Context-Dependent Scan Matching for Aided Navigation

For robust and autonomous navigation, many different sensors have been incorporated and indeed fused to form a navigation suite in autonomous vehicles to perform tasks such as vehicular guidance, path generation, and simultaneous localization and mapping. The paper investigates the use of laser ranger data for aided navigation. The objective is to seek a fair assessment of the laser ranger data, from the perspective of navigation, so that the data can be blended into the navigation system. At any measurement epoch, a laser scanner provides a scan map in an array format that contains the angles and the corresponding ranges. Typically, a scan matching problem is solved to find the relationship between two scan maps, leading to the increments in translation and rotation. A known error source in scan matching is the correspondence error in associating measurement data at two different epochs. As the points being scanned at two epochs may not be identical, the displacement between the two scanned points thus induces error in map building and localization. Further, the context of the environment in terms of the richness of features may also affect the quality of the resulting navigation solution. In the paper, a quality index based on the analysis of intra-frame point clouds is proposed to assess the context to provide a way for quantify the scan matching results. In addition to the increments in translation and rotation, the corresponding quality indices are obtained. As a result, the data can then be better accounted in the navigation suite as a robust aided navigation data is provided. Simulation and experimental results will be used to illustrate the proposed approach.