A probabilistic scan matching algorithm applied to grid-based FastSLAM

A novel scan matching algorithm called Combined probabilistic scan matching (CPSM) in which Rao-Blackwellized particle filters (RBPF) is combined with Probabilistic scan matching (PSM) is proposed for grid based FastSLAM. The algorithm projects the raw laser data into occupied grids directly, and measures the similarity between maps using the Probability distribution function (PDF) which is determined by the Mahalanobis distance from each occupied grid in the map computed at the current robot position to the closest occupied grid in a previous scan map of the environment. So the algorithm can reduce the computational costs of establishing correspondence, since it is not necessary to establish an explicit correspondence between raw laser data. Further, the algorithm can prevent the problem of particle depletion and decrease the map generating time, since the lower variance in the corrected odometry by probability scan matching reduces the number of necessary resampling steps. Experimental results in three kinds of indoor environments prove the advantages of the proposed method over previous approaches with respect to robustness and the ability to generate high quality maps in real time.