Color Based Saccades for Attention Control

This paper addresses a simplified version of attention control - where a robot is asked to attend to scene points with a priori specified color. Differing from the classical approaches, in which generating fixations is based on explicit search, we introduce the requirement that the search strategy must be accompanied by a series of saccades whose nature control the fixation process. The robot must start from an arbitrary initial fixation point, start looking through the scene and find the target color points as it is doing so. In the explicit search approach, first, the scene point whose color is most similar to the ”looked-for” color is determined and then the camera is made to move to that point. In the artificial potential functions approach, the two stages are merged together where the camera simply starts moving towards a point whose color is similar to the target color – although not necessarily the most similar. We present working implementations of the two approaches – reporting actual experiments with an attentive robot and comparing the resulting search behaviors using quantitative measures on saccadic selectivity and optimization.