Adaptive feature-based self-localization for autonomous mobile robots

Model based methods of object recognition like Alignment, Pose Clustering or Geometric Basing can be used for the global self-localization problem. For each algorithm parameters must be set in order to achieve good localization results. We present an adaptive algorithm: first for each algorithm and parameter set we classify a set of good localization positions representing scans. Then we extract characteristic properties from the scans and learn a descision tree. In a localization query, a generated scan will be decomposed in its characteristic attributes. After that a good parameter set can be found for each algorithm such that localisation queries can be computed. Present practical tests show promising results.