Coordinating Teams of Mobile Sensors for Monitoring Environmental Phenomena

Stranders, Ruben (2009) Coordinating Teams of Mobile Sensors for Monitoring Environmental Phenomena s.n.


[img] PDF mini_thesis.pdf - Author's Original
Download (3MB)


Mobile wireless sensors can play a vital role in achieving situational awareness in uncertain and changing environments by keeping track of environmental phenomena, such as temperature, gas concentration and radiation, that exhibit spatial and temporal correlations. Examples of such environments are commonly found in disaster response, where the safety and effectiveness of response units critically depends on the accuracy of estimation of the state of the world. In these environments, mobile sensors operating in a team can improve situational awareness by offering a high sensing resolution in a timely and efficient way. In order to do this efficiently, however, they need to coordinate their movements. This coordination is a challenging task, since the sensors operate in an environment that is highly uncertain and dynamic, have a limited perception of their surroundings, and have limited communication with adjacent sensors. Consequently, coordination mechanisms need to address the challenges involved in maximising the collective information gain of the entire team, in the presence of uncertainty and different world views. Previous work in this area has focused on the use mobile and fixed wireless sensors for environmental monitoring, but fails to provide a principled online, decentralised coordination mechanism for such settings. In this report, we study the challenge of coordinating teams of mobile sensors for monitoring environmental phenomena. In order to do so, we review the literature on wireless (mobile) sensor networks, information processing, target tracking, and localisation and mapping. In particular, we focus on the key concept of adaptive sampling, which encompasses a set of techniques that aim to maximise information gain subject to movement constraints and the limited resources at a sensor's disposal. Based on this review, we present a general architecture for a sensor that makes a clear distinction between information processing, information valuing and maximising information gain. In more detail, we show that the state of the art in adaptive sampling falls short of providing robust, scalable, decentralised coordination algorithms. To address these shortcomings, we develop two online, decentralised coordination algorithms for monitoring spatial phenomena. The first algorithm operates in an un-negotiated coordination mode, in which coordination is achieved exclusively through the exchange of observations; sensors need not coordinate (negotiate) about the actions they are about to take, but base their decisions solely on the picture of the state of the environment that they compiled using their own observations and those received from their neighbours. This algorithm is based on two techniques found in previous work. Firstly, Gaussian process regression (Rasmussen2006a, Osborne2008), which is used for processing the raw observations obtained by the sensors and for predicting unobserved measurements. Secondly, myopic information-theoretic control (as found in Grocholsky2002), which is used for maximising the informativeness of the samples that are obtained by moving the sensors to locations where the environment is more uncertain. The second algorithm extends the first by adding a negotiation stage, which results in negotiated coordination. This algorithm is based on the max-sum message passing algorithm for decentralised control (Farinelli2008), which allows the sensors to maximise a team objective function in a decentralised fashion. To make the max-sum algorithm suitable for solving the mobile sensor coordination problem, we develop two pruning algorithms that drastically reduce the amount of computation required. These pruning algorithms are generic in the context of applying the max-sum algorithm, and are thus not limited to the mobile sensor setting. Finally, we extend the negotiated algorithm for sensors that are characterised by continuous control parameters (for example their heading and velocity). To this end, we generalise the discrete max-sum algorithm to the continuous case in which the interactions between sensors are characterised by continuous piecewise linear functions.

Item Type: Monograph (Project Report)
Organisations: Agents, Interactions & Complexity
ePrint ID: 267535
Date :
Date Event
May 2009Accepted/In Press
Date Deposited: 08 Jun 2009 11:08
Last Modified: 17 Apr 2017 18:46
Further Information:Google Scholar

Actions (login required)

View Item View Item