Motion-Aware LiDAR Correction

LiDARs are widely used in autonomous naviga- tion as they provide input to several functionalities such as Localization, obstacle detection, and Simultaneous Localization and Mapping. In such applications, the most common type is the mechanical LiDAR, which in turn works by deflecting the laser beam with a rotating mirror. This technology might result in artifacts affecting the measurements when the sensor is moved during the acquisition. This leads to the well-known skewing phenomenon. In this paper, we approach the problem of deskewing the scans by estimating the sensor's motion each time a single range measurement is done; in fact, we treat with the scan as a continuous stream of measurements rather than a single rigid point cloud. We propose a variant of ICP that rather than estimating the relative position of a pair of scans, attempts to estimate the slowly changing velocities that render the stream of measurement maximally consistent.