Distributed AI for Ambient Intelligence: Issues and Approaches Theodore Patkos 1 , Antonis Bikakis 1 , Grigoris Antoniou 1 , Maria Papadopouli 1 , and Dimitris Plexousakis 1 Institute of Computer Science, FO.R.T.H. Heraklion, Greece {patkos, bikakis, antoniou, mgp, dp}@ics.forth.gr Abstract. Research in many fields of AI, such as distributed plan- ning and reasoning, agent teamwork and coalition formation, cooper- ative problem solving and action theory has advanced significantly over the last years, both from a theoretical and a practical perspective. In the light of the development towards ambient, pervasive and ubiqui- tous computing, this research will be tested under new, more demanding realistic conditions, stimulating the emergence of novel approaches to handle the challenges that these open, dynamic environments introduce. This paper identifies shortcomings of state-of-the-art techniques in han- dling the complexity of the Ambient Intelligence vision, motivated by the experience gained during the development and usage of a context- aware platform for mobile devices in dynamic environments. The paper raises research issues and discusses promising directions for realizing the objectives of near-future pervasive information systems. Key words: Ambient Intelligence, Distributed AI, Context Awareness, Action Theories, Multi-agent Cooperation 1 Introduction The vision of Ambient Intelligence assumes a shift in computing towards a mul- tiplicity of communicating devices disappearing into the background, providing an intelligent, augmented environment, where the emphasis is on the human fac- tor. Realizing this vision requires the integration of expertise from a multitude of disciplines; distributed intelligence, dynamic networks and ubiquitous commu- nications, human-computer interaction and intuitive user-friendly interfacing, robotics, and hardware design are embraced under the influence of the Ambient Intelligence paradigm. This paradigm implies a seamless medium of interaction, advanced networking technology and efficient knowledge management, in order to deploy an environment, where entities describe themselves, are aware of each other and can figure out ways to interoperate at syntactic and semantic levels. Arranging a physical environment, where mobile and stationary devices com- municate and cooperate to achieve common objectives, has proven to be a labori- ous task for the research community. Although much success has been achieved in defining theoretical frameworks for the fields of distributed AI, agent teamwork