Team : SMA
Departure date : 07/01/2013
Supervision : Amal EL FALLAH SEGHROUCHNI Co-supervision : CORRUBLE Vincent
Coordination in Multi-Agent Systems : The timed patrolling problem with an open system setting
The timed multi-agent patrolling task is an interesting problem to test and compare coordination strategies on a cooperative task: visiting repetitively and frequently areas of interest in a graph. However, until now this task was limited by strong hypothesis: the system is closed, and the communications are synchronous. These hypotheses limit strongly both its complexity and applicability. In this thesis, we propose to lift these hypotheses and study the behaviour of various coordination strategies on the new task. In a first time, we propose a framework for the new problem: formal description, metrics, and strategies to use as references. In a second time, we show the impact of lifting these hypotheses on auction-based, decentralized strategies. We propose various mechanisms to cope with the open system setting and decrease its impact. We also show that it is possible to apply computational social choice theory concepts to the patrolling task and create cooperative, decentralized auction-based strategies that perform as well as centralized, reference strategies. Finally, we open the patrolling problem on the scaling perspective, both in terms of graph size and population size.
C. Poulet, V. Corruble, A. El Fallah‑Seghrouchni, G. Ramalho : “The Open System Setting in Timed Multiagent Patrolling”, IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT), vol. 2, Lyon, France, pp. 373-376, (IEEE) (2011)