@inproceedings{7175, abstract = {Deep reinforcement learning (RL) policies for robotic control typically overfit to a single hardware configuration. Any change to the kinematic chain breaks the learned mapping and requires complete retraining. This paper investigates whether combining self-attention mechanisms with explicit spatial observations can improve a policy’s adaptability to varying kinematic topologies. We train a single RL agent to control a robotic manipulator across different degrees of freedom (DOF), ranging from a restricted 4-DOF mode to a fully redundant 7-DOF configuration. Instead of fixed-length state vectors, the method processes the active joints as a variable-length sequence in an attention buffer, enriching each joint’s representation with its relative spatial routing and geometric Jacobian influence. This structure allows the agent to dynamically evaluate the physical utility of its available actuators. The proposed Architecture reaches a success rate of 82.1% across trained topologies and 66.6% on unseen configurations, outperforming MLP and generic attention baselines while remaining nearly collisionfree. Finally, we demonstrate successful sim-to-real transfer by deploying the simulation-trained agent on a physical Franka Emika Panda manipulator.}, author = {Mayer, Patrick Thomas and Rexilius, Jan}, keywords = {Reinforcement Learning, Robotic Manipulation, Morphology Generalization, Sim-to-Real Transfer}, location = {Barcelona}, title = {{Adapting to Variable Kinematic Configurations: A Causal-Kinematic Attention Approach for Robotic Control}}, year = {2026}, } @inproceedings{7174, abstract = {This paper presents a hybrid control architecture for dynamic robotic picking tasks. The framework combines a Deep Reinforcement Learning policy for high-level interception with a dedicated Inverse Kinematics controller for precise terminal grasping, while mitigating precision limitations of monolithic learning-based approaches. The framework utilizes a Proximal Policy Optimization agent to approach moving targets, seamlessly transitioning to an Inverse Kinematics solver that reduces terminal orientational and positional errors while minimizing cumulative control effort. To facilitate deployment on physical hardware, a robust sim-to-real pipeline incorporating system identification, domain randomization, and latency injection is employed. Experimental results on a physical Franka Emika Panda manipulator validate this hybrid architecture. The system achieves an 80% success rate in pick-and-place tasks, compared to 60.8% for unadapted baselines, with no safety-critical violations such as joint limit breaches or collisions observed during testing.}, author = {Mayer, Patrick Thomas and Rexilius, Jan}, keywords = {Reinforcement Learning, Hybrid Control, Sim-to-Real, Robotic Manipulation, Inverse Kinematics.}, location = {Sapporo}, title = {{A Hybrid Control Framework Using Reinforcement Learning for Dynamic Robotic Manipulation and Sim-to-Real Transfer}}, year = {2026}, } @inproceedings{6998, author = {Vovchenko, Viktoriia and Barberi, Vincenzo and Schultenkämper, Sergej and Bäumer, Frederik}, booktitle = {2026 The First International Conference on Security and Cybersecurity in the AI and Digital Context}, isbn = {978-1-68558-449-8}, keywords = {Deepfakes, Image forensics, Diffusion models, Computer vision}, location = {Porto, Portugal}, title = {{ADRIAN InstructFace-Edit - Towards Robust Detection of AI-Manipulated Face Images}}, year = {2026}, } @inproceedings{7002, abstract = {Selecting an appropriate semantic segmentation model for a given application domain remains a challenging and time-consuming task for practitioners and researchers. This paper presents an interactive, web-based platform that enables side-by-side visual comparison of multiple neural network segmentation models applied to identical images. The system integrates three transformer-based segmentation models: a face-parsing network producing 19 semantic classes, a SegFormer-B3 clothing segmentation model with 18 classes, and a Mask2Former model for general-purpose scene segmentation spanning 150 ADE20K categories. Key contributions include side-by-side evaluation of model outputs across multiple architectures and image categories, with real-time segment highlighting and a scalable inference caching system that enables model comparisons without requiring repeated graphics processing unit (GPU) computation. The platform organizes a curated dataset of images under a hierarchical category taxonomy, supporting structured evaluation across demographic and contextual variables. As a practical use case, the system is applied within the ADRIAN project to assist in verifying identity consistency across images through segmentation-based analysis. The platform thus contributes a specialized artificial intelligence (AI) tool for systematic segmentation and object detection model evaluation within media analysis pipelines, where selecting appropriate models is a recurring challenge across tasks from identity verification to content moderation. It is available under https://github.com/vika-v-v/neural-networks-for-image-segmentation and designed to lower the barrier for comparative model evaluation in applied computer vision workflows.}, author = {Vovchenko, Viktoriia and Schultenkämper, Sergej and Bäumer, Frederik}, booktitle = {2026 The Second International Conference on AI-based Media Innovation}, editor = {Böhm, Stephan and Ahrweiler, Petra and Gutenberg, Johannes and Leung, Clement }, isbn = {978-1-68558-403-0}, keywords = {segmentation, model comparison, face parsing}, location = {Nizza, Frankreich}, publisher = {IARIA}, title = {{A Web-Based Platform for Interactive Comparison of Neural Network Image Segmentation Models}}, year = {2026}, } @inproceedings{7153, author = {Schulze, Jan-Philipp and Wette, Philip}, booktitle = {Proceedings of the 31st IEEE International Conference on Emerging Technologies and Factory Automation}, location = {Västerås, Sweden}, publisher = {IEEE}, title = {{Online Reconfiguration of Smart Material Recovery Facilities using Mixed-Integer Linear Programming}}, year = {2026}, } @inproceedings{7173, abstract = {Reinforcement learning has achieved state-of-the-art performance in MAV control, waypoint flight, and obstacle avoidance. However, existing RL approaches often assume fixed objectives and constraints, with flight behavior largely limited by vehicle dynamics and orientation considered only when required for locomotion. Classical planning and model predictive control handle such constraints explicitly, but require optimization or replanning. This motivates methods that combine learned local control with explicit constraint handling. We combine reinforcement learning with control barrier functions to improve constraint-aware execution. We propose parameterized waypoints that encode orientation, velocity, and corridor constraints. Simulation and real-world experiments show that a single policy can execute different constraint-parameterized navigation scenarios, revealing scenario-dependent trade-offs between traversal time, tracking accuracy, and constraint satisfaction.}, author = {Kirsch, André and Rexilius, Jan}, keywords = {MAV navigation, Reinforcement learning, Constrained navigation, Control barrier functions}, location = {Barcelona}, title = {{Reinforcement Learning-based MAV Navigation With Parameterized Waypoints: Incorporating Orientation, Speed Limits, and Corridor Constraints}}, year = {2026}, } @inproceedings{7172, author = {Battermann, Sven and Halbe, Mirko}, booktitle = {Proceedings EMV Kongress 2026 : Internationale Fachmesse und Kongress für Elektromagnetische Verträglichkeit}, location = {Köln}, publisher = {Apprimus}, title = {{In-situ EMV Messung und Analyse eines Photovoltaik Balkonkraftwerks mit Mikrowechselrichter}}, doi = {10.15488/21110}, year = {2026}, } @inproceedings{7171, abstract = {Edge-based Artificial Intelligence (AI) acceleration has recently improved progress in real-time object detection. Object detection on edge devices requires a balance between accuracy, speed, and power efficiency. This paper proposes a customized Deep Learning Processor Unit (DPU)-aware architecture for attention-based YOLO variants deployed on AMD FPGAs. Specifically, we evaluate and benchmark YOLOv26 and YOLOv11, two modern attention-based YOLO variants, on the Xilinx ZCU104 across both standard and oriented object detection tasks. We replace unsupported activation functions, substitute split operations with 1x1 convolutions, and approximate the spatial attention mechanism in a DPU-compatible way. All models are then trained and evaluated across six benchmark datasets such as COCO, Pascal VOC, KITTI, DOTA, DIOR-R, and an in-house human presence dataset, and benchmarked across all eight DPU configurations (B512 to B4096) in terms of mAP, FPS, latency, power, and resource utilization. Notably, YOLOv26n and YOLOv26n-OBB deliver the highest end-to-end throughput at 34.05 and 29.55 FPS for standard and oriented detection, respectively, with an average of 5% absolute reduction in accuracy due to quantization while achieving up to approximately 3x lower power consumption compared with the state of the art.}, author = {Karki, Suraj and Ahmed, Qazi Arbab and Jungeblut, Thorsten}, booktitle = {arXiv:2607.13106}, title = {{No Attention, No Problem: DPU-Aware Attention Approximation in Modern YOLO on FPGA}}, year = {2026}, } @inproceedings{7166, author = {Seidler-Rolf, Kristina and Knoblauch, Flavian and Ilskens, Karina}, location = {Dresden}, title = {{Integration von Planetary Health in die Ausbildung: Lehrende durch ein Train-the-Trainer-Konzept unterstützen}}, year = {2026}, } @inproceedings{7162, abstract = {Heart rate variability biofeedback has emerged as a validated intervention for stress management and emotional regulation through coherence breathing techniques. Existing biofeedback systems typically present cardiac and respiratory signals as separate feedback channels or rely on discrete interface elements that fragment attention and disrupt immersion. We present a design framework that extends Viscereality, a bioresponsive VR framework originally developed for breath mapping, to integrate multi-channel biofeedback including heart rate within immersive VR, centered on a particle-based environment that functions as a spatial body-mapping interface. By leveraging bioresponsive breath interactions for embodiment and weak representational aesthetics for cardiac coherence, the framework aims to encode both streams within a shared visual substrate, reducing the attentional task-switching and experiential fragmentation that multi-channel displays otherwise risk. The system translates lung volume into spatial expansion and contraction of the environment itself, maps cardiac coherence onto affective visual dimensions (angularity, brightness, saturation, symmetry), and embeds breathing guidance within the same geometric substrate. We present the design rationale and implementation of this framework, with empirical validation of its effectiveness to be conducted in future studies.}, author = {Fejer, George and Holzapfel, Till and Hirvonen, Taru and Lalidis Mateo, Anestis-Pere and Blum, Johannes and Gaebler, Michael and Lenggenhager, Bigna}, booktitle = {Proceedings of the 1st International Conference on Human-Computer Interaction in the Alps}, isbn = {9798400719486}, keywords = {heart rate variability, biofeedback, virtual reality, affective state modulation, affective visualization, coherence breathing}, location = {Ascona Switzerland}, pages = {79--86}, publisher = {ACM}, title = {{Breathing Space: Spatial Mapping of Breath and Cardiac Biofeedback for Affective State Representation and Coherence Training in Viscereality}}, doi = {10.1145/3780045.3780061}, year = {2026}, } @misc{7165, author = {Fejer, George and Holzapfel, Till and Hirvonen, Taru and Lalidis Mateo, Anestis-Pere and Blum, Johannes and Gaebler, Michael and Lenggenhager, Bigna}, publisher = {OSF}, title = {{Viscereality: Bioresponsive Virtual Reality for Box Breathing and Altered-State Experience}}, doi = {10.17605/OSF.IO/64MWK}, year = {2026}, } @inproceedings{7155, abstract = {Dieser Beitrag untersucht die Potenziale von Agentic AI und Vibe Coding im Finanz- und Rechnungswesen mittelständischer Unternehmen. Angesichts von Fachkräftemangel und Prozessineffizienzen rücken autonome KI-Systeme und neue Programmierparadigmen zunehmend in den Fokus. Da diese Technologien noch eine geringe Marktdurchdringung aufweisen, nutzt die Studie sieben Large Language Models (LLMs) als exploratives Forschungsinstrument zur systematischen Identifikation und Bewertung von Anwendungsfeldern, Chancen und Risiken. Die Ergebnisse zeigen, dass Agentic AI als Schlüsseltechnologie für die autonome Prozessautomatisierung fungiert, während Vibe Coding die agile Entwicklung individueller Lösungen demokratisiert. Im Vergleich zur Vorstudie 2024/2025 zeigt sich eine signifikante technologische Konkretisierung bei gleichzeitig gestiegenen Anforderungen an Governance und Qualitätssicherung. Die Untersuchung unterstreicht den Wandel hin zu agentischen Architekturen und betont die Notwendigkeit von Hybridwissen, um Effizienzgewinne mit Sicherheitsanforderungen in Einklang zu bringen.}, author = {Uphaus, Andreas and Kampe, Tim}, booktitle = {CARF Luzern 2026 - Controlling. Accounting. Risiko. Finanzen. Luzern, 10.-11.09.}, keywords = {Digitalisierung, Künstliche Intelligenz, Automatisierung, Rechnungswesen, Finanzwesen}, location = {Luzern}, title = {{Agentic AI und Vibe Coding im Finanz- und Rechnungswesen: Anwendungsfälle, Chancen und Risiken für KMU und Familienunternehmen}}, year = {2026}, } @inproceedings{7157, author = {Schröder, Nadine and Schamp, Christina and Reutterer, Thomas}, location = {Berlin, Germany}, title = {{How Shopping Clubs Work: Differential Effects of Nudges Across the Customer Decision Journey }}, year = {2026}, } @inproceedings{7156, author = {Schröder, Nadine and Schamp, Christina and Reutterer, Thomas}, location = {Carcavelos, Portugal}, title = {{How Shopping Clubs Work: Differential Effects of Nudges Across the Customer Decision Journey}}, year = {2026}, } @article{7159, abstract = {Background: Smart home technology powered by AI can detect anomalies and make emergency calls, enabling residents to live safely and independently. However, the adoption of such technologies for medical emergency detection remains limited. Objective: This study aimed to explore consumer preferences for AI-based smart home technology for medical emergency detection and identify predictors such as sociodemographic variables, AI literacy, and technology affinity. Method: A sample of 300 participants (172/300, 57.33% female; 128/300, 42.67% male, aged 18‐69 years) completed a choice-based conjoint analysis (CBCA). Participants evaluated 15 choice sets describing smart home variants based on cost, location, emergency detection rate, type of sensor, and data processing. Results: Cost was the most important attribute (relative importance [RI]=41%), followed by emergency detection rate (RI=19%), data processing (RI=14%), and location (RI=14%). The type of sensor was the least important attribute (RI=9%). The preferred configuration combined an annual subscription of €70 (US $82), a 95% detection rate, wearable sensors, personalized AI, and installation in both intimate and shared rooms. Notably, 68.3% (205/300) of participants showed a positive none utility, indicating that even the optimal configuration did not overcome general reluctance to adopt such systems. While most expected correlations between sociodemographic variables and attribute importances were not observed, a significant correlation between self-reported health status and emergency detection rate was found (r=.16, P=.007). Interestingly, 61% (167/270) of participants preferred AI over human involvement in data processing, but logistic regression revealed that a significant correlation between self-reported health status and emergency detection rate was found (OR=0.58, P=.04). Conclusions: These findings highlight the need to align smart home development with user preferences, emphasizing cost-effectiveness. Additionally, AI literacy plays an important role in technology adoption in the context of AI-based smart home technology. Further research is needed to understand and address the reluctance to adopt AI for medical emergency detection.}, author = {Jagemann, Inga and Baudisch, Justin and Jungeblut, Thorsten and Maier, Günter W and Hirschfeld, Gerrit}, issn = {2817-1705}, journal = {JMIR AI}, pages = {e94589--e94589}, publisher = {JMIR Publications Inc.}, title = {{Consumer Preferences for AI-Based Smart Home Medical Emergency Detection Among German Adults: Choice-Based Conjoint Analysis}}, doi = {10.2196/94589}, volume = {5}, year = {2026}, } @article{7146, author = {Wehlage, Daria and Göcht, Mia and Ehrmann, Andrea and Andrée, Birgit}, issn = {2688-2663}, journal = {MedComm}, number = {9}, publisher = {Wiley}, title = {{Electrospun Tubing as Shaping Material for Central Vessels in Perfusable, Vascularized Constructs}}, doi = {10.1002/mco2.70985}, volume = {7}, year = {2026}, } @techreport{7152, abstract = {Property prediction for molecules and materials is bottlenecked by label scarcity, and in this regime the operative failure mode is not a low mean but an unpredictable tail: individual training runs collapse. The three dominant molecular representations fail in complementary ways?physicochemical descriptors are exact but fixed, graph neural networks learn topology but degenerate under sparse supervision, and language models supply semantic context but cannot compute exact quantities? yet combining them is itself the learning problem, because modality-level scalar gating commits the whole model to one trust weight per source. We introduce TAME, a tri-modal encoder whose element-wise Mixture-of-Experts router assigns an independent modality mixture to every hidden coordinate, stabilized by a balance?entropy regularizer, a closed-form gate initialization (? = log τ ) that is intended to remove router burn-in, and two-stage self-supervised graph pretraining. On scaffold-split BACE (≈ 1.5k molecules, 100 seeds) we evaluate two fusion topologies, each against its own graph-only control. In both, adding the text and descriptor experts contracts the seed-to-seed distribution of the threshold-free metrics?ROC-AUC σ falls five-fold in the flat topology and 1.6-fold in the hierarchical one, whose control was already the tighter of the two?and the collapse tail reaching ROC-AUC ≈ 0.45 that the weaker single-modality configurations carry is absent from the fused models while accuracy is retained. Each comparison is matched on everything but the fusion stage, isolating its effect. The router allocates a distinct mixture to each coordinate, which no scalar gate can represent, and the entropy weight moves that allocation continuously between graded mixing and a near-binary regime in which each coordinate is claimed outright by one modality. Fused-representation alignment tracks encoder quality without supervision, shifting from text to topology once pretraining lifts the graph embedding out of rank-1 collapse. The same three experts map onto crystal graphs, composition?structure descriptors and literature-mined text. These three modalities recur unchanged across molecular and inorganic chemistry, so TAME is a single interpretable recipe for turning volatile low-data predictors into reliable, self-explaining ones, wherever data are scarce and every failed prediction costs a real experiment.}, author = {Schiller, Robert P. and Weisser, Christoph and Müller, Klaus-Robert and Ochsenfeld, Christian and Semnani, Parastoo}, title = {{TAME: Element-wise Mixture-of-Experts Fusion for Reliable and Interpretable Molecular Property Prediction}}, doi = {10.26434/chemrxiv.15008270/v2}, year = {2026}, } @article{7038, author = {Nadolny, Stephan and Grunwald, Marie and Gröndahl, Annika and Palmdorf, Sarah and Ilskens, Karina and Borchardt, Nina and Maaß, Annika and Schulenkorf, Tessa and Schniering, Stefanie and Afroukhte, Dariusch and Richter, Miriam Tariba and Offermanns, Matthias and Parloh, Ann Kathrin and Grebe, Christian and Latteck, Änne-Dörte}, journal = {Preprint}, publisher = {Elsevier BV}, title = {{Case management and care expertise as a prevention approach for adults with intellectual disabilities: results of a randomized-controlled trial}}, year = {2026}, } @inproceedings{7145, author = {Berlik, Stefan and Seidpisheh, Mohammad}, booktitle = {Integrated Systems: AI-Augmented Engineering}, editor = {Alam, Mohammad-Reza and Fathi, Madjid and Pulatov, Alim}, isbn = {978-3-032-21980-0}, location = {Tashkent, Usbekistan}, pages = {46--56}, publisher = {Springer Nature Switzerland}, title = {{Multi-stage Knowledge Graph-Augmented LLMs for Reliable Product Configuration}}, doi = {10.1007/978-3-032-21981-7_4}, year = {2026}, } @inproceedings{7144, abstract = {The increasing complexity of product configurations demands intelligent systems that effectively integrate customer requirements, dependencies, and uncertainties. This paper introduces a unified framework combining Knowledge Graphs (KGs) and Bayesian Networks (BNs) to enhance the efficiency and adaptability of product configuration processes. KGs provide a semantic foundation for product information, ensuring interoperability and explicit relationship modeling. BNs enhance this through probabilistic reasoning, allowing the system to manage uncertainties and dynamically generate optimal configurations. The integration of deterministic, rule-based reasoning from ontologies with the probabilistic nature of BNs automates suggestions, predicts user preferences, and reduces complexity. This framework streamlines user interactions through intelligent form pre-filling and contextually relevant suggestions, even under uncertainty. By employing an ontology-based representation of BNs, the components fit seamlessly into the KG, creating a cohesive and unified framework that balances scalability and user-centric design to address modern configuration challenges. }, author = {Berlik, Stefan and Seidpisheh, Mohammad}, booktitle = {New Paradigms for Anticipated Uncertainty. Proceedings of the 10th Changeable, Agile, Reconfigurable and Virtual Production Conference (CARV 2025) and the 12th World Mass Customization and Personalization Conference (MCPC 2025), Siegen, Germany, September 2025}, editor = {Manns, Martin}, isbn = {978-3-032-16888-7}, issn = {2195-4364}, location = {Siegen}, pages = {131--141}, publisher = {Springer Nature Switzerland}, title = {{A Unified Framework for Intelligent Product Configuration Using Knowledge Graphs and Bayesian Networks}}, doi = {10.1007/978-3-032-16889-4_13}, year = {2026}, }