Catálogo de publicaciones - libros

Various toolkits have been developed in the research community for constructing multi-agent systems. These toolkits differ in the types of programming languages they support, particular agent services they can be used to implement, and support for visualisation tools to specify agent behaviours. The focus of most tools is on software engineering support for constructing Multi-Agent Systems (MAS) examples of toolkits discussed in this volume are DECAIF, Agora and MADKIT. The approach in these toolkits is to operate at a higher level of abstraction than programming paradigms, enabling the description of particular features which contribute towards “agency”. Some research projects have adopted the operating system based approach of specifying an “agent kernel” which provides core low level services for interacting agents. Such core services can include agent interaction, and state management for an executing agent, for instance. Again the emphasis is to shift the focus of creating MAS away from low level aspects such as socket creation and message structure, to defiing the relationships between agents, and the role each agent plays within such a relationship. Although such an abstraction is desirable when constructing an agent system for solving a particular problem, it may not be if performance issues, such as response time, are of interest. Performance itself needs a clear definition in the context of MAS, and could be related to (1) wall clock time associated with low level agent infrastructure, contributing towards such application themes as Quality of Service (QoS), and response times to users, or it could be (2) associated with the functionality of the system, such as the efficiency, utility, correctness or capacity to undertake and complete a given role or action.