defended: 20 december 2006
Localization is one of the basic skills of a mobile robot. Much progress has been made in this field over the past years. In particular, the RoboCup competitions have focused their research within robotics on a unified challenge problem, and the yearly rule adaptations aim at increasingly approximating robotic soccer to the real world.
Until now, most approaches, however, still rely on special sensors (like laser range scanners, preferably used in the RoboCup rescue leagues) or artificial environments (like color-tagged soccer fields, as used by the RoboCup soccer leagues).
In this thesis, a novel approach is presented that can provide compass information purely based on the visual appearance of a room. A robot using such a visual compass can quickly learn the appearance of previously unknown environments. By the level of resemblance, a robot can additionally recognize qualitatively how far it is away from the former training spot. This can be used for visual homing.
The visual compass algorithm is efficient, scaleable and can therefore work in real-time on almost any contemporary robotic platform. Therefore, the approach has been implemented on the popular Sony entertainment robot Aibo.
Extensive experiments have validated that the approach works in a vast variety of environments. It has been shown that a visual compass can supply a mobile robot in natural environments with accurate heading estimates. Finally, it is shown in experiments that a robot using multiple compasses is able to estimate its translational pose.
last modified on 2008/09/18 16:57