Reducing Drift of Lidar Odometry by Incorporating OpenStreetMap Building Data

Lidar odometry is the problem of estimating the movement of a mobile robot based on the alignment of consecutive lidar scans. Due to the accumulation of errors in the individual alignments, lidar odometry is prone to drift. The incorporation of prior maps has proven an effective strategy for mitigating this problem. While existing methods use an additional optimization procedure, we introduce a closed-form approach for the fusion of pose estimates from the lidar odometry and map. In particular, we motivate and implement a complementary filter-inspired fusion method that simultaneously reduces drift introduced by frame-to-frame alignments as well as errors introduced by potential inaccurate prior maps. An exhaustive evaluation of our method on the KITTI odometry and Boreas data sets, as well as on our own recorded data, demonstrates the benefits of our approach in a wide variety of urban and suburban situations, outperforming other state-of-the-art OpenStreetMap-assisted odometry approaches.