Selected work

Representative Papers

MR-Compare: A Mixed-Reality Framework for Spatially Grounded Visual Comparison of 3D Gaussian Splatting and Mesh Reconstructions with the Physical Environment

Jul 22, 2026

This work addresses the lack of effective methods for spatially aligning and visually comparing 3D Gaussian Splatting (3DGS) reconstructions with mesh-based results in real-world environments. We propose MR-Compare, the first cross-modal 3D reconstruction comparison framework supporting spatial anchoring, built upon the Meta Quest 3’s video passthrough mixed reality platform. Our approach integrates a two-stage registration pipeline enhanced by a zero-shot anisotropic filter to improve the robustness of 3DGS alignment and introduces an interactive 3D slider for real-time visual comparison. Experiments demonstrate centimeter-level registration accuracy in real indoor scenes, validating the superiority of the 3DGS-MCMC workflow, while user studies confirm the system’s high usability and low cognitive load.

0 citationsRead paper

U-Net-Accelerated Quality-Diversity Optimization for Climate-Adaptive Urban Layouts

Jun 03, 2026

This study addresses the challenge of optimizing climate-resilient urban layouts by balancing building density and cool-air ventilation, a task traditionally hindered by the high computational cost of high-fidelity physical simulations that limits design-space exploration. To overcome this, the authors propose embedding a U-Net deep learning surrogate model—leveraging its spatial inductive bias—into an offline MAP-Elites quality-diversity optimization framework. Trained solely on a one-time Sobol random sampling dataset, the surrogate achieves highly accurate prediction of microclimate responses (R² = 0.996) without requiring costly active sampling. The approach maintains strong rank-order consistency in layout fitness (Spearman’s ρ = 0.994) and enables the rapid generation of thousands of diverse, high-performing urban design alternatives within minutes, substantially enhancing both optimization efficiency and generalization capability.

0 citationsRead paper

COFFAIL: A Dataset of Successful and Anomalous Robot Skill Executions in the Context of Coffee Preparation

Apr 20, 2026

Existing robotic manipulation datasets are often limited to successful executions or single skills, lacking coverage of anomalous behaviors and thus hindering the development of robust policy learning. To address this gap, this work introduces COFFAIL, a novel dataset collected in a real kitchen environment using a physical dual-arm robot performing coffee-making tasks. COFFAIL encompasses multimodal recordings of diverse manipulation skills executed both successfully and under various failure conditions, along with coordinated bimanual actions—all situated within a unified task context. This dataset is the first to jointly capture multi-skill execution, multiple outcome types (success and anomalies), and bimanual coordination in a realistic setting. Experimental results demonstrate that imitation learning policies trained on COFFAIL exhibit strong effectiveness and generalization capabilities.

0 citationsRead paper

A Study of Performance and Interaction Patterns in Hand and Tangible Interaction in Tabletop Mixed Reality

Nov 14, 2025

This study investigates performance differences between hand-based and tangible interaction for manipulating virtual 3D objects with four degrees of freedom (3D translation + 1D rotation) in desktop mixed reality. We designed a cylindrical tangible proxy integrating a physical knob to support translation, rotation, and scaling, and compared both modalities across isolated and composite manipulation tasks, measuring accuracy, efficiency, and correction behavior. Results show that tangible interaction—leveraging desktop support and structural constraints—significantly improves precision, particularly in composite tasks, where error growth remains markedly lower than with gesture-based interaction. Although hand-based interaction yields marginally smaller rotational errors, it requires frequent corrective actions, undermining overall stability. This work is the first to systematically uncover the mechanistic basis for tangible proxies’ accuracy advantages in desktop MR multi-DOF manipulation, providing empirical evidence and design guidelines for principled interaction modality selection in mixed reality interfaces.

0 citationsRead paper

GG-BBQ: German Gender Bias Benchmark for Question Answering

Jul 22, 2025

This study addresses gender bias in German large language models (LLMs) on question-answering tasks, a critical yet underexplored issue due to German’s grammatical gender system. Method: We construct the first high-quality, German-specific bias evaluation benchmark, designed explicitly for grammatical gender phenomena. The dataset comprises two subsets—group terms and proper names—generated by translating English templates and rigorously refined by native German linguists to avoid machine-translation artifacts. We employ a question-answering evaluation framework to quantify both accuracy and gender bias across multiple German LLMs. Contribution/Results: All evaluated models exhibit significant gender bias—some amplifying, others contradicting societal stereotypes—revealing systemic fairness deficiencies in current German LLMs. This work establishes the first standardized, grammar-aware methodology for bias evaluation in German, providing a reproducible paradigm and publicly available benchmark for fairness research in non-English LLMs.

0 citationsRead paper

Recent publications

Latest Papers

MR-Compare: A Mixed-Reality Framework for Spatially Grounded Visual Comparison of 3D Gaussian Splatting and Mesh Reconstructions with the Physical Environment

Jul 22, 2026

This work addresses the lack of effective methods for spatially aligning and visually comparing 3D Gaussian Splatting (3DGS) reconstructions with mesh-based results in real-world environments. We propose MR-Compare, the first cross-modal 3D reconstruction comparison framework supporting spatial anchoring, built upon the Meta Quest 3’s video passthrough mixed reality platform. Our approach integrates a two-stage registration pipeline enhanced by a zero-shot anisotropic filter to improve the robustness of 3DGS alignment and introduces an interactive 3D slider for real-time visual comparison. Experiments demonstrate centimeter-level registration accuracy in real indoor scenes, validating the superiority of the 3DGS-MCMC workflow, while user studies confirm the system’s high usability and low cognitive load.

0 citationsRead paper

U-Net-Accelerated Quality-Diversity Optimization for Climate-Adaptive Urban Layouts

Jun 03, 2026

This study addresses the challenge of optimizing climate-resilient urban layouts by balancing building density and cool-air ventilation, a task traditionally hindered by the high computational cost of high-fidelity physical simulations that limits design-space exploration. To overcome this, the authors propose embedding a U-Net deep learning surrogate model—leveraging its spatial inductive bias—into an offline MAP-Elites quality-diversity optimization framework. Trained solely on a one-time Sobol random sampling dataset, the surrogate achieves highly accurate prediction of microclimate responses (R² = 0.996) without requiring costly active sampling. The approach maintains strong rank-order consistency in layout fitness (Spearman’s ρ = 0.994) and enables the rapid generation of thousands of diverse, high-performing urban design alternatives within minutes, substantially enhancing both optimization efficiency and generalization capability.

0 citationsRead paper

COFFAIL: A Dataset of Successful and Anomalous Robot Skill Executions in the Context of Coffee Preparation

Apr 20, 2026

Existing robotic manipulation datasets are often limited to successful executions or single skills, lacking coverage of anomalous behaviors and thus hindering the development of robust policy learning. To address this gap, this work introduces COFFAIL, a novel dataset collected in a real kitchen environment using a physical dual-arm robot performing coffee-making tasks. COFFAIL encompasses multimodal recordings of diverse manipulation skills executed both successfully and under various failure conditions, along with coordinated bimanual actions—all situated within a unified task context. This dataset is the first to jointly capture multi-skill execution, multiple outcome types (success and anomalies), and bimanual coordination in a realistic setting. Experimental results demonstrate that imitation learning policies trained on COFFAIL exhibit strong effectiveness and generalization capabilities.

0 citationsRead paper

A Study of Performance and Interaction Patterns in Hand and Tangible Interaction in Tabletop Mixed Reality

Nov 14, 2025

This study investigates performance differences between hand-based and tangible interaction for manipulating virtual 3D objects with four degrees of freedom (3D translation + 1D rotation) in desktop mixed reality. We designed a cylindrical tangible proxy integrating a physical knob to support translation, rotation, and scaling, and compared both modalities across isolated and composite manipulation tasks, measuring accuracy, efficiency, and correction behavior. Results show that tangible interaction—leveraging desktop support and structural constraints—significantly improves precision, particularly in composite tasks, where error growth remains markedly lower than with gesture-based interaction. Although hand-based interaction yields marginally smaller rotational errors, it requires frequent corrective actions, undermining overall stability. This work is the first to systematically uncover the mechanistic basis for tangible proxies’ accuracy advantages in desktop MR multi-DOF manipulation, providing empirical evidence and design guidelines for principled interaction modality selection in mixed reality interfaces.

0 citationsRead paper

GG-BBQ: German Gender Bias Benchmark for Question Answering

Jul 22, 2025

This study addresses gender bias in German large language models (LLMs) on question-answering tasks, a critical yet underexplored issue due to German’s grammatical gender system. Method: We construct the first high-quality, German-specific bias evaluation benchmark, designed explicitly for grammatical gender phenomena. The dataset comprises two subsets—group terms and proper names—generated by translating English templates and rigorously refined by native German linguists to avoid machine-translation artifacts. We employ a question-answering evaluation framework to quantify both accuracy and gender bias across multiple German LLMs. Contribution/Results: All evaluated models exhibit significant gender bias—some amplifying, others contradicting societal stereotypes—revealing systemic fairness deficiencies in current German LLMs. This work establishes the first standardized, grammar-aware methodology for bias evaluation in German, providing a reproducible paradigm and publicly available benchmark for fairness research in non-English LLMs.

0 citationsRead paper