Selected work

Representative Papers

Simulation-Based Performance Evaluation of 3D Object Detection Methods with Deep Learning for a LiDAR Point Cloud Dataset in a SOTIF-related Use Case

Mar 05, 2025International Conference on Vehicle Technology and Intelligent Transport Systems

Current SOTIF (Safety of the Intended Functionality) validation for autonomous driving lacks systematic assessment of LiDAR perception robustness under adverse weather conditions. Method: This work formally defines and models SOTIF-critical use cases, and constructs a high-fidelity simulated LiDAR point cloud dataset—comprising 547 frames across 21 weather-illumination combinations—using CARLA and PreScan. Within the MMDetection3D and OpenPCDet frameworks, it conducts a comprehensive performance evaluation of state-of-the-art 3D object detectors using AP and Recall metrics for cross-model comparison. Contribution/Results: Results reveal significant performance degradation under rain, fog, and low-light conditions, with AP dropping by up to 62%. To our knowledge, this is the first study to provide a reproducible, quantitative benchmark for evaluating LiDAR perception robustness in SOTIF contexts. The dataset and empirical findings bridge a critical gap in SOTIF validation, supporting algorithmic refinement and standardization efforts.

2 citationsRead paper

Design and Evaluation of a Touchscreen-Based Teleoperation Interface for Robotic Manipulators

Aug 06, 2026

This study addresses the limitations of conventional joystick-based teleoperation in scenarios such as nuclear industry applications, where precise path tracking, force control, and obstacle avoidance during surface contact tasks are critical, yet impose high cognitive load on operators. To overcome these challenges, this work proposes a touchscreen-based teleoperation interface that directly maps continuous finger motion to the Franka Emika Panda robotic arm, integrating control and visualization for more intuitive motion mapping and fine-grained velocity modulation. User studies demonstrate that, compared to both joystick and one-button autonomous modes, the proposed method reduces median task completion time by 53.5% (2.50 vs. 5.38 minutes), achieves a 90.7% coverage rate on sinusoidal paths with lower overshoot, and decreases NASA-TLX cognitive workload scores by 17.3%, significantly enhancing operational efficiency and naturalness.

0 citationsRead paper

Robot Learning from Human Demonstrations: Handwritten Alphabet Trajectories and Human-Likeness Evaluation

Aug 06, 2026

This work proposes a framework for learning human-like motion trajectories from human demonstrations to enhance trust and naturalness in human-robot collaboration. Leveraging a touch-based teleoperation interface, the approach captures position, contact force, and temporal data, which are integrated into an extended Gaussian Mixture Model (GMM) that incorporates both force and normalized time dimensions, enabling modeling of discontinuous, multi-segment trajectories. Smooth and dynamically consistent robot trajectories are then generated via Gaussian Mixture Regression (GMR). A user study demonstrates the efficacy of the method, with synthesized trajectories scoring 71.50 out of 100 on human-likeness; geometric position and trajectory sequencing were identified as key perceptual factors, confirming the framework’s expressive capability and validity.

0 citationsRead paper

Trajectory-Level Data Augmentation for Offline Reinforcement Learning

May 13, 2026

This work addresses the challenges of data scarcity and suboptimal quality in offline reinforcement learning, which arise from reliance on a limited number of imperfect trajectories. To mitigate these issues, the paper proposes a trajectory-level data augmentation method that leverages the geometric relationships among the reward function, value function, and behavior policy. By incorporating the intrinsic geometry of the task, the approach remains compatible with suboptimal data-collecting policies and provides theoretical justification for trajectory augmentation. This is the first study to integrate trajectory-level augmentation with task-specific geometric structure, demonstrating significant improvements in both performance and data efficiency across a range of high-dimensional and partially observable navigation tasks.

0 citationsRead paper

Feasibility of AI-Assisted Programming for End-User Development

Dec 05, 2025

Non-programmer end users face limitations in development capability and vendor lock-in when using low-code/no-code (LCNC) platforms. Method: This study investigates generative AI–assisted end-user development (AI-TUD) as an alternative or complementary paradigm, proposing a large language model (LLM)–centric, natural language–driven conversational programming framework that enables users to generate executable code directly from natural language instructions. Contribution/Results: Experiments demonstrate that most non-professional users can successfully build basic web applications within reasonable timeframes. Compared to conventional LCNC platforms, AI-TUD significantly enhances development flexibility, task coverage breadth, and technological neutrality while mitigating platform dependency risks. The study empirically validates AI-TUD’s practical efficacy and scalability in end-user development, providing theoretical foundations and empirical evidence for next-generation human–AI collaborative software engineering.

0 citationsRead paper

Recent publications

Latest Papers

Design and Evaluation of a Touchscreen-Based Teleoperation Interface for Robotic Manipulators

Aug 06, 2026

This study addresses the limitations of conventional joystick-based teleoperation in scenarios such as nuclear industry applications, where precise path tracking, force control, and obstacle avoidance during surface contact tasks are critical, yet impose high cognitive load on operators. To overcome these challenges, this work proposes a touchscreen-based teleoperation interface that directly maps continuous finger motion to the Franka Emika Panda robotic arm, integrating control and visualization for more intuitive motion mapping and fine-grained velocity modulation. User studies demonstrate that, compared to both joystick and one-button autonomous modes, the proposed method reduces median task completion time by 53.5% (2.50 vs. 5.38 minutes), achieves a 90.7% coverage rate on sinusoidal paths with lower overshoot, and decreases NASA-TLX cognitive workload scores by 17.3%, significantly enhancing operational efficiency and naturalness.

0 citationsRead paper

Robot Learning from Human Demonstrations: Handwritten Alphabet Trajectories and Human-Likeness Evaluation

Aug 06, 2026

This work proposes a framework for learning human-like motion trajectories from human demonstrations to enhance trust and naturalness in human-robot collaboration. Leveraging a touch-based teleoperation interface, the approach captures position, contact force, and temporal data, which are integrated into an extended Gaussian Mixture Model (GMM) that incorporates both force and normalized time dimensions, enabling modeling of discontinuous, multi-segment trajectories. Smooth and dynamically consistent robot trajectories are then generated via Gaussian Mixture Regression (GMR). A user study demonstrates the efficacy of the method, with synthesized trajectories scoring 71.50 out of 100 on human-likeness; geometric position and trajectory sequencing were identified as key perceptual factors, confirming the framework’s expressive capability and validity.

0 citationsRead paper

Trajectory-Level Data Augmentation for Offline Reinforcement Learning

May 13, 2026

This work addresses the challenges of data scarcity and suboptimal quality in offline reinforcement learning, which arise from reliance on a limited number of imperfect trajectories. To mitigate these issues, the paper proposes a trajectory-level data augmentation method that leverages the geometric relationships among the reward function, value function, and behavior policy. By incorporating the intrinsic geometry of the task, the approach remains compatible with suboptimal data-collecting policies and provides theoretical justification for trajectory augmentation. This is the first study to integrate trajectory-level augmentation with task-specific geometric structure, demonstrating significant improvements in both performance and data efficiency across a range of high-dimensional and partially observable navigation tasks.

0 citationsRead paper

Feasibility of AI-Assisted Programming for End-User Development

Dec 05, 2025

Non-programmer end users face limitations in development capability and vendor lock-in when using low-code/no-code (LCNC) platforms. Method: This study investigates generative AI–assisted end-user development (AI-TUD) as an alternative or complementary paradigm, proposing a large language model (LLM)–centric, natural language–driven conversational programming framework that enables users to generate executable code directly from natural language instructions. Contribution/Results: Experiments demonstrate that most non-professional users can successfully build basic web applications within reasonable timeframes. Compared to conventional LCNC platforms, AI-TUD significantly enhances development flexibility, task coverage breadth, and technological neutrality while mitigating platform dependency risks. The study empirically validates AI-TUD’s practical efficacy and scalability in end-user development, providing theoretical foundations and empirical evidence for next-generation human–AI collaborative software engineering.

0 citationsRead paper

SAT-sampling for statistical significance testing in sparse contingency tables

Nov 07, 2025

Exact conditional inference for sparse contingency tables—especially those with structural zeros—is hindered by conventional Markov chain Monte Carlo (MCMC) methods based on Markov bases, which often fail due to intractable basis computation and slow convergence. Method: We propose SAT-Driven MCMC: a framework that encodes the fixed-margin fiber as a Boolean satisfiability (SAT) problem and leverages modern SAT samplers to generate global proposals; these are combined with local moves to form a hybrid proposal mechanism. Crucially, this approach eliminates the need for precomputing a Markov basis or assuming graph connectivity, and natively accommodates structural zeros. We further introduce sampling bias diagnostics and correction to ensure convergence to the correct stationary distribution. Results: Experiments on multiple high-dimensional sparse benchmarks demonstrate that our method consistently yields reliable p-values, outperforming existing basis-dependent approaches in both stability and accuracy.

0 citationsRead paper