Agreement Technologies for Adaptive, Service-Oriented Multi-Agent Systems J. Santiago Pérez 1 , Carlos E. Cuesta 2 , and Sascha Ossowski 1 1 Centre for Intelligent Information Technologies (CETINIA), and 2 Kybele Research Group, Dept. Comp. Languages and Systems II Rey Juan Carlos University, 28933 Móstoles (Madrid), Spain {josesantiago.perez,carlos.cuesta,sascha.ossowski}@urjc.es Abstract. Multi-Agent Systems (MAS) are increasingly popular in Artificial Intelligence (AI) to solve complex problems. They can be conceived flexible and able to adapt to different situations. However, these features are often compromised by the characteristics of the problem itself. On the other hand, MAS have not had a lot of success in the industry, probably due to a different development culture. To solve this, MAS techniques should be more accessible to the general public, and have a shorter learning curve. The proposed approach is to use service-oriented concepts, which are popular in industry, to simplify this step. Moreover, if this approach manifests also self-adaptive capabilities, it will fulfil the notion's original promise: to guarantee that the system is able to adapt to changing conditions of the problem. This work proposes a service-oriented framework, consisting on a supporting agent-oriented architecture, a development methodology for service-oriented MAS, and an infrastructure based on the concept of agreement, which makes it adaptive. The first section provides a brief introduction and summarizes the paper goals. This is followed by the description of the base architecture, designed to support the agreement structure. Next section discusses concepts about service layers and the role of organizations. After that, the service-oriented methodology as well as the agreement structure itself is presented. Finally, a real-world case study, in the domain of medical emergencies, is analyzed, some conclusions are drawn, and further lines of work are outlined. Keywords: Multi-Agent Systems, Service-Oriented Architecture, agreement, coordination, adaptability. 1 Introduction The concept of agent has evolved, and nowadays MAS are increasingly popular in AI as a generic approach to solve complex problems. Different development strategies have been proposed in order to make them flexible and able to adapt to different situations. However, these features are often compromised by the heterogeneity of components, the nature of problems themselves, or the dynamism in the environment. On the other hand, MAS have not had a lot of success in the industry [14][36], probably due to a different development culture. To solve this, MAS techniques should be more accessible to the general software community. The proposed approach is to use service-oriented concepts, which are popular in industry, to simplify this step. Moreover, if this approach demonstrates also self-adaptive capabilities, it 118