Integrating Self-organisation into Dynamic Coalition Formation (Extended Abstract) Dayong Ye University of Wollongong NSW 2522 AU dy721@uow.edu.au Minjie Zhang University of Wollongong NSW 2522 AU minjie@uow.edu.au Danny Sutanto University of Wollongong NSW 2522 AU danny@elec.uow.edu.au ABSTRACT In some real systems, e.g., sensor networks, individual agents will often need to form coalitions to accomplish complex tasks. Due to communication or computation constrains, it is infeasible for agents to directly interact with all other peers to form coalitions. Most current coalition formation works, however, overlooked this aspect. Those works usually did not provide an explicitly modeled agent network or as- sumed agents in a fully connected network, where an agent can communicate with all other agents. Thus, to alleviate this problem, it is necessary to provide a neighbourhood sys- tem within which agents can directly interact only with their neighbours. Towards this end, in this paper, we propose a dynamic coalition formation mechanism, incorporated with self-organisation, in a structured agent network. Based on self-organisation principles, this mechanism enables agents to dynamically adjust their degrees of involvement in differ- ent coalitions and to join new coalitions at any time. Categories and Subject Descriptors I.2.11 [Artificial Intelligence]: Distributed Artificial In- telligence General Terms Algorithms Keywords Coalition Formation, Self-organisation 1. INTRODUCTION In many applications of multi-agent systems, agents will need to dynamically join together in a coalition to complete a complex task which none of them can complete indepen- dently. Recently, many efforts have been done on coalition formation and have achieved very great results. There is a common assumption in these studies that the agent network underlying structure is either not explicitly modeled or the network structure is based on some regular structures, e.g., a fully connected network or a hierarchical network. How- ever, in many real circumstances, particularly in large and distributed environments, it is infeasible for each individu- al agent to consider all the other agents to form coalition- Appears in: Proceedings of the 11th International Con- ference on Autonomous Agents and Multiagent Systems (AAMAS 2012), Conitzer, Winikoff, Padgham, and van der Hoek (eds.), 4-8 June 2012, Valencia, Spain. Copyright c  2012, International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved. s due to time, communication and computation constrains [4]. One approach to overcome this limitation is to impose some sort of network structure on the agents and require that agents can directly communicate only with their neigh- bours when forming coalitions. Gaston and desJardins [2, 3], and Glinton et al. [4] made many efforts in this way. The common limitation in [2, 3, 4] is that an agent can join only one coalition and once a coalition is formed for a task, the coalition is fixed and agents cannot leave the coali- tion, until the task is finished. Against this background, in this paper, our research concentrates on designing a dy- namic coalition formation mechanism in a structured agent network, where each agent has only a limited view about it- s neighbours in the environment and makes decisions based only on this view. In addition, we integrate self-organisation notion into coalition formation which enables agents to dy- namically adjust their degrees of involvement in different coalitions and to join new coalitions, via negotiation, at any time if necessary. In that case, agents have more autonomy and flexibility when they execute tasks. 2. COALITION FORMATION In the agent network, agents make decisions based on- ly on local information about the system, and the deci- sion making process of agents is autonomous without ex- ternal control. Hence, we define a set P = {P1, ..., Pn}. P is defined as a partition of the Compatible Relation R, where 〈ai ,aj 〉∈ R if and only if aj is a neighbour of ai . Accordingly, it can be obtained that  1≤i≤n Pi = R and ∀Pi ,Pj ∈ P : i = j ⇒ Pi ∩ Pj = ∅. The set P can be generated by using Algorithm 1. Algorithm 1: Create a partition P on relation R begin: (1) for each a i , a i ∈ A, in sequential order (2) if ∃a j ∈ A : 〈a i ,a j 〉∈ R then (3) P i ← P i ∪{〈a i ,a j 〉}; end The coalition formation mechanism is illustrated in Algo- rithm 2 as follows. Algorithm 2: Coalition Formation Mechanism begin: (1)Call Algorithm 1 to generate P; (2)for each θ i , θ i ∈ Θ, in sequential order /*θ i is a subtask of Θ*/ (3) randomly select an IDLE agent, a i ∈ A, as Initiator;