In this Habilitation Thesis, I present my research activity in the field of multiagent simulation within the SMAC team (LIFL, Université Lille 1), since the beginning of my career of teacher and researcher in 2002. My work focuses on the design and implementation of methods and tools aimed at facilitating the modeling of large-scale complex systems. Therefore, I developped with my colleagues an " interaction-oriented " approach, characterized by the unification of several concepts used in the field of MAS. It is supported by a large methodological and algorithmic elaboration (the IODA method), in which each entity of the model is represented by an agent and each behavior by a rule called an interaction. This method relies upon the separation between declarative and procedural features, which facilitates the acquisition of thematic expertise. In addition, several software tools have been built throughout this research (e.g. the JEDI platform and the IODA exension for the NetLogo platform). This method has also been applied to various fields (cell biology, serious games, marketing, cartography). To conclude, I present my research project for the coming years, which proposes to combine issues raised by recent work: on the one hand, the multi-level simulation (aimed at defining an operational framework for automatizing the change of observation scale or point of view upon the subsystems of a complex phenomenon) ; and on the other hand, the automatic information retrieval from real data, in order to increase the behavioral realism of populations of agents. In addition, a collaboration with the IGN on using those techniques for cartographic generalization opens perspectives towards the transposition of those simulation methods to the field of problem solving.
