Reducing localization errors by scan-based multiple hypothesis tracking

The purpose of this approach is to significantly reduce the accumulation of odometric errors, while an autonomous robot moves through a completely unknown environment. The information of the displacement of the robot is obtained by analyzing the trajectories (tracks) of selected scan-points and Hessian parameters of line segments in the local coordinate frame. The origin of each track is transformed into the reference coordinate system and is regarded as a dynamic landmark, as long it is detected by the laser rangefinder. While the robot moves, the algorithm tries to determine certain scan-points, line segment parameters respectively, as focal representatives of this landmark. To handle the association ambiguity of scan data for tracking, a multiple hypotheses tracking approach is used. Hypotheses for the global robot position are computed by using local tracks and their associated global landmarks. The fusion of these hypotheses is done by using a weighted sum fusion approach.