A central problem in robot navigation is recognizing when a robot is
somewhere that it has been before. Without ``loop closing'', the
robot's position uncertainty increases without bound; consequently
navigation, map-building, and other common robot tasks become
impossible. We present an algorithm that considers groups of several dozen loop
closure hypotheses and robustly rejects the incorrect hypotheses. This
paper's central contribution is showing how to map the loop-closing
problem onto the Single Cluster Graph Partitioning (SCGP) problem,
which has an efficient solution.
This paper has been revised and expanded. Click here to go to the updated paper.
Best Paper
@inproceedings{olson2006csw, AUTHOR = {Edwin Olson}, TITLE = {Recognizing Places with Weak Evidence}, BOOKTITLE = {CSAIL Student Workshop Proceedings}, YEAR = {2006}, }