Automatic modeling can be seen as a process of selection and assembling of predefined elementary sub-models called fragments. These fragments model a fraction of the application domain (a phenomenon, an equation, a physical component, ...) and are organized into a library.

Fragments often rely on hypotheses made at different levels which are partially or totally contracditory (hypotheses on the structure of the modeled system, choices in modeling, ...)

The objective of the automatic modeling based on the assembling of fragments is to determine, in the library, the set of convenient fragments which can constitute the more simple, complete and consistent model solving a given problem.

Existing works on model construction generally offer an automatic approach: once the fragments library has been built, the user provides a description of the system to be modelled and gets, if the library is correct, a model of this system.

In practice, this lacks from flexibility as soon as the library reveals itself incomplete or not adequate for a portion of the model.

The ROMANS Action proposes a semi-automatic model construction. The user can use the library to design certain parts of the model and to build by herself the other parts of the model under the control of the composition system which is in charge of the global consistency of the model. A formalism  for describing model fragments and the specifications of the composition engine are currently studied. 

In collaboration with the geographs of the P.A.R.I.S research team (CNRS URA 1243), and with the support of the "Modeling and Numeric Simulation" CNRS program, some experiments are carried on on the dynamics of town systems. It consists in simulating the emergence of towns and town systems from an initial seedling of places of living and several hypotheses about the migrations between these places.