13979 research outputs found
Sort by
On Uncertainty-Aware Perception, Prediction, and Planning
No full textUrban environments are inherently dynamic and multifaceted, creating conditions of complexity that challenge traditional approaches to sensing, forecasting, and decision-making. In response, artificial intelligence (AI) is increasingly used to power urban systems, enabling them to interpret sensor data, anticipate future states, and plan appropriate actions. A common structure for such AI-driven processes is the Perception–Prediction–Planning (PPP) pipeline, which decomposes complex urban tasks into sequential stages. However, despite its growing role, AI, particularly in the form of deep learning, remains fundamentally limited in its ability to handle uncertainty. Deep learning models often rely on incomplete data, exhibit miscalibrated confidence, and lack mechanisms to enforce logical consistency, allowing uncertainty to accumulate and propagate through the pipeline. Inadequate uncertainty handling, spanning modeling, quantification, and reduction, can result in overconfident yet flawed predictions. Existing uncertainty estimation methods are primarily statistical and frequently overlook the structured relationships and domain constraints that characterize urban environments. Incorporating explicit reasoning mechanisms, such as logical rules or semantic dependencies, refines uncertainty estimates, constrains outputs to plausible interpretations, and ensures coherence across tasks.
This dissertation proposes approaches that integrate domain knowledge and symbolic logic to handle uncertainty at each stage of the PPP pipeline, thereby yielding outputs that are calibrated in their confidence and verifiable in their semantics. The contributions are outlined in Chapter 1.4, with detailed summaries of research approaches and key findings presented in Chapters 4 to 7, while the full publications are provided in Chapters P1 to P7.
First, we introduce Knowledge-Refined Prediction Sets (KRPS), which embed domain-aware logical constraints into conformal prediction for multi-task perception, ensuring semantically consistent and probabilistically rigorous outputs. Experimental results on the ROAD and Waymo/ROAD++ datasets demonstrate that KRPS outperforms state-of-the-art conformal prediction methods by reducing uncertainty by up to 80% and boosting semantic consistency by up to 30%, all while retaining formal coverage guarantees.
Second, we propose NeSyMoF (Neuro-Symbolic Model for Motion Forecasting), a deep learning model, which fuses deep learning with symbolic logic to generate multi-modal trajectory forecasts along with their symbolic rules. Evaluations on Argoverse reveal that NeSyMoF not only outperforms state-of-the-art interpretable baselines for single-mode predictions but also provides first-order logic explanations for its forecasts, articulating the reasoning behind each predicted path. We develop an approach to quantify the uncertainty of NeSyMoF and devise a novel active learning strategy leveraging the discrepancies between the learned rules and the generated paths, which prioritizes high-uncertainty scenarios during training. With our active learning strategy, we reach 99% of the model performance with only 60% of the data.
Third, we present SafePath, a path planning approach that combines Large Language Models, conformal prediction for uncertainty handling, and domain-specific safety constraints. We prove theoretically that SafePath limits collision risk to a user-defined probability. Experiments on the nuScenes dataset and the Highway-env simulator show a 77% reduction in planning uncertainty, a 95% decrease in user interventions, and up to a 70% reduction in collision rates compared to state-of-the-art approaches.
Finally, we conceptualize and evaluate novel approaches for AI systems to handle uncertainty through interactions with their external environment, with a particular focus on human stakeholders. We investigate this across three key aspects: knowledge transfer and data annotation, where we introduce PointCloudLab, an immersive platform that enhances labeling efficiency and accuracy; communication of reasoning uncertainty, where we develop AR-CP, an augmented reality-based system for conveying uncertainty in perception models; and communication of decision uncertainty, where we explore AR-based representations of planned trajectory uncertainty to improve human trust and decision-making in autonomous vehicles
Enhancing Life Satisfaction through Eudaimonic, Hedonic, and Combined Interventions: New Training Approaches Relevant to Theory and Practice
No full textIn recent scientific debates, eudaimonia and hedonia have been discussed as either complementary or opposing pathways to well-being. If they are opposites, a combination of the two would not have a positive effect. If they are complementary, their combination is of particular interest. Research to date has often been based on correlational designs that do not allow any conclusions to be drawn about causality. Therefore, we used randomized control designs not only to demonstrate the effectiveness of interventions for eudaimoina and hedonia but also to see whether or not a combination of hedonia and eudaimonia will lead to life satisfaction (full-life effectivity) or even outperforms single-component interventions (full-life superiority). Two randomized controlled studies were conducted with pre-, post- and follow-up measurements. In Study 1 (N = 265), we compared four groups: hedonia training, eudaimonia training, combined training and a control group. In Study 2 (N = 76), we compared three groups: eudaimonia training, combined training and a control group. Results showed positive effects on life satisfaction in the eudaimonia and hedonia groups. The combined training worked (full-life effectivity), although not more so than the single-component trainings (no full-life superiority). The expected mediating role of the art-of-living (a set of individual behavioral strategies) for training effects on life satisfaction was also supported. Results are discussed with reference to the synergetic change model, which offers further ideas to improve combined trainings
Shape Function-Based Strain Determination in DIC for Solids and Lattice Structures
No full textBackground: Additive Manufacturing offers the opportunity to build lattice structures with benefits in manufacturing efficiency and weight. For the determination of the fatigue properties of lattice structures, it lacks a method to determine the deformation under mechanic stress.
Objective: A digital image correlation (DIC) algorithm was implemented. The algorithm determines strains within a subset in an uncommon way by physically interpreting the subset shape function and does not need neighboring subsets, therefore.
Method: With a monochrome background this shape function-based strain determination is able to determine the deformation of a whole lattice unit cell, even if the background is visible in sectors of the subset. The implementation is validated by comparing the results in quasi-static tests on bulk material specimens to the results tactile sensors and a conventional DIC program. Then the deformation of lattice unit cells in fatigue tests is evaluated.
Results: The shape function-based strain determination performs well in quasi-static tests even for large deformations. The deformation of lattice unit cells is determined successfully, whereby conventional DIC algorithms can be challenged if the lattice’s strut diameter becomes close to the image resolution. The determined strains are appropriate for lifetime prediction and fractures can be detected.
Conclusion: The shape function-based strain determination is a suitable tool for determination of large local strains as well as strains in lattice structures, which do partially not cover the background in the whole region of interest due to periodic empty spaces between the lattice struts. For determination of strain fields, conventional DIC algorithms will still be more efficient in this state of development
Pollution affects Arabian and Saharan dust optical properties in the eastern Mediterranean
No full textUncertainties in the direct radiative effect of mineral dust result from the variability in its optical properties but are also influenced by mixing with anthropogenic aerosols ("pollution"), e.g., black carbon or sulfates. This study investigates the effect of mixing pollution with mineral dust from different source regions on the intensive optical properties. The Ångström exponents of scattering and absorption, the single-scattering albedo, and the asymmetry parameter are determined from in situ measurements during the A-LIFE aircraft field experiment over the eastern Mediterranean, where Arabian and Saharan dust mixed with pollution. Our results show that all intensive dust optical properties change significantly with increasing pollution content, while differences between Arabian and Saharan dust are not statistically significant. We discuss the implications of these results for the identification of mineral dust events and for their direct radiative effects. The pollution masks the mineral dust signal, which calls for caution when using Ångström exponents to identify mineral dust events. Furthermore, the asymmetry parameter and single-scattering albedo change from pure to polluted mineral dust layers (e.g., at 525 nm, median values decrease from 0.67 to 0.56 and from 0.96 to 0.89, respectively). These changes have opposing effects on the shortwave direct radiative effect efficiency (i.e., the direct radiative effect per unit of aerosol optical depth), potentially canceling each other out. Still, the effect can be significant depending on surface albedo. In conclusion, quantifying pollution content in mineral dust layers is essential for accurately assessing their local direct radiative effect
Der Architekturprofessor Heinrich Walbe (1865 - 1954)
No full textVor wenigen Wochen jährte sich zum 150. Mal der Geburtstag des früheren Architekturprofessors und Denkmalpflegers Heinrich Rudolf Walbe. Walbe wurde im Alter von 37 Jahren im Herbst 1902 als Nachfolger von Erwin Marx (1841 – 1901) auf die Professur Baukunst IV an die Architekturabteilung der TH Darmstadt berufen. Bis zu
seinem Ausscheiden im Herbst 1933 hatte er diese Professur inne
Ottilie Rady (1890 - 1987) - mit Willen und Beharrlichkeit zum Ziel
No full textVor 125 Jahren, am 13. April 1890, wurde Ottilie Rady in Darmstadt geboren. Als sie mit 39 Jahren, am 22. Juni 1929, im Fach Kunstgeschichte an der Technischen Hochschule (TH) Darmstadt habilitiert wurde, schrieb sie Geschichte, war sie doch die erste habilitierte Kunsthistorikerin in Deutschland
Wolfgang Finkelnburg: eine Physikerkarriere im 20. Jahrhundert
No full textDer Physiker Wolfgang Finkelnburg (1905 – 1967) hätte in diesem Jahr seinen 110. Geburtstag gefeiert. Finkelnburg war am Institut für theoretische Physik der TH Darmstadt ab 1936 zunächst als Oberassistent und dann als Extraordinarius tätig. Er stellte sich in den Dienst des Nazi-Regimes – von 1938 bis 1942 hatte er das Amt eines NS-Dozentenbundführers inne
Delay adaptation does not transfer between discrete button press actions and continuous control
No full textWhen interacting with technology, humans often deal with delays between an action and the desired action outcome. Through delay adaptation these delays will become less detrimental to visuomotor performance over time. Delay adaptation has been shown for a variety of tasks and control modes, from simple button presses causing a beep or flash to continuous target-tracking tasks. Here we investigated whether the delay adaptation is specific for the control mode used, when the task itself remained unaltered. To this end, participants performed a target tracking task in which they controlled a cursor item either by moving a stylus on a graphics tablet or by pressing the arrow keys on a keyboard. We found that delay adaptation occurred for both these types of control modes, but observed no transfer to the other control mode. This indicates that delay adaptation is specific to the control mode used during adaptation
SASVi: segment any surgical video
Get PDFPurpose: Foundation models, trained on multitudes of public datasets, often require additional fine-tuning or re-prompting mechanisms to be applied to visually distinct target domains such as surgical videos. Further, without domain knowledge, they cannot model the specific semantics of the target domain. Hence, when applied to surgical video segmentation, they fail to generalise to sections where previously tracked objects leave the scene or new objects enter.
Methods: We propose SASVi, a novel re-prompting mechanism based on a frame-wise object detection Overseer model, which is trained on a minimal amount of scarcely available annotations for the target domain. This model automatically re-prompts the foundation model SAM2 when the scene constellation changes, allowing for temporally smooth and complete segmentation of full surgical videos.
Results: Re-prompting based on our Overseer model significantly improves the temporal consistency of surgical video segmentation compared to similar prompting techniques and especially frame-wise segmentation, which neglects temporal information, by at least 2.4%. Our proposed approach allows us to successfully deploy SAM2 to surgical videos, which we quantitatively and qualitatively demonstrate for three different cholecystectomy and cataract surgery datasets.
Conclusion: SASVi can serve as a new baseline for smooth and temporally consistent segmentation of surgical videos with scarcely available annotation data. Our method allows us to leverage scarce annotations and obtain complete annotations for full videos of the large-scale counterpart datasets. We make those annotations publicly available, providing extensive annotation data for the future development of surgical data science models
Algorithmic Accountability: An Analysis of AI Developers' Perceptions and Behavioral Responses
Get PDFThe increasing integration of Information Systems (IS) based on Artificial Intelligence (AI) into diverse societal and organizational domains has made algorithmic accountability a critical concern in IS research and practice. As these systems assume greater roles in high-stakes decision-making, such as in healthcare, finance, and criminal justice, they raise pressing questions about ethics, governance, and, ultimately, algorithmic accountability. Algorithmic accountability aims to clarify who is obligated to justify the design, use, and outcomes of AI systems and who bears responsibility for their potential negative consequences. While policymakers, organizations, and the public emphasize the need for algorithmic accountability, much of the existing discourse has mainly remained conceptual, raising the question of how algorithmic accountability and perceptions of it materialize in practice and what concrete effects they have. Understanding these manifestations is crucial, particularly concerning AI developers, who directly shape AI design and whose accountability perceptions influence their development decisions. Against this backdrop, this dissertation examines how accountability triggers foster accountability perceptions among AI developers, how these perceptions manifest, and how they influence AI developers’ behavior in AI systems development. The findings reveal that direct indications, such as accountability arguments embedded in IS engineering tools, effectively evoke accountability perceptions among AI developers. However, these perceptions are not uniform but rather multifaceted, varying in intensity and reference points. While they often lead AI developers to favor more cautious designs of AI systems, unclear accountability attributions can negatively impact work-related affective states. These insights highlight the importance of designing algorithmic accountability mechanisms that trigger accountability perceptions and clarify their scope and implications, ensuring both responsible AI systems development and sustainable work environments for AI developers.
This dissertation consists of four peer-reviewed articles (Article A–D) that address the socio-technical and behavioral dimensions of algorithmic accountability in AI systems development. The first part of this dissertation explores how organizations can trigger and shape accountability perceptions. Given the limitations of established governance mechanisms such as AI principles and AI audits, Article A introduces accountability arguments as embedded accountability triggers within IS engineering tools. Using a mixed-method research approach, the article demonstrates that AI developers differentiate between accountability perceptions related to development processes (process accountability) and those concerning the outcomes of AI systems (outcome accountability). The findings reveal that process accountability is more immediately perceived, while outcome accountability requires targeted interventions to be internalized equally effectively by AI developers. These insights advance IS research by conceptualizing accountability arguments as a dynamic governance mechanism that actively shapes AI developers’ accountability perceptions in AI systems development.
The second part of this dissertation examines how different forms of accountability perceptions manifest among AI developers. Through an online survey, Article B highlights the consequences of incongruence in intrapersonal accountability perceptions, differentiating between self-attributed accountability and others-attributed accountability, referring to accountability assigned by others. The article demonstrates that misalignment between these perceptions increases role ambiguity and reduces job satisfaction, underscoring the need for clear and transparent algorithmic accountability communication within organizations. Through qualitative interviews, Article C further refines this understanding by distinguishing between two conceptualizations of algorithmic accountability: one as an intrinsic ethical virtue shaping AI developers’ decision-making and the other as an external governance mechanism ensuring compliance with organizational and regulatory standards. The findings reveal that AI developers’ ethical orientations influence whether they proactively integrate algorithmic accountability into their decision-making or adapt a more reactive, compliance-driven approach. This differentiation is essential for organizations seeking to cultivate a shared algorithmic accountability culture within AI systems development teams.
The third part of this dissertation explores how accountability perceptions shape AI developers’ behavior, especially related to AI design. While prior IS research has predominantly focused on the effects of accountability perceptions on users’ behavior, Article D shifts the focus to AI developers as decision-makers by employing a scenario-based survey, revealing that heightened accountability perceptions lead to more cautious and risk-averse AI design preferences. AI developers who perceive strong accountability tend to reduce AI systems’ autonomy and inscrutability while prioritizing their learnability. This article advances IS research by demonstrating that algorithmic accountability is not only a governance mechanism but also a factor that actively shapes AI design. These findings call for organizations to carefully balance algorithmic accountability mandates with innovation goals, as excessive algorithmic accountability pressure may constrain exploratory design decisions.
Taken together, the articles in this dissertation contribute to IS research by providing a more holistic understanding of how accountability triggers evoke accountability perceptions among AI developers, how these perceptions take shape in diverse and multifaceted ways, and how they ultimately influence AI systems development practices and decision-making. In doing so, this dissertation conceptualizes algorithmic accountability as a multi-layered construct, examining how AI developers internalize algorithmic accountability, how inconsistencies in accountability perceptions affect work-related affective states, and how these perceptions shape AI developers’ behavior. By differentiating between process and outcome accountability within AI systems development, self- and others-attributed accountability, and algorithmic accountability as a virtue versus a mechanism, this dissertation advances a more nuanced perspective on algorithmic accountability and its broader implications. These insights lay the groundwork for future IS research on algorithmic accountability as a dynamic and evolving governance mechanism within IS development practices. From a practical perspective, this dissertation offers valuable guidance for organizations and policymakers. For organizations, the findings suggest that integrating embedded algorithmic accountability interventions into development workflows can enhance clarity and consistency in algorithmic accountability communication, helping to minimize perceptual misalignment among AI developers. Rather than merely imposing mandates, effective algorithmic accountability frameworks must actively shape how algorithmic accountability is understood, internalized, and applied in practice, ensuring that AI developers engage with it as an embedded and actionable aspect of their work. For policymakers, this dissertation underscores that regulatory approaches must not only mandate algorithmic accountability but also consider how AI developers perceive and internalize these requirements. Ambiguously framed algorithmic accountability mandates risk creating unintended and potentially counterproductive consequences, as ambiguous understandings of algorithmic accountability may negatively impact AI developers’ ability to adhere to algorithmic accountability standards in practice.
These findings call for closer collaboration between researchers, organizations, and policymakers to ensure that algorithmic accountability remains both theoretically sound and practically implementable. Future IS research should explore how accountability perceptions evolve over time, how interactions between AI stakeholders shape algorithmic accountability, and how algorithmic accountability mechanisms influence AI system adoption and long-term societal outcomes. Ultimately, this dissertation lays the groundwork for developing more effective governance strategies for AI systems, enabling organizations to proactively shape accountability perceptions, and ensuring that AI systems are not only technically advanced but also aligned with ethical and societal expectations