TU Dortmund University
Eldorado - Ressourcen aus und für Lehre, Studium und ForschungNot a member yet
19914 research outputs found
Sort by
Guided Reinforcement Learning for dynamic control
No full textAutonomous mobile robots have the potential to transform industries by performing various tasks, such as material handling and transportation. As technology advances towards dynamic robots capable of interacting with their environments, controlling these systems becomes increasingly challenging. Learning-based control offers a promising approach, enabling robots to adapt to new tasks based on their experiences in dynamic environments. Such adaptability is crucial for success in real-world applications.
This dissertation explores Guided Reinforcement Learning (RL) methodologies aimed at addressing common challenges such as training speed, policy performance, and the transfer from simulation to reality, all of which are essential for deploying robots in real-world scenarios. The first part of this thesis introduces a novel taxonomy for Guided RL that systematically incorporates external knowledge sources, including expert demonstrations and prior experiences. This approach enhances the efficiency and effectiveness of the training process. The dissertation also includes a comprehensive survey of Guided RL methods and practical guidelines for their application in various robotics contexts, allowing researchers and practitioners to choose the most suitable strategies for their specific needs. The second part includes an extensive case study on learning-based control for evoBOT, a dynamic two-wheeled robot developed by the Fraunhofer Institute for Material Flow and Logistics IML. This study illustrates how Guided RL techniques can significantly improve learning-based control, resulting in rapid training, robustness against disturbances, and successful transfer of trained policies to the physical robot. Lastly, the thesis introduces a generalized action space for robot control, providing a formulation that operates directly at the torque level. Benchmarking evaluates the impact of different action spaces on training efficiency and policy performance across multiple robotic platforms, offering valuable insights into optimizing control strategies.
Overall, this work contributes to the advancement of robot learning, offering valuable insights and methodologies for effectively training and implementing learning-based control strategies in complex, dynamic robotic systems. Therefore, this work lays the groundwork for improving the capabilities of autonomous robots in real-world applications, promoting greater integration of robotics into everyday life. Key results of this work, including publications, the robot simulation model, and the training code, are available for open access to encourage further research in this field.Autonome mobile Roboter haben das Potenzial die Industrie zu transformieren, indem sie verschiedene Aufgaben wie Materialhandhabung und Transport übernehmen. Mit dem technologischen Fortschritt hin zu dynamischen Robotern, die mit ihrer Umgebung interagieren können, wird die Steuerung dieser Systeme immer komplexer. Lernbasierte Regelung bietet einen vielversprechenden Ansatz, der es den Robotern ermöglicht, sich auf der Grundlage ihrer Erfahrungen in dynamischen Umgebungen an neue Aufgaben anzupassen. Eine solche Anpassungsfähigkeit ist entscheidend für den Erfolg in realen Anwendungen.
Diese Dissertation bietet eine umfassende Untersuchung von Methoden des geführten Verstärkungslernens (Guided Reinforcement Learning, RL), welche darauf abzielen die Trainingsgeschwindigkeit und Performance zu verbessern, sowie die Übertragbarkeit der lernbasierten Regler von der Simulation auf die realen Roboter zu ermöglichen. Der erste Teil dieser Arbeit stellt eine neuartige Taxonomie für Guided RL vor, die systematisch externe Wissensquellen einbezieht. Dieser Ansatz steigert die Effizienz und Effektivität des Trainingsprozesses. Die Dissertation enthält auch eine umfassende Studie zu Guided RL Methoden sowie praktische Richtlinien für deren Anwendung in verschiedenen Kontexten der Robotik, die es Forschern und Anwendern ermöglichen, die am besten geeigneten Strategien für ihre individuellen Anforderungen auszuwählen. Der zweite Teil der Arbeit enthält eine umfassende Fallstudie zur lernbasierten Regelung von evoBOT, einem dynamischen zweirädrigen Roboter, der vom Fraunhofer-Institut für Materialfluss und Logistik IML entwickelt wurde. Diese Studie veranschaulicht, wie Guided-RL-Techniken die gelernten Regler erheblich verbessern können, was zu schnellerem Training, Robustheit gegenüber Störungen und der erfolgreichen Übertragung der trainierten Regler auf den physischen Roboter führt. Schließlich wird im dritten Teil dieser Arbeit ein verallgemeinerter Aktionsraum für die Robotersteuerung eingeführt, der eine Formulierung bietet, die direkt auf Drehmomentebene arbeitet. Dieser Aspekt der Arbeit bewertet die Auswirkungen verschiedener Aktionsräume auf die Performance gelernter Regler für verschiedene Roboterplattformen und bietet wertvolle Einblicke in die Optimierung von Regelungsstrategien.
Insgesamt leistet diese Arbeit einen Beitrag zur Weiterentwicklung des Roboterlernens und bietet wertvolle Einblicke und Methoden für ein effektives Training und die Implementierung lernbasierter Regelstrategien in komplexen, dynamischen Robotern. Somit legt diese Arbeit den Grundstein für die Verbesserung der Fähigkeiten autonomer Roboter in realen Anwendungen und das tägliche Leben. Zentrale Ergebnisse dieser Arbeit, einschließlich der Veröffentlichungen, des Simulationsmodells und des Trainingscodes, sind frei zugänglich, um die weitere Forschung auf diesem Gebiet zu fördern
Micro-scratch tests as a method for determining the wear resistance of high-speed blanked surfaces
No full textCompared to conventional cutting processes such as normal or fine blanking, high-speed
blanking (HSB) is an economically and ecologically relevant alternative to produce sheet
metal parts. Depending on the material and the process parameters, the microstructure is
significantly affected, and adiabatic shear bands (ASB) can form due to the localisation of
the induced deformation. The resulting HSB surfaces are characterised by high hardness,
small rollover depth and width, low roughness, and almost no burr. Therefore, HSB is a
promising strategy for producing blanked surfaces, which can be used directly as functional
surfaces without further post-processing steps. Besides the surface quality, properties like
the wear resistance are relevant for technical applications. To enable the production of
functional surfaces, which meet the requirements for the application, the understanding of
the relationships between the blanking process parameters, the resulting microstructure,
and the surface properties is essential. As standardised macroscopic wear testing methods
are often not applicable to the geometrically complex shear-cut surfaces, in this study a first
approach was taken utilising the micro-scratch test method for determining the scratch
resistance directly measured on the blanked surface
Drop dilution enables the use of PEG-derived detergents for membrane protein purification
No full textPEG detergents are important tools in the biophysical characterization of membrane protein whose utility is often limited by their intrinsic denaturing properties. This work addresses the question of whether changing the linker between PEG headgroup and nonpolar tail can modulate the denaturing properties of these detergents. To address this question, herein, we introduce the modular architecture of PEG550 detergents and explore its utility for protein purification from membranes and detergent exchange. Our results indicate that PEG550 detergents cannot efficiently solubilize proteins from lysed bacterial membranes. Varying the linker cannot eliminate the denaturing properties that PEG550 detergents can have on a protein during extraction and affinity purification. Interestingly, we find that PEG550 detergents can preserve the secondary structure and activity of the model membrane protein vitamin B12 transporter as good as the reference detergent n-dodecyl-β-D-maltoside following detergent exchange via drop dilution. Our findings clarify that denaturing properties of PEG550 detergents depend on both their chemical structure and the detergent exchange method with which proteins and detergents are brought together. Our drop dilution conditions are representative of those frequently employed in the biophysical characterization of membrane proteins. We anticipate PEG550 detergents will deliver a starting point for the optimization of sample properties in the biophysical characterization of membrane proteins
BNE als konkreter Lerninhalt im Mathematikunterricht: mit Komplexität und Unsicherheit bewusst umgehen lernen
No full textInhalte zum Thema einer nachhaltigeren Entwicklung sind insbesondere durch zwei Aspekte gekennzeichnet: Komplexität und Ungewissheit. Beides macht Nachhaltigkeitsfragen im (Mathematik-)Unterricht besonders herausfordernd, eröffnet aber auch Chancen sowohl für den Mathematikunterricht (MU) als auch für Bildung für Nachhaltige Entwicklung (BNE)
CAS-unterstützte Beweisprozesse in der Hochschulmathematik: Ein Weg vom experimentellen zum formalen Beweis
No full textZiel des Mathematikstudiums ist, Studierende in eine Disziplin einzuführen, in der formal-deduktive Beweise eine besondere Rolle spielen. Da in der Schule Aussagen oft durch Plausibilitätsüberlegungen begründet werden, haben Studierende Schwierigkeiten zum formal-deduktiven Beweisbegriff durchzudringen. In dem Vortrag werden Beispiele aufgezeigt, wie Beweisprozesse in der Hochschulmathematik mit Hilfe von CAS unterstützt werden können und wie hierbei das Modell von Wittmann und Müller (1988) genutzt werden kann, um von Plausibilitätsüberlegungen zu formal-deduktiven Beweisen zu gelangen
Assessing the Effectiveness of STEM Enrichment Programs: A Comparison between Switzerland and Germany
No full textThis paper outlines the data collection of a pilot study conducted in the summer 2024, addressing the gap in the literature regarding the impact of STEM (Science, Technology, Engineering, and Mathematics) enrichment programs with a focus on computational thinking. While existing research highlights the effectiveness of such programs on cognitive and psychological factors, much of the evidence is context-specific studies. This project compares two educational systems in Switzerland and Germany, offering insights into how enrichment programs impact students in different educational environments
Optimal design theory of dose-response experiments in toxicology
No full textNew research approaches of sciences like statistics and toxicology have been coming up through rapid development in the last years. However, those research approaches often refer to only one of the two sciences, not considering important aspects of the other discipline. Especially in the crucial part of planning an experiment, the laboratory routine in toxicology does not consider optimal design theory of statistics, although there is already much research present, which could help to improve the research results. This demonstrates a huge gap between practical applications in toxicological research and existing statistical theory. On the one hand, this gap exists due to the missing statistical methods specifically tailored to toxicological applications, and on the other hand, if those statistical methods exist, they are not reported in a clear manner for non-statisticians. The consequences for toxicological experiments are a waste of observations, or even worse animals, and non-optimal results in terms of precision. Therefore, this is an important aspect in the field of statistics and toxicology, which needs to be addressed. Optimal design approaches specifically tailored to toxicology must be developed and reported appropriately.
This cumulative thesis is based on three works, that all present optimal design approaches for diverse toxicological applications. The first manuscript highlights the importance of considering optimal design approaches for classical cytotoxicity experiments in a userfriendly manner. Here, different optimal design approaches are compared to typically used designs in practice based on an extensive case study. Moreover, a guideline for cytotoxicity experiments and an R-Shiny software tool are presented, which both facilitate the planning of upcoming cytotoxicity experiments. In the second manuscript a new design approach for the precise estimation of effective dose sets in drug combination studies is developed. For that matter, the performance of the corresponding developed criterion is investigated in a simulation study based on various scenarios including a case study. Finally, a new design approach for the analysis of high-dimensional gene-expression data is developed in the third manuscript. While two of the manuscripts are already published, the second one is attached in its current, unpublished form
Investigating patterns of language dominance and mixed dominance among Farsi-German bilingual children in Germany
No full textIntroduction: This study aimed to operationalize language dominance based on relative language proficiency across vocabulary and morphosyntax and to classify children into dominant and balanced groups. These language dominance classifications were compared with those based on relative language experience, which is characterized by two child-external factors: relative language exposure and use in Farsi. This study further explored child-internal and -external factors contributing to mixed language dominance, defined as a divergence in dominance classification across linguistic domains.
Methods: Thirty-two Farsi-German bilingual children (age range: 3.10–8.9 years, mean = 6.9 years, SD = 16.8), who speak Farsi as their heritage language (HL) in Germany, participated in the study. All children were tested on vocabulary and morphosyntax in both Farsi and German using the LITMUS-Crosslinguistic Lexical Tasks (CLTs) and Sentence Repetition Tasks (SRTs). Children's relative language experience was documented based on parental ratings.
Results: The findings indicated that the 0.5 SD-based classification is a reliable method for identifying language dominance. In contrast, relative language experience in the HL only partially predicted Farsi-dominant status when compared to German-dominant and balanced children across domains. Mixed dominance was observed in 45% of the children and was influenced by relative language use in the HL and length of exposure (LoE) to German.
Discussion: Overall, this study highlighted that using a 0.5 SD threshold provides a more consistent approach to determining relative language proficiency and that mixed dominance is a characteristic feature of bilingualism. Recognizing this feature and its contributing factors may help reduce the risk of misdiagnosing developmental language disorder (DLD) in bilingual children
Die Wahrnehmung von mathematischen Enrichment-Programmen durch Schülerinnen und Schüler der Grund- und Sekundarschule
No full textDie Notwendigkeit von Enrichment-Programmen für mathematisch interessierte und begabte Schüler:innen kann theoretisch begründet werden, wird aber kontrovers diskutiert, obwohl Erfahrungsberichte und qualitative Forschung regelmäßig positive Auswirkungen unter anderem auf die Zufriedenheit und Selbstwirksamkeit der Teilnehmenden belegen. In diesem Beitrag wird die subjektive Wahrnehmung bezogen auf Wahlfreiheit, Kompetenz und Interesse der teilnehmenden Schüler:innen zweier Enrichment-Programme im Vergleich zur Wahrnehmung derselben Aspekte bezogen auf den schulischen Unterricht dargestellt