Design of a Hybrid-Range-LiDAR SLAM algorithm
a graph-based approach to SLAM using ultra-wideband radios
More Info
expand_more
Abstract
A fundamental prerequisite for many robot-tasks is the availability of an environment model. The ability of a robot to create such a model itself is crucial to having truly autonomous robots around in our daily lives. For basic robot-tasks, a two-dimensional grid-map that describes a distinction between inhabitable and uninhabitable space is a commonly used model. Creating grid-maps, or any environment model, is usually done by simultaneously estimating the robots location and its environment.
Modern SLAM algorithms are based on incremental non-linear optimization and the robustness highly depends on the used sensors and their proneness to environmental ambiguities. In this thesis an affordable laser-range-finder (often referred to as LiDAR is used in combination with ultra-wideband radios to improve the localization and mapping performance in large ambiguous environments.
In SLAM-literature it is common that LiDAR measurements are used to both correct the robot's ego-motion and to build a grid-map. In this thesis, however, the LiDAR is used solely for creating a grid-map while the ego-motion is corrected using UWB range-measurements. To quantitatively assess the mapping performance using this strategy, a HRL-SLAM algorithm was designed and implemented as a real-time application. The application is benchmarked on both synthetic- and real data and compared to similar range-based SLAM implementations.
The result of this thesis indicates that UWB radios can dramatically improve the robustness of indoor localization and grid-mapping with affordable LiDARs. Secondary results are a scalability enhancement for an existing beacon-localization algorithm, a graph-topology inspired submapping-strategy and the introduction of a reference-frame independent benchmark-metric for range-SLAM.
Finally, a few recommendations are done for improving the HRL-SLAM algorithm. The thesis concludes with an outlook on how to proceed from SLAM's graph-representation to a hierarchical metric-topological set of spatial representations, this can be seen as a tertiary result.