Fakultät Informatik
To develop truly human-centered automated systems, it is essential to acknowledge that human reasoning is prone to systematic deviations from rational judgment, known as Cognitive Biases. The present study investigated such flawed reasoning in the context of automated driving. In a multi-step study with N = 34 participants, the occurrence of four Cognitive Biases was examined: Truthiness Effect, Automation Bias, Action Bias, and Illusory Control. Additionally, the study explored how the Explainability of the automation’s behavior and the driver’s Mental Model influenced the manifestation of these biases. The findings indicate a notable susceptibility to the Truthiness Effect and Illusory Control, although all biases appeared highly dependent on the specific driving context. Moreover, Explainability strongly impacted the perceived credibility of information and participants’ agreement with the system’s behavior. Given the exploratory nature of the study, this work aims to initiate a discussion on how Cognitive Biases shape human reasoning and decision-making in interactions with automated vehicles. Based on the results, several directions for future research are proposed: (1) investigation of additional cognitive biases, (2) analysis of biases across different levels of automation, (3) exploration of mitigation strategies versus deliberate use of biases, (4) examination of dynamic and context-dependent manifestations, and (5) validation in high-fidelity simulations or real-world settings.
This paper investigates the effects of motion mismatches on simulator sickness and subjective ratings of the motion. In an open-loop driving simulator experiment, participants were driven through a recorded urban drive twelve times, in which mismatches were induced by manipulating the following three aspects in motion cueing: (i) mismatches in specific vehicle axes, (ii) mismatch types (scaling, missing, and false cues), and (iii) inconsistent scaling between different motion axes. Subjects (N=52) reported simulator sickness post-hoc (after each drive), as well as continuously during each drive, a first in simulator sickness research. Furthermore, subjective post-hoc motion incongruence ratings on the quality of the motion were extracted. Results show that longitudinal motion mismatches lead to the most simulator sickness and the highest ratings, followed by mismatches in lateral motion, then yaw rate. False cues induce the most sickness, followed by missing and then scaled motion. Inconsistent scaling between the axes has no significant effect. The continuous sickness ratings support that the occurrence and severity of simulator sickness are indeed related to mismatches in simulator motion of specific maneuvers. This paper contributes to an improved understanding of the relationship between simulator motion and sickness, allowing for more targeted motion cueing strategies to prevent and reduce sickness in driving simulators. These strategies may include the appropriate selection of the simulator, the motion cueing, and the sample of participants, following the presented results.
This guidebook is offered as part of the THI Lehrvideocommunity Project. It is designed to assist both students and professors in producing high-quality videos for various purposes - from work and project presentation to peer education. It offers a useful starting point if you are facing such task for the first time, as well as potential avenues for improvement for those already more experienced.
Neural networks are vulnerable to adversarial attacks. Existing robustness evaluation methods have notable limitations, which makes robustness assessment challenging. This work explores robustness evaluation techniques and identifies key factors, including distance metrics, loss functions, attack generation algorithms, attacker models, specificity, and computational resources. Building on those factors, a novel robustness metric for classification tasks is proposed. Our metric accounts for both, targeted and untargeted attacks across three attacker models, while incorporating accuracy and loss into a weighted aggregation. The scoring includes robustness-versus-perturbation and loss-versus-perturbation curves. Our robustness metric offers a more reliable evaluation and deeper insights into model vulnerability compared to previous approaches.
Driving automation aims to enhance comfort, safety, and traffic flow by removing the human driver from the control loop. However, the human experience of commuting involves more than just reaching a destination or assuming the role of a driver. Factors like personal driving style and courtesy towards fellow road users are integral to the driving experience but often overlooked in the development of driving algorithms for automated vehicles. In this study, we explored the needs of passengers in highly automated vehicles. A qualitative use case analysis was conducted (N=16). In a second study, N=15 participants experienced the resulting use cases in an automated vehicle. In these scenarios, they were able to interact with the automation through a cooperation HMI. Results indicate that most participants expressed a desire for cooperative driving, albeit varying with the driving situation. Moreover, allowing cooperation improves passengers’ overall experience by satisfying psychological needs for autonomy, security, competence, and relatedness.
Artificial intelligence (AI) tools are increasingly employed in clinical genetics to assist in diagnosing genetic conditions by assessing photographs of patients. For medical uses of AI, explainable AI (XAI) methods offer a promising approach by providing interpretable outputs, such as saliency maps and region relevance visualizations. XAI has been discussed as important for regulatory purposes and to enable clinicians to better understand how AI tools work in practice. However, the real-world effects of XAI on clinician performance, confidence, and trust remain underexplored. This study involved a web-based user experiment with 31 medical geneticists to assess the impact of AI-only diagnostic assistance compared to XAI-supported diagnostics. Participants were randomly assigned to either group and completed diagnostic tasks with 18 facial images of individuals with known genetic syndromes and unaffected individuals, before and after experiencing the AI outputs. The results show that both AI-only and XAI approaches improved diagnostic accuracy and clinician confidence. The effects varied according to the accuracy of AI predictions and the clarity of syndromic features (sample difficulty). While AI support was viewed positively, users approached XAI with skepticism. Interestingly, we found a positive correlation between diagnostic improvement and XAI intervention. Although XAI support did not significantly enhance overall performance relative to AI alone, it prompted users to critically evaluate images with false predictions and influenced their confidence levels. These findings highlight the complexities of trust, perceived usefulness, and interpretability in AI-assisted diagnostics, with important implications for developing and implementing clinical decision-support tools in facial phenotyping for rare genetic diseases.
Leveling the Playing Field
(2025)
Participating in studies with lengthy questionnaires can be exhausting, especially for individuals with disabilities or neurodivergent individuals. Gamification can potentially enhance user experience (UX) and, thus, create a more inclusive engagement with studies. This study investigates gamified studies’ UX for neurodivergent persons. A gamified version of the Quality of Life Enjoyment and Satisfaction Questionnaire – Short Form (Q-LES-Q-SF) was created using motivational theory and the gamification frameworks Marczewski’s Periodic Table of Gamification Elements and Gamification User Types Hexad Scale. First results of a preliminary comparative survey with N = 72 (gamified n = 31) suggest that gamification increases the UX for neurodivergent participants.
In public transport, arrival times are typically communicated as point forecasts, aiming to present precise estimates. However, current prediction models are unable to provide such precise and reliable estimates due to unpredictable events. This results in arrival times on passenger information systems appearing inaccurate due to the lack of communicated uncertainty. We therefore investigated interval forecasts as an alternative in an online study, aiming to better communicate uncertainty in arrival times. Our findings indicate that interval forecasts improve the communication of uncertainty. Further, user satisfaction was driven primarily by waiting time, and this relationship was moderated by the forecast concept. Point forecasts were only well received when the bus arrived as predicted, otherwise users preferred the broader interval forecasts. Participants valued accuracy over precision when judging arrival times.
As virtual reality (VR) technology advances, its potential for widespread application increases. However, VR sickness (VRS) remains a major obstacle to broader adoption. This study investigates the effect of auditory stimuli on VRS during VR experiences. It contrasts the effects of music and podcasts, with a no-sound environment serving as reference. Twenty-three participants experienced three different auditory conditions during a VR roller coaster ride in a randomized, balanced order. VRS was quantified using the Simulator Sickness Questionnaire (SSQ) and physiological data. The SSQ results indicate that music significantly reduced oculomotor disturbances and disorientation, while podcasts had no positive effects. The physiological data demonstrated no significant effects. The majority of participants preferred the music scenario, describing it as relaxing and pleasant. This highlights the potential of music, especially when self-selected and perceived as pleasant, to improve VR experience by significantly reducing VRS. This effect appears to be independent of physiological data.
With the emergence of Urban Air Mobility (UAM), eVTOLs are set to revolutionize the way of traveling. However, there will be situations where things do not go as planned, such as medical emergencies during a flight. Due to the lack of cabin crew, tailored support measures must be identified to ensure passenger safety and well-being. To explore how potential passengers perceive medical emergencies on board as bystanders and what kind of assistance they expect, we conducted a VR study. The results show significant increases in negative emotions during a medical in-flight emergency. Moreover, support measures involving human interaction were rated significantly higher than those relying solely on information displayed on onboard screens. Particularly well-received were pilot announcements. For UAM to enter the market successfully, it is essential to address passenger needs in critical situations. Our findings provide valuable insights into shaping effective and user-centered emergency support strategies for eVTOL operations.
Too Close for Comfort? The Impact of eVTOL-Overflights in Residential Areas on Non-Users' Acceptance
(2025)
Urban Air Mobility (UAM) has the potential to revolutionize commuting by allowing passengers to travel quickly and efficiently within and between cities and airports. However, this innovation also raises concerns for residents on the ground, who are expected to tolerate frequent eVTOL overflights above their homes - an issue that this paper seeks to address. To investigate acceptance of eVTOLs from the perspective of residents on the ground being overflown at 1000 ft, 1500 ft, and 2000 ft, a virtual reality study was conducted. Results showed significant differences in emotions, the feeling of being disturbed by the noise, the spatial proximity, and the presence of the eVTOL in lower altitudes. Additionally, privacy concerns were expressed. The findings help the scientific community and regulators in developing guidelines for operating eVTOLs in residential areas in an acceptable manner for non-passengers.
Deep learning has recently been shown to be effective in various tasks related to static binary analysis. One important analysis task is the binary function similarity problem: Given the binary code of two functions compiled with different compilers, different settings, and different processor architectures, the goal is to decide whether the functions are semantically equivalent (i.e. "similar") or not. This problem has numerous applications for embedded systems, for example plagiarism detection, validation of compliance restrictions with usable software licenses, more efficient reverse engineering of existing binary codebases, or vulnerability scanning by detecting known vulnerable functions. In this paper, we propose a novel training scheme for the popular transformer neural network architecture to learn function embeddings directly from instruction listings. Unlike existing approaches, our solution explicitly considers the cross-architecture scenario: we propose a training method to adapt the model to different instruction set architectures (ISA) without having to train a new model from scratch, which allows the model to also be used efficiently for embedded systems, where there are a variety of different processor architectures. We show that our solution achieves a similarity classification accuracy of 89.6% on a dataset consisting of several real-world open source software projects. Finally, we conduct extensive experiments to demonstrate the effectiveness of knowledge distillation in increasing the computational efficiency of the embedding model. We demonstrate a reduction in the number of parameters from 87M to 23M, while still maintaining a classification accuracy of 87.8%. Our code and artifacts are available as open source.
In the last two decades, the popularity of self-adaptive systems in the field of software and systems engineering has drastically increased. However, despite the extensive work on self-adaptive systems, the literature still lacks a common agreement on the definition of these systems. To this day, the notion of self-adaptive systems is mainly used intuitively without a precise understanding of the terminology. Using terminology only by intuition does not suffice, especially in engineering and science, where a more rigorous definition is necessary. In this paper, we investigate the existing formal definitions of self-adaptive systems and how these systems are characterised across the literature. Additionally, we analyse and summarise the limitations of the existing formal definitions in order to understand why none of the existing formal definitions is used more broadly by the community. To achieve this, we have conducted a systematic literature review in which we have analysed over 1400 papers related to self-adaptive systems. Concretely, from an initial pool of 1493 papers, we have selected 314 relevant papers, which resulted in nine primary studies whose primary objective was to define self-adaptive systems formally. Our systematic review reveals that although there has been an increasing interest in self-adaptive systems over the years, there is a scarcity of efforts to define these systems formally. Finally, as part of this paper, based on the analysed primary studies, we also elicit requirements and set a foundation for a potential (formal) definition in the future that is accepted by the community on a broader range.
Background: Due to their diversity, complexity, and above all importance, safety-critical and dependable systems must be developed with special diligence. Criticality increases as these systems likely contain artificial intelligence (AI) components known for their uncertainty. As software and reference architectures form the backbone of any successful system, including safety-critical dependable systems with learning-enabled components, choosing the suitable architecture that guarantees safety despite uncertainties is of great eminence.
Aim: We aim to provide the missing overview of all existing architectures, their contribution to safety, and their level of maturity in AI-based safety-critical systems.
Method: To achieve this aim, we report a systematic mapping study. From a set of 1,639 primary studies, we selected 38 relevant studies dealing with safety assurance through software architecture in AI-based safety-critical systems. The selected studies were then examined using various criteria to answer the research questions and identify gaps in this area of research.
Results: Our findings showed which architectures have been proposed and to what extent they have been implemented. Furthermore, we identified gaps in different application areas of those systems and explained these gaps with various arguments.
Conclusion: As the AI trend continues to grow, the system complexity will inevitably increase, too. To ensure the lasting safety of the systems, we provide an overview of the state of the art, intending to identify best practices and research gaps and direct future research more focused.
Background: Contract-based Design (CbD) is a valuable methodology for software design that allows annotation of code and architectural components with contracts, thereby enhancing clarity and reliability in software development. It establishes rules that outline the behaviour of software components and their interfaces and interactions. This modular approach enables the design process to be segmented into smaller, independently developed, tested, and verified system components, ultimately leading to more robust and dependable software. Aim: Despite the significance and well-established theoretical background of CbD, there is a need for a comprehensive systematic mapping study for reliable software systems. Our study provides an evidence-based overview of a method and demonstrates its practical feasibility.
Method: To conduct this study, we systematically searched three different databases using specially formulated queries, which initially yielded 1,221 primary studies. After voting, we focused on 288 primary studies for more detailed analysis. Finally, a collaborative review allowed us to gather relevant evidence and information to address our research questions. Results: Our findings suggest potential avenues for future research trajectories in CbD, emphasising its role in improving the dependability of software systems. We highlight maturity levels across different domains and identify areas that may benefit from further research.
Conclusion: Although CbD is a well-established software design approach, a more comprehensive literature review is needed to clarify its theoretical state about dependable systems. Our study addresses this gap by providing a detailed overview of CbD from various perspectives, identifying key gaps, and suggesting future research directions.
In recent years an increasing amount of electric vehicle fast charging stations have been built to meet the growing demand from rising electric vehicle numbers. The standard for fast charging communication in europe is ISO 15118. In theory the standard includes security controls for authentication and transport encryption. In reality difficulties with implementing those security controls as well as insecure design within the standard lead to multiple possible attack vectors compromising the confidentiality and authenticity of charging communication sessions. The goal of this research is to present different approaches towards performing sniffing and man-in-the-middle attacks against charging communication. We also provide a novel approach, which does not rely on race conditions and thus is more reliable than previous approaches.
Shared automated mobility-on-demand promises efficient, sustainable, and flexible transportation. Nevertheless, security concerns, resilience, and their mutual influence - especially at night - will likely be the most critical barriers to public adoption since passengers have to share rides with strangers without a human driver on board. As related work points out that information about fellow travelers might mitigate passengers' concerns, we designed two user interface variants to investigate the role of this information in an exploratory within-subjects user study (N = 24). Participants experienced four automated day and night rides with varying personal information about co-passengers in a simulated environment. The results of the mixed-method study indicate that having information about other passengers (e.g., photo, gender, and name) positively affects user experience at night. In contrast, it is less necessary during the day. Considering participants' simultaneously raised privacy demands poses a substantial challenge for resilient system design.
The main approaches for simulating FMCW radar are based on ray tracing, which is usually computationally intensive and do not account for background noise. This work proposes a faster method for FMCW radar simulation capable of generating synthetic raw radar data using generative adversarial networks (GAN). The code and pre-trained weights are open-source and available on GitHub. This method generates 16 simultaneous chirps, which allows the generated data to be used for the further development of algorithms for processing radar data (filtering and clustering). This can increase the potential for data augmentation, e.g., by generating data in non-existent or safety-critical scenarios that are not reproducible in real life. In this work, the GAN was trained with radar measurements of a motorcycle and used to generate synthetic raw radar data of a motorcycle traveling in a straight line. For generating this data, the distance of the motorcycle and Gaussian noise are used as input to the neural network. The synthetic generated radar chirps were evaluated using the Frechet Inception Distance (FID). Then, the Range-Azimuth (RA) map is calculated twice: first, based on synthetic data using this GAN and, second, based on real data. Based on these RA maps, an algorithm with adaptive threshold and edge detection is used for object detection. The results have shown that the data is realistic in terms of coherent radar reflections of the motorcycle and background noise based on the comparison of chirps, the RA maps and the object detection results. Thus, the proposed method in this work has shown to minimize the simulation-to-reality gap for the generation of radar data.
Adaptive Message Prioritization: How to Prioritize VRU Awareness Messages in a Congested V2X Network
(2025)
Foundation models (FMs), i.e., models trained on a vast amount of typically unlabeled data, have become popular and available recently for the domain of histopathology. The key idea is to extract semantically rich vectors from any input patch, allowing for the use of simple subsequent classification networks potentially reducing the required amounts of labeled data, and increasing domain robustness. In this work, we investigate to which degree this also holds for mitotic figure classification. Utilizing two popular public mitotic figure datasets, we compared linear probing of five publicly available FMs against models trained on ImageNet and a simple ResNet50 end-to-end-trained baseline. We found that the end-to-end-trained baseline outperformed all FM-based classifiers, regardless of the amount of data provided. Additionally, we did not observe the FM-based classifiers to be more robust against domain shifts, rendering both of the above assumptions incorrect.
Driving under the influence of alcohol (DUI) remains a leading cause of accidents globally, with accident risk rising exponentially with blood alcohol concentration (BAC). This study aims to distinguish between sober and intoxicated drivers using driving behavior analysis and driver monitoring system (DMS), technologies that align with emerging EU regulations. In a driving simulator, twenty-three participants (average age: 32) completed five drives (one practice and two each while sober and intoxicated) on separate days across city, rural, and highway settings. Each 30-minute drive was analyzed using eye-tracking and driving behavior data. We applied significance testing and classification models to assess the data. Our study goes beyond the state of the art by a) combining data from various sensors and b) not only examining the effects of alcohol on driving behavior but also using these data to classify driver impairment. Fusing gaze and driving behavior data improved classification accuracy, with models achieving over 70% accuracy in city and rural conditions and a Long Short-Term Memory (LSTM) network reaching up to 80% on rural roads. Although the detection rate is, of course, still far too low for a productive system, the results nevertheless provide valuable insights for improving DUI detection technologies and enhancing road safety.
Efficient nutrient management and precise fertilization are essential for advancing modern agriculture, particularly in regions striving to optimize crop yields sustainably. The AgroLens project endeavors to address this challenge by develop ing Machine Learning (ML)-based methodologies to predict soil nutrient levels without reliance on laboratory tests. By leveraging state of the art techniques, the project lays a foundation for acionable insights to improve agricultural productivity in resource-constrained areas, such as Africa. The approach begins with the development of a robust European model using the LUCAS Soil dataset and Sentinel-2 satellite imagery to estimate key soil properties, including phosphorus, potassium, nitrogen, and pH levels. This model is then enhanced by integrating supplementary features, such as weather data, harvest rates, and Clay AI-generated embeddings. This report details the methodological framework, data preprocessing strategies, and ML pipelines employed in this project. Advanced algorithms, including Random Forests, Extreme Gradient Boosting (XGBoost), and Fully Connected Neural Networks (FCNN), were implemented and finetuned for precise nutrient prediction. Results showcase robust model performance, with root mean square error values meeting stringent accuracy thresholds. By establishing a reproducible and scalable pipeline for soil nutrient prediction, this research paves the way for transformative agricultural applications, including precision fertilization and improved resource allocation in underresourced regions like Africa.