PhD Thesis proposal Université d’Angers, France Reactive scheduling for target tracking with wireless networks networks under uncertainty André Rossi and Marc Sevaux May 2017 Overview This thesis is to develop a decision support system for tracking mobile targets with a network of wireless sensors deployed in an uncertain environment. This will be achieved by developing optimization methods for exploiting these networks, and dynamic online algorithms to deal with large amplitude uncertainties, like sensor failures and targets unexpected trajectories. The thesis is jointly supervised by Prof. André Rossi, LERIA lab, University of Angers, France, and by Prof. Marc Sevaux, Lab-STICC, University of South Brittany, Lorient, France. Context and working directions This work is to extend the PhD thesis of Charly Lersteau, defended in 2016 [1]. We consider a wireless sensor network deployed in a potentially hostile environment where energy supply is impossible. This network is dedicated to monitor mobile targets, that may be allied, neutral or enemies. The present thesis builds upon an article published in EJOR [2] as well as on a second article currently in revision. The input model of the problem is based on a procedure called discretization, that is now available to explore more realistic problem variants, and to interface the results with a wireless sensor network simulator called CupCarbon http://www.cupcarbon.com/, designed at Lab-STICC and currently used in ANR project Persepteur. This simulator will allow to represent and check and play the solutions under different realistic scenarios. This thesis will in particular consider the cases where uncertainty is beyond the threshold for which existing robust solutions [3] can offer performance guarantees. To this end, reactive methods will be proposed, based on off-line generation of partial solutions. These partial solution will then be implemented online, when necessary, possibly using CupCarbon as a prof of concept platform. These partial solutions, that can be regarded as paths in a directed graph, will be used only if necessary, i.e. in the case where the robust solution is no longer appropriate to large magnitude uncertainties. It is then necessary to combine the advantages of a robust solution and the required flexibility that allows to timely implement appropriate reactions. Another important feature is the long term management of the network, that allows the network owners to maintain a high coverage for a given set of geographical zones. Doing so avoids the bias of short-term optimization, where energy minimization of the current target tracking mission would the only concern. Hence, we propose to optimize the use of the network both at strategic and operational levels. Expected work The wireless sensor network can be used to perform two kinds of missions: reconnaissance missions and target tracking missions. A reconnaissance mission is to identify all the targets inside the sensing zone of the network. It is usually the first mission once the network is deployed. The most natural (but highly energy inefficient) way to do that is to activate all the sensors at the same time. However, static target coverage can be used [4] to achieve a reconnaissance mission at minimum energy cost. A variant of this mission is to focus the reconnaissance mission to a restricted geographical zone. After a reconnaissance mission (that is typically very short), the network managers designate one or more targets to track among those that have been detected. This is the starting point of a target tracking mission, that can be seen as a series of static target coverage missions, thanks to the discretization step. Efficient hybrid column- generation based algorithms are available to address this series of problems. A target tracking mission terminates when all the targets have left the coverage zone of the network, or if the network managers decide to abort it. A 1/3