Monte Carlo localization in outdoor terrains using multilevel surface maps

We propose a novel combination of techniques for robustly estimating the position of a mobile robot in outdoor environments using range data. Our approach applies a particle filter to estimate the full six-dimensional state of the robot and utilizes multilevel surface maps which, in contrast to standard elevation maps, allow the robot to represent vertical structures and multiple levels in the environment. We describe probabilistic motion and sensor models to calculate the proposal distribution and to evaluate the likelihood of observations. We furthermore describe an active localization approach which actively selects the sensor orientation of the 2D laser range scanner to improve the localization results. To efficiently calculate the appropriate orientation we apply a clustering operation on the particles and only evaluate potential orientations based on these clusters. Experimental results obtained with a mobile robot in large-scale outdoor environments indicate that our approach yields robust and accurate position estimates. The experiments also demonstrate that multilevel surface maps lead to a significantly better localization performance than standard elevation maps. They additionally show that further accuracy is obtained from the active sensing approach. This is a preprint of an article published in Journal of Field Robotics, Vol. 25, Issue 6-7, pp. 346-359, June July 2008, available online http://www3.interscience.wiley.com/journal/111090262/home Copyright c ©2008 Wiley Periodicals, Inc., A Wiley Company Figure 1: Elevation map (left) and multilevel surface (MLS) map (right) of the Freiburg campus. The MLS map represents vertical structures more accurately and can deal with multiple surfaces that can be traversed by the robot. Figure 2: Advantage of the MLS map approach in comparison to the standard elevation maps. In contrast to the MLS map (right) the elevation map (left) lacks the ability to model vertical structures, because it averages over all measured height values. Since the distance of the endpoint of a laser beam to the closest point in the elevation map can have substantial deviations from the true distance, localization becomes harder.