A context and task dependent visual attention system to control a mobile robot
In this paper an artificial neural network model is proposed for the selection of spatially relevant visual information present in an unknown environment. An attentional mechanism is used to control the orientation of a mobile robot. In particular the system is built to be independent of the nature of the target but dependent on the surrounding context and on the task in which the agent is involved. The computational model take into account neurobiological and psychophysical data related to pop-out and attentional search studies. It consists in a bottom-up process leading to an unique salient map. To constrain the selection of expected object, an additional top-down bias is introduced to favor the meaningful features according to the task. We will show how such a neuronal architecture is able to rapidly direct the gaze of a mobile robot towards particular objects according to the attentional bias. This behavioral approach favoring the dynamical interplay between environment and task, brings new insights to the nature of mechanisms such as object recognition.