Der vorliegende Beitrag stellt einen Entwurf für eine flexible und robuste Steuerungsarchitektur für Roboter- und Fertigungssysteme vor. Dabei wurde versucht ein offenes Konzept zu realisieren, welches einen vereinfachten Engineeringprozess ermöglicht. Hierzu wird innerhalb der Steuerung eine Trennung zwischen einem funktionellen verhaltensbasierten und einem ablauforientierten Modell vorgeschlagen. Dieser Ansatz wird durch die Verwendung von Aktionsprimitiven innerhalb einer hybriden Robotersteuerung ermöglicht. Diese garantieren durch ihre ausgeprägte Modularität eine hohe Flexibilität und Erweiterbarkeit des entstandenen Systems. Im Beitrag wird sowohl der entstandene Entwurf diskutiert als auch eine prototypische objektorientierte Implementierung vorgestellt sowie erste Ergebnisse präsentiert. / This paper presents a framework for a flexible and robust control architecture for robotic systems. The design incorporates an application independent system concept which allows a simplified engineering process. For this purpose a distinction between a functional behavioural and a sequential control system model is proposed. This approach is based on the utilisation of action primitives within a hybrid control architecture. The use of these primitives affords a high level of modularity through increasing flexibility and expandability of the resulting system. In this paper the proposed framework will be discussed as well as a prototypical object-oriented implementation and first results.

Der vorliegende Beitrag stellt einen Entwurf für eine flexible und robuste Steuerungsarchitektur für Roboter- und Fertigungssysteme vor. Dabei wurde versucht ein offenes Konzept zu realisieren, welches einen vereinfachten Engineeringprozess ermöglicht. Hierzu wird innerhalb der Steuerung eine Trennung zwischen einem funktionellen verhaltensbasierten und einem ablauforientierten Modell vorgeschlagen. Dieser Ansatz wird durch die Verwendung von Aktionsprimitiven innerhalb einer hybriden Robotersteuerung ermöglicht. Diese garantieren durch ihre ausgeprägte Modularität eine hohe Flexibilität und Erweiterbarkeit des entstandenen Systems. Im Beitrag wird sowohl der entstandene Entwurf diskutiert als auch eine prototypische objektorientierte Implementierung vorgestellt sowie erste Ergebnisse präsentiert. / This paper presents a framework for a flexible and robust control architecture for robotic systems. The design incorporates an application independent system concept which allows a simplified engineering process. For this purpose a distinction between a functional behavioural and a sequential control system model is proposed. This approach is based on the utilisation of action primitives within a hybrid control architecture. The use of these primitives affords a high level of modularity through increasing flexibility and expandability of the resulting system. In this paper the proposed framework will be discussed as well as a prototypical object-oriented implementation and first results.

The navigation tasks in advanced home robotic applications incorporating reliable revisiting strategies are dependent on very low cost but nevertheless rather accurate localization systems. In this paper a localization system based on the principle of trilateration is described. The proposed system uses only a single small base station, but achieves accuracies comparable to systems using spread beacons and it performs sufficiently for map building. Thus it is a standalone system and needs no odometry or other auxiliary sensors. Furthermore a new approach for the problem of the reliably detection of areas without direct line of sight is presented. The described system is very low cost and it is designed for use in indoor service robotics. The paper gives an overview on the system concept and special design solutions and proves the possible performances with experimental results.

This thesis offers a probabilistic solution to robust estimation using a novel adaptive estimator. Reliable state estimation is a mandatory prerequisite for autonomous systems interacting with the real world. The presence of outliers challenges the Gaussian assumption of numerous estimation algorithms, resulting in a potentially skewed estimate that compromises reliability. Many approaches attempt to mitigate erroneous measurements by using a robust loss function – which often comes with a trade-off between robustness and numerical stability. The proposed approach is purely probabilistic and enables adaptive large-scale estimation with non-Gaussian error models. The introduced Adaptive Mixture algorithm combines a nonlinear least squares backend with Gaussian mixtures as the measurement error model. Factor graphs as graphical representations allow an efficient and flexible application to real-world problems, such as simultaneous localization and mapping or satellite navigation. The proposed algorithms are constructed using an approximate expectation-maximization approach, which justifies their design probabilistically. This expectation-maximization is further generalized to enable adaptive estimation with arbitrary probabilistic models. Evaluating the proposed Adaptive Mixture algorithm in simulated and real-world scenarios demonstrates its versatility and robustness. A synthetic range-based localization shows that it provides reliable estimation results, even under extreme outlier ratios. Real-world satellite navigation experiments prove its robustness in harsh urban environments. The evaluation on indoor simultaneous localization and mapping datasets extends these results to typical robotic use cases. The proposed adaptive estimator provides robust and reliable estimation under various instances of non-Gaussian measurement errors.

In recent years semantic segmentation models utilizing Convolutional Neural Networks (CNN) have seen significant success for multiple different segmentation problems. Models such as U-Net have produced promising results within the medical field for both regular 2D and volumetric imaging, rivalling some of the best classical segmentation methods. In this thesis we examined the possibility of using a convolutional neural network-based model to perform segmentation of discrete bone fragments in CT-volumes with segmentation-hints provided by a user. We additionally examined different classical segmentation methods used in a post-processing refinement stage and their effect on the segmentation quality. We compared the performance of our model to similar approaches and provided insight into how the interactive aspect of the model affected the quality of the result. We found that the combined approach of interactive segmentation and deep learning produced results on par with some of the best methods presented, provided there were adequate amount of annotated training data. We additionally found that the number of segmentation hints provided to the model by the user significantly affected the quality of the result, with convergence of the result around 8 provided hints.

In large scale productions of metal sheets, it is important to maintain an effective way to continuously inspect the products passing through the production line. The inspection mainly consists of detection of defects and tracking of ID numbers. This thesis investigates the possibilities to create an automatic inspection system by evaluating different machine learning algorithms for defect detection and optical character recognition (OCR) on metal sheet data. Digit recognition and defect detection are solved separately, where the former compares the object detection algorithm Faster R-CNN and the classical machine learning algorithm NCGF, and the latter is based on unsupervised learning using a convolutional autoencoder (CAE). The advantage of the feature extraction method is that it only needs a couple of samples to be able to classify new digits, which is desirable in this case due to the lack of training data. Faster R-CNN, on the other hand, needs much more training data to solve the same problem. NCGF does however fail to classify noisy images and images of metal sheets containing an alloy, while Faster R-CNN seems to be a more promising solution with a final mean average precision of 98.59%. The CAE approach for defect detection showed promising result. The algorithm learned how to only reconstruct images without defects, resulting in reconstruction errors whenever a defect appears. The errors are initially classified using a basic thresholding approach, resulting in a 98.9% accuracy. However, this classifier requires supervised learning, which is why the clustering algorithm Gaussian mixture model (GMM) is investigated as well. The result shows that it should be possible to use GMM, but that it requires a lot of GPU resources to use it in an end-to-end solution with a CAE.

Semantic segmentation is a key approach to comprehensive image data analysis. It can be applied to analyze 2D images, videos, and even point clouds that contain 3D data points. On the first two problems, CNNs have achieved remarkable progress, but on point cloud segmentation, the results are less satisfactory due to challenges such as limited memory resource and difficulties in 3D point annotation. One of the research studies carried out by the Computer Vision Lab at Linköping University was aiming to ease the semantic segmentation of 3D point cloud. The idea is that by first projecting 3D data points to 2D space and then focusing only on the analysis of 2D images, we can reduce the overall workload for the segmentation process as well as exploit the existing well-developed 2D semantic segmentation techniques. In order to improve the performance of CNNs for 2D semantic segmentation, the study has used input data derived from different modalities. However, how different modalities can be optimally fused is still an open question. Based on the above-mentioned study, this thesis aims to improve the multistream framework architecture. More concretely, we investigate how different singlestream architectures impact the multistream framework with a given fusion method, and how different fusion methods contribute to the overall performance of a given multistream framework. As a result, our proposed fusion architecture outperformed all the investigated traditional fusion methods. Along with the best singlestream candidate and few additional training techniques, our final proposed multistream framework obtained a relative gain of 7.3\% mIoU compared to the baseline on the semantic3D point cloud test set, increasing the ranking from 12th to 5th position on the benchmark leaderboard.

Die mobile Robotik stellt ein sehr junges und komplexes Forschungsfelder unserer Zeit dar. Innerhalb der letzten Jahrzehnte wurde es Robotern möglich, sich innerhalb ihrer Umgebung zu bewegen, zu navigieren und mit ihrer Umwelt zu interagieren. Aufgrund der Tatsache, dass die Welt von Unsicherheit geprägt ist und ein Roboter immer nur partielle Information über sie erhalten kann, wurden probabilistische Navigationsverfahren entwickelt, mit denen sich Roboter lokalisieren und Objekte ihrer Umgebung modellieren können. Weiterhin wurden in letzter Zeit Verfahren untersucht, die die kooperative Exploration der Umgebung durch eine Gruppe von Robotern zum Ziel haben. In der vorliegenden Arbeit wird ein neuartiges Konzept, welches sich Perzeptrelationen für die kooperative Umweltmodellierung zu Nutze macht, vorgestellt und evaluiert. Einen zweiten Beitrag der Arbeit stellen constraintbasierte Lokalisierungstechniken dar, die es einem oder mehreren Robotern auf effiziente Art und Weise ermöglichen, sich zu lokalisieren und ihre Umwelt zu modellieren. / Mobile autonomous robotics is a very young and complex field of research. Only in recent decades have robots become able to explore, to move, navigate and to interact with their environment. Since the world is uncertain and since robots can only gain partial information about it, probabilistic navigation algorithms have become very popular whenever a robot has to localize itself or surrounding objects. Furthermore, cooperative exploration and localization approaches have become very relevant lately, as robots begin to act not just alone but in groups. Within this thesis a new approach using the concept of spatial percept-relations for cooperative environment modeling is presented and evaluated. As a second contribution, constraint based localization techniques will be introduced for having a robot or a group of robots efficiently localized and to model their environment.

"Die Bedeutung eines Wortes ist sein Gebrauch in der Sprache". Ludwig Wittgenstein führte diese Idee in der ersten Hälfte des 20. Jahrhunderts in die Philosophie ein und in verwandten Disziplinen wie der Psychologie und Linguistik setzte sich vor allem in den letzten Jahrzehnten die Ansicht durch, dass natürliche Sprache ein dynamisches System arbiträrer und kulturell gelernter Konventionen ist. Forscher um Luc Steels übertrugen diesen Sprachbegriff seit Ende der 90er Jahre auf das Gebiet der Künstlichen Intelligenz, indem sie zunächst Software-Agenten und später Robotern mittels sogenannter Sprachspiele gemeinsame Kommunikationssysteme bilden liessen, ohne dass Agenten im Voraus mit linguistischem und konzeptionellen Wissen ausgestattet werden. Die vorliegende Arbeit knüpft an diese Forschung an und untersucht vertiefend die Selbstorganisation von geteiltem lexikalischen Wissen in humanoiden Robotern. Zentral ist dabei das Konzept der "referential uncertainty", d.h. die Schwierigkeit, die Bedeutung eines bisher unbekannten Wortes aus dem Kontext zu erschliessen. Ausgehend von sehr einfachen Modellen der Lexikonbildung untersucht die Arbeit zunächst in einer simulierten Umgebung und später mit physikalischen Robotern systematisch, wie zunehmende Komplexität kommunikativer Interaktionen komplexere Lernmodelle und Repräsentationen erfordert. Ein Ergebnis der Evaluierung der Modelle hinsichtlich Robustheit und Übertragbarkeit auf Interaktionszenarien mit Robotern ist, dass die in der Literatur vorwiegenden selektionistischen Ansätze schlecht skalieren und mit der zusätzlichen Herausforderung einer Verankerung in visuellen Perzeptionen echter Roboter nicht zurecht kommen. Davon ausgehend wird ein alternatives Modell vorgestellt. / "The meaning of a word is its use in the language". In the first half of the 20th century Ludwig Wittgenstein introduced this idea into philosophy and especially in the last few decades, related disciplines such as psychology and linguistics started embracing the view that that natural language is a dynamic system of arbitrary and culturally learnt conventions. From the end of the nineties on, researchers around Luc Steels transferred this notion of communication to the field of artificial intelligence by letting software agents and later robots play so-called language games in order to self-organize communication systems without requiring prior linguistic or conceptual knowledge. Continuing and advancing that research, the work presented in this thesis investigates lexicon formation in humanoid robots, i.e. the emergence of shared lexical knowledge in populations of robotic agents. Central to this is the concept of referential uncertainty, which is the difficulty of guessing a previously unknown word from the context. First in a simulated environments and later with physical robots, this work starts from very simple lexicon formation models and then systematically analyzes how an increasing complexity in communicative interactions leads to an increasing complexity of representations and learning mechanisms. We evaluate lexicon formation models with respect to their robustness, scaling and their applicability to robotic interaction scenarios and one result of this work is that the predominating approaches in the literature do not scale well and are not able to cope with the challenges stemming from grounding words in the real-world perceptions of physical robots. In order to overcome these limitations, we present an alternative lexicon formation model and evaluate its performance.

In this thesis, three different neural network architectures are investigated to detect the action of a shot within a football game using video data. The first architecture uses con- ventional convolution and pooling layers as feature extraction. It acts as a baseline and gives insight into the challenges faced during shot detection. The second architecture uses a pre-trained feature extractor. The last architecture uses three-dimensional convolution. All these networks are trained using short video clips extracted from football game video streams. Apart from investigating network architectures, different sampling methods are evaluated as well. This thesis shows that amongst the three evaluated methods, the ap- proach using MobileNetV2 as a feature extractor works best. However, when applying the networks to a video stream there are a multitude of challenges, such as false positives and incorrect annotations that inhibit the potential of detecting shots.