The human brain is able to process a huge quantity of information, thanks to a massively parallel functioning and distributed representations. However, it is not able to process each information with a similar degree of accuracy, choices have to be made, hence one of the roles of selective attention. Getting inspired from cerebral structures reported as involved in that attentional mechanism, we will describe a model of visual attention, based on fully distributed architecture and functioning, that enables a robot to explore a dynamic visual scene in an efficient way. We will also explain how that mechanism is linked to the ability to anticipate the consequences of one's action on the external world.