006 Spezielle Computerverfahren
Refine
Departments, institutes and facilities
- Fachbereich Informatik (118)
- Institute of Visual Computing (IVC) (45)
- Institut für KI und Autonome Systeme (A2S) (25)
- Fachbereich Wirtschaftswissenschaften (23)
- Institut für Technik, Ressourcenschonung und Energieeffizienz (TREE) (23)
- Institut für Verbraucherinformatik (IVI) (22)
- Institut für Sicherheitsforschung (ISF) (11)
- Fachbereich Ingenieurwissenschaften und Kommunikation (7)
- Graduierteninstitut (4)
- Institut für Cyber Security & Privacy (ICSP) (4)
Document Type
- Conference Object (76)
- Article (57)
- Part of a Book (10)
- Dataset (6)
- Preprint (6)
- Report (6)
- Doctoral Thesis (5)
- Book (monograph, edited volume) (3)
- Contribution to a Periodical (3)
- Other (3)
Year of publication
Keywords
- Virtual Reality (7)
- Augmented Reality (6)
- deep learning (5)
- virtual reality (5)
- Machine Learning (4)
- Machine learning (4)
- Knowledge Graphs (3)
- Large Language Models (3)
- UAV (3)
- facial expression analysis (3)
Interactions between consumers and companies are increasingly relying on technologies such as chatbots and voice assistants that are based on natural language processing (NLP) techniques. With the advent of more sophisticated technologies such as transformers and generative artificial intelligence, this trend will likely continue and further solidify. To our knowledge, this study is the first to systematically review the current scientific discourse on NLP-based technologies in the context of the customer journey and attempts to outline existing knowledge and identify gaps before the onset of a new era in NLP sophistication. Employing the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) method and co-occurrence analysis, we offer new and nuanced insights into the prevailing discourse. From a sample of 734 articles, 41 studies were selected and analyzed. Our findings shed light on the current research focus, exploring various technologies, concepts, and challenges. We also offer a starting point for how emerging NLP-based technologies could impact the customer journey, as well as future research directions.
Arm's Platform Security Architecture (PSA) is a family of hardware and firmware security specifications, along with open-source reference implementations, aimed at helping device makers and chip manufacturers integrate best-practice security into their products. Devices that comply with PSA can generate attestation tokens as described in this document, which serve as the foundation for various protocols, including secure provisioning and network access control. This document specifies the structure and semantics of the PSA attestation token.
The PSA attestation token is a profile of the Entity Attestation Token (EAT). This specification describes the claims used in an attestation token generated by PSA-compliant systems, how these claims are serialized for transmission, and how they are cryptographically protected.
This Informational document is published as an Independent Submission to improve interoperability with Arm's architecture. It is not a standard nor a product of the IETF.
The success of any agent, human or artificial, ultimately depends on their successfully accomplishing the given goals. Agents may, however, fail to do so for many reasons. With artificial agents, such as robots, this may be due to internal faults or exogenous events in the complex, dynamic environments in which they operate. The bottom line is that plans, even good ones, can fail. Despite decades of research, effective methods for artificial agents to cope with plan failure remain limited and are often impractical in the real world. One common reason for failure that plagues agents, human and artificial alike, is that objects that are expected to be used to get the job done are often found to be missing or unavailable. Humans might, with little effort, accomplish their tasks by making substitutions. When they are not sure if an object is available, they may even proceed optimistically and switch to making a substitution when they confirm that an object is indeed unavailable. In this work, the system uses Description Logics to enable open-world reasoning --- making it possible to distinguish between cases where an object is missing/unavailable and cases where the failure to even generate a plan is due to the planner's use of the closed-world assumption (where the fact stating that something is true is missing from its knowledge base and so it is assumed to be not true). This ability to distinguish between something being missing and having incomplete information enables the agent to behave intelligently: recognising whether it should identify and then plan with a suitable substitute or create a placeholder, in the case of incomplete information. By representing the functional affordances of objects (i.e. what they are meant to be used for), socially-expected and accepted object substitutions are made possible. The system also uses the Conceptual Spaces approach to provide feature-based similarity measures that make the given task a first-class citizen in the identification of a suitable substitute. The generation of plans to `get the job done' is made possible by incorporating the Hierarchical Task Network planning approach. It is combined with a robust execution/monitoring system and contributes to the success of the robot in achieving its goals.
Unmanned Aerial Vehicles (UAVs) are increasingly used for reforestation and forest monitoring, including seed dispersal in hard-to-reach terrains. However, a detailed understanding of the forest floor remains a challenge due to high natural variability, quickly changing environmental parameters, and ambiguous annotations due to unclear definitions. To address this issue, we adapt the Segment Anything Model (SAM), a vision foundation model with strong generalization capabilities, to segment forest floor objects such as tree stumps, vegetation, and woody debris. To this end, we employ parameter-efficient fine-tuning (PEFT) to fine-tune a small subset of additional model parameters while keeping the original weights fixed. We adjust SAM's mask decoder to generate masks corresponding to our dataset categories, allowing for automatic segmentation without manual prompting. Our results show that the adapter-based PEFT method achieves the highest mean intersection over union (mIoU), while Low-rank Adaptation (LoRA), with fewer parameters, offers a lightweight alternative for resource-constrained UAV platforms.
The first Data Competence College was hosted from March 27th to 28th, 2025 at the IT center of RWTH Aachen. Based on the concept of the Wissenschaftskolleg in Berlin or the Institute of Advanced Studies in Princeton, we invited two individuals with high data competence from different scientific fields (“Data Experts”) to participate as part of the data competence college:
Prof. Sebastian Houben (Hochschule Bonn-Rhein-Sieg, specialist in AI and autonomous systems)
Dr. Moritz Wolter (University of Bonn, expert in high performance computing and machine learning)
For two days we aimed to create a space where not only local scientists, and especially early career researchers, learn from the data experts and each other regarding research data and methods but also data experts could inspire each other. The schedule included keynote presentations by all data experts, poster and group presentations by the participants, 1:1 sessions between data experts and early career researchers, as well as a method- and data-related workshop. We aimed foremost to create an environment in which everyone feels safe to give input, share their knowledge and learn from the other participants and experts.
Second International Workshop on Perception-driven Graphics and Displays for VR and AR (PerGraVAR)
(2025)
Generative AI can considerably speed up the process of producing narrative content including different media. This may be particularly helpful for the generation of modular variations on narrative themes in hypermedia, crossmedia, or transmedia contexts, thereby enabling personalized access to the content by heterogenous target groups. We present an example where GenAI has been applied for image creation and translation of a text to multiple languages for a crossmedia edutainment project transferring IT security knowledge to vulnerable groups. GenAI still seems inadequate to produce interesting narrative text integrating dedicated educational content. AI-generated illustrations often require manual rework. However, LLM support in multilingual translations displays more intelligent solutions than expected, including the implementation of a password generation process from a narrated description.
Implementation of the 3D Face Recognition algorithm described in M. Jribi, S. Mathlouthi, and F. Ghorbel, "A geodesic multipolar parameterization-based representation for 3D face recognition," Signal Processing: Image Communication, vol. 99, 2021.
There is a C++ implementation and a python wrapper (Attention, this underlies BSD-3 Clause licensing) available.
As the Code was only tested with a single STL Face Image yet, it's currently only a Beta. Should you find any bugs, please report them back to me via mail: alexandra.mielke@smail.emt.h-brs.de or add issues in Github.
If you use this software, please cite it as below.
To ensure reliable performance of Question Answering (QA) systems, evaluation of robustness is crucial. Common evaluation benchmarks commonly only include performance metrics, such as Exact Match (EM) and the F1 score. However, these benchmarks overlook critical factors for the deployment of QA systems. This oversight can result in systems vulnerable to minor perturbations in the input such as typographical errors. While several methods have been proposed to test the robustness of QA models, there has been minimal exploration of these approaches for languages other than English. This study focuses on the robustness evaluation of German language QA models, extending methodologies previously applied primarily to English. The objective is to nurture the development of robust models by defining an evaluation method specifically tailored to the German language. We assess the applicability of perturbations used in English QA models for German and perform a comprehensive experimental evaluation with eight models. The results show that all models are vulnerable to character-level perturbations. Additionally, the comparison of monolingual and multilingual models suggest that the former are less affected by character and word-level perturbations.
This paper presents dCTIDH, a CSIDH implementation that combines two recent developments into a novel state-of-the-art deterministic implementation. We combine the approach of deterministic variants of CSIDH with the batching strategy of CTIDH, which shows that the full potential of this key space has not yet been explored. This high-level adjustment in itself leads to a significant speed-up. To achieve an effective deterministic evaluation in constant time, we introduce Wombats, a new approach to performing isogenies in batches, specifically tailored to the behavior required for deterministic CSIDH using CTIDH batching.
Furthermore, we explore the two-dimensional space of optimal primes for dCTIDH, with regard to both the performance of dCTIDH in terms of finite-field operations per prime and the efficiency of finite-field operations, determined by the prime shape, in terms of cycles. This allows us to optimize both for choice of prime and scheme parameters simultaneously. Lastly, we implement and benchmark constant-time, deterministic dCTIDH. Our results show that dCTIDH not only outperforms state-of-the-art deterministic CSIDH, but even non-deterministic CTIDH: dCTIDH-2048 is faster than CTIDH-2048 by 17 percent, and is almost five times faster than dCSIDH-2048.
This article introduces a model-based design, implementation, deployment, and execution methodology, with tools supporting the systematic composition of algorithms from generic and domain-specific computational building blocks that prevent code duplication and enable robots to adapt their software themselves. The envisaged algorithms are numerical solvers based on graph structures. In this article, we focus on kinematics and dynamics algorithms, but examples such as message passing on probabilistic networks and factor graphs or cascade control diagrams fall under the same pattern. The tools rely on mature standards from the Semantic Web. They first synthesize algorithms symbolically, from which they then generate efficient code. The use case is an overactuated mobile robot with two redundant arms.
This specification defines a method for computing a hash value over a CBOR Object Signing and Encryption (COSE) Key. It specifies which fields within the COSE Key structure are included in the cryptographic hash computation, the process for creating a canonical representation of these fields, and how to hash the resulting byte sequence. The resulting hash value, referred to as a "thumbprint", can be used to identify or select the corresponding key.
This dataset contains multimodal data recorded on robots performing robot-to-human and human-to-robot handovers. The intended application of the dataset is to develop and benchmark methods for failure detection during the handovers. Thus the trials in the dataset contain both successful and failed handover actions. For a more detailed description of the dataset, please see the included Datasheet.
RoboCup @work 2023 dataset
(2023)
RoboCup 2023 dataset
(2023)
The relevance of introducing digital twin technology in robotics is substantiated, which allows testing and modelling the capabilities of robots, such as manipulation, grasping, etc., using virtual robot prototypes that are identical copies of physical robot prototypes. An overview of the key components of the digital twin framework for robotics, including the physical element, virtual element, middleware, service and transport components, is provided. A technology for designing a robot using digital twins is proposed, including the design of a computer model of a robot using a computer-aided design system or three-dimensional graphics packages, the use of robot simulation systems, data management, data analysis and human-machine interaction. The further development of the research is the implementation of digital twin technology for a rescue robot according to the proposed stages: building a computer model, programming robot behaviour in a simulation system, developing a mathematical and digital model of the robot, implementing human-machine interaction between a physical robot and its digital replica, which will allow testing the interaction of the main components of the digital twin, performing data exchange between the physical and digital replica, and building a digital data model to verify the main operations.
Ventricular Pressure-Volume Loops Obtained by 3D Real-Time Echocardiography and Mini-Pressure Wire
(2013)
Neuromorphic computing mimics computational principles of the brain in silico and motivates research into event-based vision and spiking neural networks (SNNs). Event cameras (ECs) exclusively capture local intensity changes and offer superior power consumption, response latencies, and dynamic ranges. SNNs replicate biological neuronal dynamics and have demonstrated potential as alternatives to conventional artificial neural networks (ANNs), such as in reducing energy expenditure and inference time in visual classification. Nevertheless, these novel paradigms remain scarcely explored outside the domain of aerial robots. To investigate the utility of brain-inspired sensing and data processing, we developed a neuromorphic approach to obstacle avoidance on a camera-equipped manipulator. Our approach adapts high-level trajectory plans with reactive maneuvers by processing emulated event data in a convolutional SNN, decoding neural activations into avoidance motions, and adjusting plans using a dynamic motion primitive. We conducted experiments with a Kinova Gen3 arm performing simple reaching tasks that involve obstacles in sets of distinct task scenarios and in comparison to a non-adaptive baseline. Our neuromorphic approach facilitated reliable avoidance of imminent collisions in simulated and real-world experiments, where the baseline consistently failed. Trajectory adaptations had low impacts on safety and predictability criteria. Among the notable SNN properties were the correlation of computations with the magnitude of perceived motions and a robustness to different event emulation methods. Tests with a DAVIS346 EC showed similar performance, validating our experimental event emulation. Our results motivate incorporating SNN learning, utilizing neuromorphic processors, and further exploring the potential of neuromorphic methods.
Grading student answers and providing feedback are essential yet time-consuming tasks for educators. Recent advancements in Large Language Models (LLMs), including ChatGPT, Llama, and Mistral, have paved the way for automated support in this domain. This paper investigates the efficacy of instruction-following LLMs in adhering to predefined rubrics for evaluating student answers and delivering meaningful feedback. Leveraging the Mohler dataset and a custom German dataset, we evaluate various models, from commercial ones like ChatGPT to smaller open-source options like Llama, Mistral, and Command R. Additionally, we explore the impact of temperature parameters and techniques such as few-shot prompting. Surprisingly, while few-shot prompting enhances grading accuracy closer to ground truth, it introduces model inconsistency. Furthermore, some models exhibit non-deterministic behavior even at near-zero temperature settings. Our findings highlight the importance of rubrics in enhancing the interpretability of model outputs and fostering consistency in grading practices.
In computer vision, a larger effective receptive field (ERF) is associated with better performance. While attention natively supports global context, its quadratic complexity limits its applicability to tasks that benefit from high-resolution input. In this work, we extend Hyena, a convolution-based attention replacement, from causal sequences to bidirectional data and two-dimensional image space. We scale Hyena’s convolution kernels beyond the feature map size, up to 191×191, to maximize ERF while maintaining sub-quadratic complexity in the number of pixels. We integrate our two-dimensional Hyena, HyenaPixel, and bidirectional Hyena into the MetaFormer framework. For image categorization, HyenaPixel and bidirectional Hyena achieve a competitive ImageNet-1k top-1 accuracy of 84.9% and 85.2%, respectively, with no additional training data, while outperforming other convolutional and large-kernel networks. Combining HyenaPixel with attention further improves accuracy. We attribute the success of bidirectional Hyena to learning the data-dependent geometric arrangement of pixels without a fixed neighborhood definition. Experimental results on downstream tasks suggest that HyenaPixel with large filters and a fixed neighborhood leads to better localization performance.
Altering posture relative to the direction of gravity, or exposure to microgravity has been shown to affect many aspects of perception, including size perception. Our aims in this study were to investigate whether changes in posture and long-term exposure to microgravity bias the visual perception of object height and to test whether any such biases are accompanied by changes in precision. We also explored the possibility of sex/gender differences. Two cohorts of participants (12 astronauts and 20 controls, 50% women) varied the size of a virtual square in a simulated corridor until it was perceived to match a reference stick held in their hands. Astronauts performed the task before, twice during, and twice after an extended stay onboard the International Space Station. On Earth, they performed the task of sitting upright and lying supine. Earth-bound controls also completed the task five times with test sessions spaced similarly to the astronauts; to simulate the microgravity sessions on the ISS they lay supine. In contrast to earlier studies, we found no immediate effect of microgravity exposure on perceived object height. However, astronauts robustly underestimated the height of the square relative to the haptic reference and these estimates were significantly smaller 60 days or more after their return to Earth. No differences were found in the precision of the astronauts’ judgments. Controls underestimated the height of the square when supine relative to sitting in their first test session (simulating Pre-Flight) but not in later sessions. While these results are largely inconsistent with previous results in the literature, a posture-dependent effect of simulated eye height might provide a unifying explanation. We were unable to make any firm statements related to sex/gender differences. We conclude that no countermeasures are required to mitigate the acute effects of microgravity exposure on object height perception. However, space travelers should be warned about late-emerging and potentially long-lasting changes in this perceptual skill.
Diese Arbeit untersucht Entwicklungspraktiken im Kontext professioneller Softwareentwicklung für Augmented und Virtual Reality (XR) Anwendungen. Die Verbindung einer Design Science Linse mit praxeologischen Ansätzen ermöglicht einen umfassenden Einblick in existierende und aufkommende Entwicklungsprozesse in der aufstrebenden XR-Softwareindustrie. Angesichts des aktuellen Mangels an Design Guidelines, Entwicklungs- und Technologiestandards sowie unterstützenden Entwicklungswerkzeugen bietet die Arbeit einen ganzheitlichen Überblick und entwickelt mögliche Lösungsansätze sowie Designvorschläge zur (softwarebasierten) Unterstützung professioneller XR-Entwickler in interdisziplinären Teams.
Virtual Reality (VR) sickness remains a significant challenge in the widespread adoption of VR technologies. The absence of a standardized benchmark system hinders progress in understanding and effectively countering VR sickness. This paper proposes an initial step towards a benchmark system, utilizing a novel methodological framework to serve as a common platform for evaluating contributing VR sickness factors and mitigation strategies. Our benchmark, grounded in established theories and leveraging existing research, features both small and large environments. In two research studies, we validated our system by demonstrating its capability to (1) quickly, reliably, and controllably induce VR sickness in both environments, followed by a rapid decline post-stimulus, facilitating cost and time-effective within-subject studies and increased statistical power, (2) integrate and evaluate established VR sickness mitigation methods — static and dynamic field of view reduction, blur, and virtual nose — demonstrating their effectiveness in reducing symptoms in the benchmark and their direct comparison within a standardized setting. Our proposed benchmark also enables broader, more comparative research into different technical, setup, and participant variables influencing VR sickness and overall user experience, ultimately paving the way for building a comprehensive database to identify the most effective strategies for specific VR applications.
Few mobile robot developers already test their software on simulated robots in virtual environments or sceneries. However, the majority still shy away from simulation-based test campaigns because it remains challenging to specify and execute suitable testing scenarios, that is, models of the environment and the robots’ tasks. Through developer interviews, we identified that managing the enormous variability of testing scenarios is a major barrier to the application of simulation-based testing in robotics. Furthermore, traditional CAD or 3D-modelling tools such as SolidWorks, 3ds Max, or Blender are not suitable for specifying sceneries that vary significantly and serve different testing objectives. For some testing campaigns, it is required that the scenery replicates the dynamic (e.g., opening doors) and static features of real-world environments, whereas for others, simplified scenery is sufficient. Similarly, the task and mission specifications used for simulation-based testing range from simple point-to-point navigation tasks to more elaborate tasks that require advanced deliberation and decision-making. We propose the concept of composable and executable scenarios and associated tooling to support developers in specifying, reusing, and executing scenarios for the simulation-based testing of robotic systems. Our approach differs from traditional approaches in that it offers a means of creating scenarios that allow the addition of new semantics (e.g., dynamic elements such as doors or varying task specifications) to existing models without altering them. Thus, we can systematically construct richer scenarios that remain manageable. We evaluated our approach in a small simulation-based testing campaign, with scenarios defined around the navigation stack of a mobile robot. The scenarios gradually increased in complexity, composing new features into the scenery of previous scenarios. Our evaluation demonstrated how our approach can facilitate the reuse of models and revealed the presence of errors in the configuration of the publicly available navigation stack of our SUT, which had gone unnoticed despite its frequent use.