Schirmer, Robert Alexander: Pose Uncertainty Aware Mobile Robot Navigation. - Bonn, 2026. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-90956
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-90956
@phdthesis{handle:20.500.11811/14252,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-90956,
doi: https://doi.org/10.48565/bonndoc-897,
author = {{Robert Alexander Schirmer}},
title = {Pose Uncertainty Aware Mobile Robot Navigation},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2026,
month = jul,
note = {In recent years, robotics has emerged as a transformative force, changing how we approach everyday activities to improve our comfort and productivity. Robotics technologies continuously push the boundaries of what can be solved autonomously, thereby improving our lives. Service robots such as autonomous vacuum cleaners and lawn mowers are well established in the market, while the automation of factory logistics and automated driving are up-and-coming areas with significant robotics impact. To be a successful product, a robot must solve a real-world problem autonomously and efficiently. For mobile robots that primarily interact with the world by moving, the main task is often navigation, which encompasses the technologies required to drive robustly and solve a problem for the user. Robot navigation in real-world applications is challenging because robustness and performance requirements conflict with the cost-minimization imperative. Thus, to perform a real service to the user, a robot must perform its task in potentially adverse conditions without compromising safety or requiring constant human intervention.
In this thesis, we address the problem of robot navigation in real-world scenarios motivated by the Bosch Indego autonomous lawn mower. Our goal is to enable robots with cheaper sensor sets to navigate effectively in challenging garden environments despite significant sensing and actuation noise. These uncertainties are particularly pronounced in the lawn-mowing scenario due to uneven terrain, featureless areas, and the need for the robot to operate cost-effectively. We handle those uncertainties in three fundamental building blocks of robot navigation. First, we consider the localization problem. Localization is particularly important for mobile robots, as it allows the use of a single coordinate frame to represent the robot's position and the environment, enabling the robot to move with a purpose. Our localization method generates accurate pose estimates in diverse environments recorded with different sensors, while requiring only little compute. Nevertheless, robots with cheap sensors operating in complex conditions, such as autonomous lawn mowers, cannot always precisely know where they are. Their ability to remain well-localized varies significantly across different regions of their workspace. Accounting for this variation in localization quality directly motivates our second contribution, which addresses how robots should plan their motion in such situations to ensure safety and robustness. Our method estimates which areas of the workspace are particularly informative or feature-poor, and exploits this information during the path planning process. Our point-to-point path planning method is fast, computationally lightweight, and efficiently accounts for the uncertainty in sensing and actuation. Beyond point-to-point path planning, we also introduce a method that leverages the localizability information for coverage path planning. This problem is central to autonomous lawn mowers as it computes a trajectory to cover the entire workspace with the end effector. Our approach leverages the expected localization accuracy in the environment to plan paths where the robot can localize well, thus improving the robustness of the coverage path and overall system performance.
Taken together, the contributions outlined in this thesis provide novel solutions to localization and planning under uncertainty. All parts of this thesis have been published in peer-reviewed proceedings of international conferences. We have implemented our approaches in efficient C++ and deployed them on robots.},
url = {https://hdl.handle.net/20.500.11811/14252}
}
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-90956,
doi: https://doi.org/10.48565/bonndoc-897,
author = {{Robert Alexander Schirmer}},
title = {Pose Uncertainty Aware Mobile Robot Navigation},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2026,
month = jul,
note = {In recent years, robotics has emerged as a transformative force, changing how we approach everyday activities to improve our comfort and productivity. Robotics technologies continuously push the boundaries of what can be solved autonomously, thereby improving our lives. Service robots such as autonomous vacuum cleaners and lawn mowers are well established in the market, while the automation of factory logistics and automated driving are up-and-coming areas with significant robotics impact. To be a successful product, a robot must solve a real-world problem autonomously and efficiently. For mobile robots that primarily interact with the world by moving, the main task is often navigation, which encompasses the technologies required to drive robustly and solve a problem for the user. Robot navigation in real-world applications is challenging because robustness and performance requirements conflict with the cost-minimization imperative. Thus, to perform a real service to the user, a robot must perform its task in potentially adverse conditions without compromising safety or requiring constant human intervention.
In this thesis, we address the problem of robot navigation in real-world scenarios motivated by the Bosch Indego autonomous lawn mower. Our goal is to enable robots with cheaper sensor sets to navigate effectively in challenging garden environments despite significant sensing and actuation noise. These uncertainties are particularly pronounced in the lawn-mowing scenario due to uneven terrain, featureless areas, and the need for the robot to operate cost-effectively. We handle those uncertainties in three fundamental building blocks of robot navigation. First, we consider the localization problem. Localization is particularly important for mobile robots, as it allows the use of a single coordinate frame to represent the robot's position and the environment, enabling the robot to move with a purpose. Our localization method generates accurate pose estimates in diverse environments recorded with different sensors, while requiring only little compute. Nevertheless, robots with cheap sensors operating in complex conditions, such as autonomous lawn mowers, cannot always precisely know where they are. Their ability to remain well-localized varies significantly across different regions of their workspace. Accounting for this variation in localization quality directly motivates our second contribution, which addresses how robots should plan their motion in such situations to ensure safety and robustness. Our method estimates which areas of the workspace are particularly informative or feature-poor, and exploits this information during the path planning process. Our point-to-point path planning method is fast, computationally lightweight, and efficiently accounts for the uncertainty in sensing and actuation. Beyond point-to-point path planning, we also introduce a method that leverages the localizability information for coverage path planning. This problem is central to autonomous lawn mowers as it computes a trajectory to cover the entire workspace with the end effector. Our approach leverages the expected localization accuracy in the environment to plan paths where the robot can localize well, thus improving the robustness of the coverage path and overall system performance.
Taken together, the contributions outlined in this thesis provide novel solutions to localization and planning under uncertainty. All parts of this thesis have been published in peer-reviewed proceedings of international conferences. We have implemented our approaches in efficient C++ and deployed them on robots.},
url = {https://hdl.handle.net/20.500.11811/14252}
}





