Droeschel, David Marcel: Efficient Methods for Lidar-based Mapping and Localization. - Bonn, 2020. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-59839
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-59839,
author = {{David Marcel Droeschel}},
title = {Efficient Methods for Lidar-based Mapping and Localization},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2020,
month = oct,

note = {An expedient aim in robotics research is to enable robot systems to enter areas which are inaccessible or too dangerous to humans, such as disaster scenarios. For autonomous navigation in these environments, robust and reliable perception is key. One fundamental perception problem is to build maps of unknown environments and to localize within them simultaneously.
This thesis presents an approach to simultaneous localization and mapping (SLAM) using local multiresolution. Local multiresolution corresponds well to the measurement density and accuracy of most range sensors and allows for efficient and concise map representations. The proposed map data structure is designed for memory-efficient aggregation of measurements and enables online mapping and localization. To align acquired sensor data, a probabilistic registration method is proposed, exploiting the properties of the map data structure.
Local multiresolution maps from different view poses are aligned with each other to create an allocentric map of the environment. Optimization of the view poses yields a globally consistent dense 3D map of the environment. Continuous registration of local maps with the global map allows for tracking the robot pose in real time.
Furthermore, a method for reassessing previously aggregated measurements, to account for registration errors due to missing information or erroneous sensor data is proposed. In order to incorporate corrections when refining the alignment, the individual sensor poses of the measurements in the local map are modeled as a sub-graph in a hierarchical graph structure. Sensor poses in the sub-graphs are optimized to account for drift and misalignments in the local maps. Each sub-graph maintains a continuous-time representation of the sensor trajectory to interpolate measurements between discrete sensor poses.
The proposed methods are evaluated on different datasets and compared to state-of-the-art methods, with the results indicating superior accuracy and efficiency. Particular applications demonstrate that they have been successfully employed on different robotic platforms, such as micro aerial vehicles and ground robots, in different research projects and robot competitions.},

url = {http://hdl.handle.net/20.500.11811/8738}

The following license files are associated with this item: