A layered control system for stereo vision head with vergence

This paper presents a stereo active vision system that is designed for a humanoid robot. The task was decomposed into three layers as saccade, tracking, and vergence. We adopted a developmental approach based on the developmental process of a human infant for the task decomposition. The three tasks were realized by using a layered network of modules implemented on a DSP network. For tracking, a search of the most similar region based on the sum of absolute difference (SAD) was utilized. The vergence module utilized a minimum SAD search for each pixel to obtain figure-ground separation in the 3D space. As a result of the combination of these modules and techniques, the system demonstrated real time tracking, vergence, and figure-ground separation in the real world