In the framework of autonomous navigation, the use of a cognitive map requires to perform fast and robust storage of different pieces of information. Classical models of auto-associative memory have been proven to be limited in such a context because of the well known catastrophic interference phenomenon. Taking strong inspiration from the inner organization of the hippocampus, we present in this paper a model of auto-associative memory based on 4 distinct structures (EC, DG, CA3 and CA1) where information is processed along a loop at three distinct levels and has been tested successfully on a real robot.