A biologically inspired active vision gaze controller

Abstract

An active vision system (AVS) has been created as a general purpose vision system for a robot. The AVS enables the selective, localized capture of imagery for visual analysis. With it, the robot can direct its gaze around the environment to track an object or to perform a visual search. The ability to react to, or probe the environment simplifies a number of the robot's interactions with the world. This work presents both the design and the implementation of a camera head controller for the AVS of the humanoid, ISAC. The camera head comprises two color cameras mounted on a four degree-of-freedom head (pan, tilt, left verge, and right verge). The camera head controls produce camera movements that are analogous to those of human eyes. Human eye-motion can be classified into three voluntary movements: saccades, smooth pursuit, and vergence, and two involuntary movements: vestibulo-ocular reflex and opto-kinetic reflex. Control methods have been designed and tested through experimentation. The Five basic human-like eye movements have been robustly attained and have resulted in demonstrable improvements in overall visual performance. Using a distributed network of personal computers which are tightly integrated through a unique software architecture, the AVS performs with more than adequate speed and accuracy. Furthermore, because of the simple structure of its hardware, this AVS is conveniently portable to other robots. The results of using the described camera head motions are faster performance, and smoother, more stable, and more robust control of the camera head than was possible with more traditional control methods. This AVS is implemented as an integral part of the humanoid robot ISAC in Intelligent Robotic Laboratory.