Multiscale pedestrian modeling with CA and agent-based approaches: Ubiquity or consistency?

The simulation of complex system often pushes the modelers to face issues related to conflicting goals, constraints and limits of the available computational instruments: we often want to simulate large scale scenarios but with very good computational performances. A way to deal with this kind of situation is to couple simple modeling approaches with more fine grained representations of portions of the simulated system requiring higher degree of fidelity. This paper describes an approach adopting this scheme for large scale pedestrian simulation and focusing on issues related to the connection of the models representing the system at different granularities. In particular, to achieve a consistent behavior of the adopted models, in certain portions of the environment a single pedestrian needs to be represented in both models at the same time.