Accurate 3D localization of legged robots using NanoVDB and High-speed 3D laser scanning system
Peña Echeverría, Andrea Maybell (2023)
2023
Master's Programme in Automation Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-11-02
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202310098705
https://urn.fi/URN:NBN:fi:tuni-202310098705
Tiivistelmä
In the field of mobile robotics, legged robots are distinguished by their capability to traverse challenging terrain and ascend stairs. The Boston Dynamics' Spot robot exemplifies this category. Its navigation is based on its "auto-walk" feature, which allows it to replicate manually pre-recorded routes. However, for truly autonomous operation not restricted to predefined routes, 3D map-based localization is essential. Therefore, this research proposes a pipeline to perform 3D map-based localization with the Spot robot.
Initially, to identify the most effective means of localization, the performance of multiple scan matching techniques and libraries were compared in terms of accuracy and registration time. The evaluation concluded that the ICP method implemented through the Libpointmatcher library, registered the most precise point cloud registration results. In contrast, the NDT method executed with FastGICP appeared as the most time-efficient.
Processing further, a localization pipeline was built with the map of the environment obtained with a high-speed 3D LiDAR, TrimbleX7. The study proved the efficiency of a ray tracing technique, developed using the NanoVDB library, in extracting a local point cloud from the map. When combined with scan matching techniques, this approach significantly improves the localization accuracy. In its optimal configuration, utilizing the NDT scan matching method from the FastGICP library, the method achieved a localization precision of 0.148 meters.
Initially, to identify the most effective means of localization, the performance of multiple scan matching techniques and libraries were compared in terms of accuracy and registration time. The evaluation concluded that the ICP method implemented through the Libpointmatcher library, registered the most precise point cloud registration results. In contrast, the NDT method executed with FastGICP appeared as the most time-efficient.
Processing further, a localization pipeline was built with the map of the environment obtained with a high-speed 3D LiDAR, TrimbleX7. The study proved the efficiency of a ray tracing technique, developed using the NanoVDB library, in extracting a local point cloud from the map. When combined with scan matching techniques, this approach significantly improves the localization accuracy. In its optimal configuration, utilizing the NDT scan matching method from the FastGICP library, the method achieved a localization precision of 0.148 meters.