A Factored PF-SLAM for Robot Localization with a Partially-Incorrect Map

K. Tanaka and E. Kondo (Japan)


SLAM, global localization, partiallyincorrect maps, non stationary environments


In this paper, we address the problem of robot self localization in large and non-stationary environments, given a partially-incorrect map. Particle filter -based SLAM is robust technique for localization in stationary en vironments. However, in partially known environments, it becomes inefficient due to multiple initial self-position hy potheses. To solve this problem, we propose a factored par ticle filter represented in local coordinate system defined for each initial self-position hypothesis. Moreover, we uti lize distance filter and object filter to reduce the initial hy potheses. We have tested the method in various environ ments that contain many self-similar movable landmarks and less fixed landmarks.

Important Links:

Go Back