Monte Carlo Localization with Mixture Proposal Distribution

Recently, Monte Carlo localization (MCL) has been applied successfully to state estimation problems mobile robotics. This paper points out a limitation of MCL that is counter-intuitive, namely that better sensors can yield worse results. An analysis of this problem leads to the formulation of the ``dual'' MCL algorithm, which works well in cases were MCL fails. Combining both, MCL and its dual, leads to an extremely robust filter that consistently outperforms MCL and dual MCL.