a mutual information utility. We quantify the loss due to the use of an approximate hop count observation model and study the effect of statistical dependence between observations. Furthermore, we discuss how to account for this dependence by adapting an integer autoregressive model for the hop count.Index Terms-Sensor decision and fusion, networkable sensors-actuators, wireless sensor networks, robotics and automation applications, localization, POMDP, Monte Carlo methods, rollout algorithms.

Policy Rollout Action Selection in Continuous Domains for Sensor Path Planning

“…Note that the absolute error is not reduced, however, only the relative error is important to make the right decision. Here we implemented CRN, such that x t aj =x t a j (21) for the same rollout number j and different actions a and a . Note thatx t aj is still randomly sampled.…”

Study of Wireless Sensor Network Location Algorithm and Data Collection Protocol

“…As shown in Figure 1, the inside of the small cycle is the round cluster head candidate zone. The node decides whether it belongs to this zone depending on formula (10).…”

“…Assuming a sink moves once each other T rounds [10]. Figure 4 shows the data flow chart of sink stop points.…”

Wireless Sensor Network Multi-Hop Positioning Algorithm Based on Continuous Regression

“…m+n } is regression coefficient and e is random error. Equation (4) shows that there are serious multiple correlations in the variables in the equation, or the quantity of sample points in the equation is less than the quantity of the variables [16]. If Equation (4) is forcibly calculated in this circumstance, the estimate will be invalid.…”