Consistent robot localization using Polar Scan Matching based on Kalman Segmentation

This work presents a revision of polar scan-matching procedure (PSM) in order to obtain an accurate robot localization in long runs without applying any other later procedure. In order to obtain an accurate robot pose estimation in long runs, PSM uses some SLAM procedure to correct error introduced by the scan-matching procedure. PSM compares scan points obtained from different positions, we have observed that if it associates points belonging to different objects, significant errors could appear in pose estimation. In our approach, an advanced line segmentation algorithm, based on Kalman filtering, is used to prevent as much as possible this kind of associations and to improve other areas in the original procedure. This new scan-matching procedure is named Polar Scan Matching based on Kalman Segmentation (PSM-KS). Experimental results in simulations show a reasonably accurate robot pose estimation, that outperform the accuracy obtained using odometry only, with low execution times. Consistent maps are obtained by simply overlapping estimated segments drawn from estimated poses. Consistent robot localization using only scan matching.Consistent robot mapping using only scan matching.Laser scan segmentation by means of regression using Kalman filtering.