Construction of Autonomous Driving Maps employing LiDAR Odometry

Autonomous driving typically relies on a priori maps for localization and path planning. In order to construct such maps, data from perception sensors such as light detection and ranging (LiDAR), global positioning system (GPS), inertial measurement unit (IMU), etc. are employed in simultaneous localization and mapping (SLAM) algorithms. Since LiDAR can currently provide the highest accuracy representation of the environment, generating mapping data from LiDAR odometry has observed significant interest in the literature. Furthermore, LiDAR based odometry can provide high quality mapping information in feature-rich GPS-denied areas like urban centers where ground level roads are occluded by tall buildings. This paper describes an experimental setup composed of hardware and software stacks required for realizing LiDAR based odometry generation in roadway environments. Subsequently, an open-source implementation that was reported to perform well on the widely accepted KITTI benchmarking dataset was experimentally evaluated. This experimentation was focused on the validation of LiDAR based mapping and odometry generation in a typical suburban environment. The corresponding experimental observations are presented and a number of propositions are made for further improvement.