Abstract
Robust autonomous navigation is a key feature of a mobile robot realizing services such as
transportation, cleaning, search and rescue, and surveillance. In addition to that, navigation is a
building block for a robot assisting humans in potentially dangerous situations, such as search-
and-rescue scenarios. Hence, navigation is one of the major research topics in the robotics
community.
To realize the above mentioned applications, we need to fulfill certain requirements, so that
a robot is regarded as useful. For example, a robot which performs pick-and-place tasks or
offers guidance in city centers needs to be aware of its own position in the environment and
it needs to have an accurate model of the environment for planning an appropriate path. A
robot which should guide a human to a certain place or has to deliver goods is only regarded as
helping hand, if the location is reliably reached within the expected time frame.
Particularly, estimating the state which describes the current situation of the navigation sys-
tem is complex. In this thesis, we focus on efficient and accurate state estimation techniques
which apply probabilistic algorithms. An example for such a state estimation task is the Si-
multaneous Localization and Mapping (SLAM) problem, in which a robot has to address both
aspects. First, it needs to estimate what the environment looks like. This is the mapping part
which deals with integrating the information obtained by the sensors of the robot into an ap-
propriate representation. Second, the localization component has to estimate the position of the
robot with respect to the model of the environment.
In the first part of this thesis, we present efficient approaches to estimate the state of the
robot while performing SLAM. Our approach allows a robot to accurately estimate the model
of the environment in an online setting and also in situations when provided with a poor initial
guess. Additionally, we provide an empirical evaluation which demonstrates the advantages of
our approach compared to other state-of-the-art methods. Subsequently, we extend our state
estimation approach to also include the unknown calibration parameters, which might change
during the lifetime of the robot, to incorporate prior information about the structure of the
environment, and to improve the fine-grained details of the estimated models.
In the second part of this thesis, we demonstrate two challenging applications which we
realized by building upon and extending the algorithms presented in the first part. In detail, we
discuss an approach which allows a car to autonomously park in a complex multi-level parking
garage. As second application we present a robotic pedestrian assistant which is able to navigate
in densely populated pedestrian zones.
All techniques presented in this thesis have been implemented and tested using both real-
world data collected with mobile robots and simulated data. To support our claims, we per-
formed an extensive collection of experiments, in which we compared the performance of our
approaches with the state-of-the-art. We believe that the proposed approaches will allow us in
the future to build systems that can assist humans in their homes and at their workplaces.
