Résumé: Large pervasive systems, deployed in dynamic environments, require flexible control mechanisms to meet the demands of chaotic state changes while accomplishing system goals. As centralized control approaches may falter in environments where centralized communication and knowledge may be impossible to implement, researchers have proposed decentralized control methods that leverage agent-driven, self-organizing behaviors, to achieve reliable, flexible systems. This article presents and compares the performance of three decentralized control approaches in the online multi-object k-assignment problem. In this domain, a set of sensors is tasked to detect and track an unknown and changing set of targets. Results showthat a proposed hybrid approach that incorporates supervisory devices within the population while allowing semi-autonomous operations in non-supervisory devices produces a flexible and reliable system capable of both high detection and coverage rates.

Langue: Anglais