Multi-Agent Itinerary Planning for Wireless Sensor Networks Min Chen 1 , Sergio Gonzalez 2 , Yan Zhang 3 , and Victor C.M. Leung 2 1 School of Computer Science & Engineering, Seoul National University, 151-744, Korea mchen@mmlab.snu.ac.kr 2 Elect & Comp Eng, University of British Columbia, V6T 1Z4, Canada sergiog,vleung@ece.ubc.ca 3 Simula Research Laboratory, 1325 Lysaker, Norway yanzhang@ieee.org Abstract. Agent-based data collection and aggregation have been proved to be efficient in wireless sensor networks (WSNs). While most of existing work focus on designing various single agent based itinerary planning (SIP) algorithms by considering energy-efficiency and/or aggregation efficiency, this paper identifies the drawbacks of this approach in large scale network, and proposes a solution through multi-agent based itinerary planning (MIP). A novel framework is pre- sented to divide our MIP algorithm into four parts: visiting central location (VCL) selection algorithm, source-grouping algorithm, SIP algorithm and its iterative algorithm. Our simulation results have demonstrated that the proposed scheme lowers delay and improves the integrated energy-delay performance compared to the existing solutions with the similar computation complexity. Keywords: Wireless sensor networks, mobile agent, itinerary planning. 1 Introduction The application-specific nature of a wireless sensor network (WSN) requires that sensor nodes have various capabilities. It would be impractical to store all the programs needed in the local memory of embedded sensors to run every possible application, due to the tight memory constraints. The intrinsically flexible features of mobile agent (MA) make it adaptable to diverse network conditions in dynamically reconfigurable WSNs. An agent deployed in a sensor network is a special kind of software that migrates among network nodes to carry out a task autonomously, in order to achieve the objec- tives of the sink node. Compared to its traditional client/server computing communications mechanism counterpart, mobile agent based computing has exhibited its unique efficiency in context- aware sensory environments [1, 2, 3, 4, 5, 6, 7, 8]. In a previous survey [1], we sepa- rated the agent design process for WSNs into four parts: architecture, itinerary planning, middleware system design, and agent cooperation for the design, development, and de- ployment of MA systems for high-level inference and surveillance in WSNs. Among the four components, itinerary planning determines the order of nodes to be visited during agent migration, which has a significant impact on energy performance N. Bartolini et al. (Eds.): QShine/AAA-IDEA 2009, LNICST 22, pp. 584–597, 2009. c  Institute for Computer Science, Social-Informatics and Telecommunications Engineering 2009