A BEHAVIOR-BASED APPROACH TO ACTIVE STEREO VISION FOR MOBILE ROBOTS

This paper describes an active stereo vision system that uses a behavi or-based approach to control gaze. The system concentrates its computa tional resources on cubic volumes of space called proximity spaces. Ea ch proximity space is controlled by a set of behaviors that combine to determine its location in the next processing frame. Multiple proximi ty spaces can be chained together using kinematic models to perform co mplex visual tasks. The vision system is mounted on a mobile robot and integrated into an intelligent control architecture. The intelligent control architecture serves to interpret the vision system output in a task context and produces appropriate robot motion. The resulting rob ot system can perform tasks such as pursuing other agents, recognizing natural gestures and avoiding obstacles. (C) 1998 Elsevier Science Lt d. All rights reserved.