Multi-Agent Systems

Multi-Agent Systems (MAS) are used to design, model, analyze and simulate complex systems involving multiple interacting entities, whose coordination helps to achieve a collective goal or to promote system stability. The autonomous entities making up an ADM, known as agents, can be distributed and heterogeneous, and act in an open, evolving environment. The discipline of MAS provides the concepts, theories and tools needed to understand such systems at the right level of abstraction.

Over the last twenty years, this discipline has made remarkable progress, thanks to a combination of factors. Firstly, the discipline has opened up to other disciplines, in particular the humanities and social sciences (geography, sociology, linguistics, cognitive psychology, economics and, more recently, law), biology and mathematics. Initially motivated by applications, this openness today enables ADM to constitute a multi-disciplinary research field based on solid theoretical knowledge, validated and shared by a community. This field has also been able to continually exploit the results of Artificial Intelligence (logic, artificial learning), from which it itself originated. At the same time, specific progress has been made in terms of engineering, with the proposal of methods (GAIA, MOISE...), logic-based formalisms, notations (e.g. AUML) and standard languages (e.g. FIPA-ACL), as well as simulation platforms (GAMA, NetLogo) and development platforms (Madkit, JADE) enabling the industrialization of applications. This evolution has also been made possible by a particularly dynamic scientific community, organized around federating projects including the European AgentLink network of excellence, the annual international conferences AAMAS and IJCAI, and the journals JAAMAS, IJAOSE and JASSS.

This maturity now makes it possible to design agents capable of reasoning, cooperating, organizing, acting, anticipating, learning and adapting to changes in their potentially evolving environment. These capabilities endow ADMs with a rich power of expression, at both social and cognitive levels, enabling them to tackle complex problems in fields as diverse as intelligent ambient systems, collaborative robotics, supply chains or social simulation... By way of example, we could be talking about a flotilla of drones coordinating to monitor the evolution of a natural disaster and assist the actors on the ground and the crisis unit, or a team of soccer robots capable of taking on another (cf. robocup competition) by implementing a strategy. It can also be software entities simulating the actors in a social network and their emotions, leading to the discovery and understanding of emerging and unanticipated group phenomena (e.g. panic). Its fields of application have broadened due to the physical changes...