O. BarakatP. Lab. d’Automatique, CNRS, Institut de Productique, Besançon, France
P. Baptiste Lab. d’Automatique, CNRS, Institut de Productique, Besançon, France
The observation of anthills has been a source of inspiration in finding
a new way of solving certain distributed problems. This general purpose
algorithm is called the Ant System. In particular it seems to be well
adapted to the travelling salesman problem. The aim of this study is
firstly to simplify this approach and then to improve the overall
dynamics of the system. After having presented the classical algorithm
briefly, we define several new modules of decision for our agents. Then
we show how such a universe can have different behaviours which we try
to define precisely. Thus, the use of a simulation tool allows us to
show how the combination of the previously defined modules leads to
different speeds of overall convergence. In particular it allows us to
put the emphasis on the role of the evaporation agent. Finally we
discuss new ways of improving our algorithm and new fields where we
could try to use it.