Landmark placement for mobile robot navigation
Maximilian Beinhofer · FreiDok plus (Universitätsbibliothek Freiburg) · 2014
Being able to navigate accurately is one of the fundamental capabilities of a mobile robot to effectively execute a variety of tasks including docking, transportation, and manipulation. To achieve the desired navigation accuracy, mobile robots are typically equipped with on-board sensors to observe persistent features in the environment, to estimate their pose from these observations, and to adjust their motion accordingly. Since real-world environments often contain changing or ambiguous areas, existing features can be insufficient for mobile robots to establish a robust navigation behavior. A popular approach to overcome this problem and to enable accurate localization is to use artificial landmarks like reflective markers or barcodes. However, depending on the type of artificial landmark, the landmarks themselves or their precise placement can be costly. Therefore, it is desirable to place as few landmarks as possible to achieve the desired accuracy, which makes selecting beneficial landmark positions especially important. In industrial settings, for example, mobile robots often have to perform repetitive tasks that include travelling along the same trajectory through a known environment. For such scenarios, we present approaches that aim at finding a minimum set of landmark positions in order to optimize the expected quality of the robot's task execution. We measure this quality either by the expected accuracy of the localization estimate of the robot or -- if the robot bases its navigation decisions directly on its localization estimate -- by the expected accuracy of the navigation behavior of the robot. In order to efficiently estimate the expected accuracy in navigation, we introduce a novel recursive calculation scheme for the expected distributions of the robot's deviation from its desired trajectory. For dealing with the generally NP-hard landmark placement problem, we use techniques from submodular function optimization to efficiently generate near-optimal landmark configurations. The resulting efficiency of our landmark placement approaches makes it possible to apply them even to large-scale scenarios. In contrast to the above-mentioned methods, landmark placement for robots travelling through unknown and unmapped environments requires different approaches. If the robot has a device to deploy a limited number of artificial landmarks itself, it will later be able to use them as fixed anchors to adjust the estimate of its relative motions between individual observations of the same landmark. We present a novel approach for learning an optimal landmark deployment policy for this scenario. We evaluated all presented methods in extensive experiments both in simulation and with real mobile robots. The experiments demonstrate that our approaches outperform baseline methods and work well on real robots. We believe that the presented landmark placement methods are a useful tool for guaranteeing a safe and reliable operation of mobile robots in practice, especially in industrial settings.