An Attentional System for a Humanoid Robot Exploiting Space Variant Vision

Abstract

This paper describes an implementation of an attentional system for a humanoid robot based completely on space variant vision (in particular log-polar). The aim is that of providing the robot with a suitable measure of position, speed and saliency of possibly interesting objects for saccading and tracking. The major advantage of log-polar based imaging is related to the reduced number of pixel while maintaining a large field of view. This arrangement is very well suited for motor control, where the high-resolution center (fovea) alows precise positioning and, at the same time, the coarse resolution periphery permits detection of potential targets. Algorithms for color processing, optic flow, and disparity computation were developed within this architecture. The attentional modules are intended as the first layer of a more complicated system, which shall include learning of object recognition, trajectory tracking, and naive physics understanding during the natural interacion of the robot with the environment.