Calibration-free Visual Navigation of Robot in a Maze

In this article we present a novel navigation approach for an autonomous robot in an unknown maze based on calibration-free visual information. On one hand, we segment images to obtain the approximate traversable region and on the other hand the lines and corners in images are also extracted to help a robot understanding the topological structure. In the experiment, our navigation approach running on a pioneer 3-DX robot is proved to be effective.