Keywords: Artificial Intelligence Systems, Robot Vision, Human-Robot Interaction
Abstract: Babies require continuous observation to ensure their safety, yet most existing action recognition systems are ill-suited for recognizing baby-specific behaviors, especially in real-world home environments with low illumination and visual clutter. In this study, we propose a unified vision-based baby behavior recognition system that supports both infant and toddler stages. Our model is trained on a custom video dataset collected from in-the-wild YouTube footage, capturing diverse lighting conditions and baby-specific actions, including high-risk behaviors such as face-down posture and chewing toys. To enhance robustness under poor lighting, we simulate low-light conditions and apply a lightweight enhancement model to recover visual clarity. We adopt a transformer-based VideoMAE V2 backbone to learn spatiotemporal features and perform behavior classification. The proposed system achieved an overall accuracy of 83.46% across 8 behavior classes under low-light conditions. We further validated the system’s applicability by connecting it to a mobile robot in a simulated environment using a baby doll, confirming its feasibility for real-time robotic monitoring. These results suggest that the system can serve as a foundation for autonomous baby monitoring in real-world settings.

Keywords: Artificial Intelligence Systems, Robot Vision, Sensors and Signal Processing
Abstract: This study investigates the feasibility of semi-autonomously building a three-dimensional model of an apple orchard using an omnidirectional camera and a laser scanner mounted on a backpack. Data was collected at an apple orchard in Japan using the LiBackpack C50, capturing 360-degree images and point clouds. We evaluated several types of methods that would make up a digital twin system. Camera localization was performed using a visual SLAM-based algorithm, and three-dimensional reconstruction was performed by UniK3D, a method that takes takes a single omnidirectional image as input. The performance of these models was assessed using images and trajectory data from our dataset. Camera pose estimation showed limited movement due to the lack of distinct features in the orchard. The 3D reconstruction results showed good point cloud generation, with errors remaining in the camera field of view estimation. This research aims to contribute to digital twin applications in smart farming, enabling rapid development and testing of strategies to mitigate climate change impacts on apple production.

Keywords: Artificial Intelligence Systems
Abstract: The increasing demand for dental services highlights the need for efficient post-operative oral surgery consultations. Many patients experience anxiety due to limited knowledge of oral care and treatment. This study introduces a chatbot prototype integrating Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs) to provide accurate, context-aware responses. The research evaluates various LLMs, embedding models, and chunking techniques to enhance chatbot performance. The multilingual-e5-large embedding model excelled in retrieval tasks due to its multilingual training, instruction tuning, and contrastive pre-training, ensuring high retrieval precision. The Hybrid Chunking method was selected for its ability to segment text contextually, combining Markdown-based, token-based, and semantic segmentation for optimal chunk relevance. The Llama3.3 (70B) model was chosen for its superior fluency, relevance, and ability to handle complex dependencies. The results demonstrate that combining the multilingual-e5-large embedding model, Hybrid Chunking technique, and Llama3.3 (70B) model improves retrieval precision, response accuracy, and relevance, enhancing patient care and operational effectiveness of dental staffs.

Keywords: Artificial Intelligence Systems, Control Theory and Applications, Robotic Applications
Abstract: The performance of Nonlinear Model Predictive Control (NMPC) is fundamentally coupled to the fidelity of its internal predictive model. While neural networks can serve as universal function approximators for system dynamics, they typically require large datasets to avoid poor generalization, making them computationally heavy to be used for control systems. We propose using a Small Lagrangian Network (SLN) as the prediction model within an NMPC framework. An SLN is a compact, physics-informed neural network that learns a system's scalar Lagrangian instead of its dynamics directly. This structural prior imbeds the Lagrangian Mechanics into the model for easier generalization to the conservation of energy. We demonstrate that we can distillate an Lagrangian Neural Network (or even other physics-informed models) to be more compact and lightweight while keeping its state estimation accuracy for NMPC framework. This shows potentials for physics-informed surrogates can be much smaller in scale and still perform robustly for controls.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control
Abstract: This paper presents a real-time detection system to enhance industrial automation in cup-filling processes by identifying the presence of plastic spoons using Deep Learning and Programmable Logic Controller (PLC) integration. The proposed system employs the YOLO (You Only Look Once) object detection model to process images captured by a vision camera. The model runs on a PC and communicates with a BECKHOFF PLC to perform automated decision-making based on detection results. A hardware simulation environment was developed, consisting of a servo-driven rotary table and camera setup, mimicking an actual production line. The system detects two classes: cups with spoons (acceptable) and cups without spoons (defective). Upon detection, the result is sent to the PLC, which initiates appropriate control actions. Experimental testing demonstrates that the proposed AI-integrated automation system improves detection accuracy, reduces errors from conventional sensors, and ensures better product quality control in real-time operations.

Keywords: Artificial Intelligence Systems, Navigation, Guidance and Control, Robot Vision
Abstract: This paper presents a centerline detection strategy for unstructured highway pavements using UAV imagery and a hybrid deep learning framework. The method integrates yellow line and slab-based lane segmentation within a hierarchical decision-making algorithm for robust centerline estimation. To improve accuracy and efficiency, image patching is applied by analyzing only a portion of each frame. The approach was evaluated on two curved highway segments and compared with manual and waypoint-based navigation. Results show the proposed method achieved the lowest path estimation errors—0.074% for Curve 1 and 1.765% for Curve 2—outperforming manual (2.791%, 3.367%) and waypoint methods (1.856%, 19.852%). These findings demonstrate its effectiveness for accurate and consistent centerline detection, with potential applications in UAV-based inspection and autonomous navigation

Keywords: Multimedia Systems, Human-Robot Interaction, Robotic Applications
Abstract: In the evolving landscape of precision agriculture, the integration of digital twin technologies with robotics offers new possibilities for enhancing plantation management. This study presents a metaverse-based digital twin environment tailored for the oil palm agriculture sector. Utilizing Unreal Engine and Cesium for Unreal, the system integrates high-resolution GIS data, 3D modeling, and immersive simulation to create a realistic and interactive representation of the plantation. The platform features unmanned ground vehicle (UGV) simulation and virtual reality (VR) interaction, providing a comprehensive tool for training, navigation, and operational planning. Rather than focusing on performance benchmarking, this study emphasizes the feasibility and practical application of digital twin technologies combined with robotics in agriculture. The resulting prototype enables terrain condition simulation, interactive tree management, and lays the groundwork for future integration with artificial intelligence (AI) and Internet of Things (IoT) components, marking a significant step toward smart plantation systems.

Keywords: Multimedia Systems, Human-Robot Interaction
Abstract: This study presents three newly developed features in HoloThrow, a virtual reality (VR) educational application designed to support understanding of projectile motion. The added features include: (1) a vector visualization feature that displays both velocity and acceleration vectors to help learners grasp motion dynamics, (2) a trajectory comparison feature that allows visual comparison of multiple projectile conditions, and (3) a tutorial feature that offers step-by-step operational guidance for VR beginners. To evaluate the usability of these features, a questionnaire-based study was conducted with seven graduate students. Participants rated the system highly in terms of ease of operation, visual clarity, and effectiveness in supporting learning. The tutorial feature was especially noted for reducing anxiety during initial use, while the acceleration vector visualization was considered useful for understanding gravitational influence. This paper details the design and implementation of the features and discusses the potential benefits of such learning support tools in VR-based physics education based on the evaluation results.

Keywords: Rehabilitation Robot
Abstract: Life-supporting technology helps elderly and disabled individuals maintain their ability to perform daily activities, such as transferring between a bed and a wheelchair. Caregivers assist care recipients in utilizing their remaining functional abilities more actively. Additionally, caregivers can undergo training to minimize their risk of lower back pain, particularly when performing transfer tasks. However, many institutions face challenges such as workforce shortages, an aging staff, and heavy care workloads. Previous studies have identified excessive pressure on the L5/S1 segment of the spine as a primary cause of lower back pain. The load on this segment is determined by the force exerted by the erector spinae muscles, as well as spinal posture, which is influenced by upper-body posture, the angle between the upper body and thighs, and knee joint angles. The proposed assistive suit features a passive elastic back blade made of carbon fiber-reinforced plastic (CFRP) that supports the caregiver's posture and reduces spinal load. Simulations show that the suit effectively mitigates compressive forces, depending on the load transmission rate. A precise mechanical model was employed to estimate the load on the L5/S1 spinal segment, which was then compared with the load generated by other types of assistive suits.

Keywords: Information and Networking, Navigation, Guidance and Control, Control Theory and Applications
Abstract: This paper proposes the problem of maintaining prescribed received signal strength indicator (RSSI) thresholds for mobile relay networks in indoor disaster environments. The proposed framework combines wireless propagation modeling with distributed model predictive control (DMPC) to ensure robust communication links. On the wireless side, the ITU-R P.1238-12 path-loss model is integrated with material-specific wall attenuation and log-normal shadow fading. Line-of-sight (LoS) and non-line-of-sight (NLoS) classifications are obtained through ray casting on material-annotated maps, providing accurate material-penetration counts. On the control side, a distributed MPC approach is formulated under nonholonomic dynamics constraints. The controller incorporates threshold-triggered hinge communication penalties and soft-consensus variables to coordinate relay motion. The constraint set encompasses system dynamics, actuation limits, and obstacle avoidance while enforcing RSSI lower bounds through nonnegative slack variables to preserve recursive feasibility. A lightweight receding-horizon implementation is developed with stepwise derivation from link budget calculations to RSSI penalty formulation. The simulation framework supports material-map protocols with configurable LoS conditions, wall losses, and fading effects, enabling Monte Carlo analysis. Performance evaluation encompasses coverage metrics, outage probability, minimum RSSI values, path efficiency, and computational runtime. The framework captures essential indoor propagation phenomena while maintaining computational tractability, inducing minimal relay motion to satisfy communication requirements for time-critical disaster response scenarios.

Keywords: Multimedia Systems
Abstract: This study proposes a new history education method called VRVN that combines Virtual Reality (VR) and Visual Novels (VN). Traditional history education in Japan has focused heavily on memorizing dates and events, failing to adequately engage students' interest and curiosity. To address this issue, we aim to create more effective historical learning experiences by combining VR's immersive capabilities with the storytelling techniques of visual novels. Specifically, we developed VRVN content about the Second Choshu expedition using Unity with C# and implemented it on the Meta Quest 3 platform. We conducted evaluation experiments with 13 university students, assessing historical knowledge retention through pre- and post-tests and verifying system effectiveness through subjective evaluations. The experimental results confirmed significant effects on historical knowledge retention, particularly in understanding complex historical contexts. Additionally, the immersive VR experience promoted understanding of historical events with a real sense of presence. Furthermore, the system was effective in improving learning motivation, with many participants responding that they ``wanted to learn more in detail," successfully inspiring continued learning interest.

Keywords: Robotic Applications, Industrial Applications of Control, Autonomous Vehicle Systems
Abstract: This paper presents real-time monitoring, teleoperation, and autonomous navigation control using external human-machine interfaces such as web and mobile applications, along with latency measurements in robotic systems based on ROS 1 and ROS 2. The study evaluates data transmission performance for various sensor and control signals, including LiDAR, RGB camera images, odometry data, and velocity commands, under three scenarios: intra-ROS communication, web-based interaction, and mobile app usage. Experiments were conducted in both simulated environments using Gazebo Classic and Ignition and on real hardware using TurtleBot 3, over local and global networks. Results show that ROS 2 exhibits lower latency compared to ROS 1, and that latency increases when interacting through mobile applications or across global networks, highlighting the influence of device capabilities and communication overhead. Timestamp-based latency measurements were performed using rosbridge, with synchronized clocks in both simulation and real hardware. Initial tests revealed that transmitting high-bandwidth data such as camera streams over rosbridge WebSocket caused significant latency in global networks due to message queuing. To address this, flow control was implemented on the client side by reducing the camera data subscription rate. This work provides practical insights into interface design, system architecture, and communication trade-offs in ROS-based robotic platforms. The experiment video is available at the following link: url{https://youtu.be/PcFoxXZFZgI}.

Keywords: Robot Vision, Sensors and Signal Processing
Abstract: Monocular vision systems are widely used in robotics and autonomous platforms for tasks such as distance estimation, object detection, and semantic understanding. However, these systems are highly sensitive to camera perturbations, such as physical misalignments caused by collisions, vibrations, or mechanical shocks during operation. Such perturbations introduce structural biases into the vision input, which significantly degrade model performance. In this work, we propose a self-adaptive, bias-aware deep learning framework designed to enhance the robustness of vision-based models under camera misalignment conditions. Our approach introduces a lightweight bias modeling module that learns to capture and compensate for systematic deviations caused by camera displacement. While our framework is applicable to various vision tasks, we evaluate it on monocular distance estimation to demonstrate its effectiveness. We construct a hybrid dataset simulating multiple camera perturbation scenarios and quantitatively assess the framework’s robustness and generalization. Experimental results show substantial improvements over baseline models in terms of accuracy and stability under misaligned conditions. This work contributes a generalizable method to improve the resilience of vision-based systems operating in dynamic and unpredictable environments.

Keywords: Robot Vision, Robotic Applications, Industrial Applications of Control
Abstract: This study presents a vision-based approach for estimating surface roughness in automotive press dies. Four feature extraction algorithms are evaluated: Grey Scale Average (GSA), FFT, Sharpness, and Texture Homogeneity. Grayscale images are collected from die surfaces, and each image is paired with a surface roughness value measured using a contact-type profilometer. Single scalar features are extracted by each algorithm and used to train regression models, including linear, polynomial, exponential, and logarithmic functions. Model performance is assessed using R-squared, root mean square error, and relative error. GSA and FFT models show the highest accuracy, with R-squared values near 0.66. These models capture global brightness and frequency-domain patterns, which exhibit stronger correlations with surface variation than local edge sharpness or texture features. The results suggest that global image characteristics are effective for surface roughness estimation. This method supports non-contact surface quality monitoring in die manufacturing, and future research may improve its accuracy and robustness through advanced modeling techniques.

Keywords: Robot Vision, Artificial Intelligence Systems, Multimedia Systems
Abstract: In the boiled egg industry, quality inspection is typically conducted twice: before eggs are transported into the conveyor-based boiling system (before boiling), and after they exit the water-based cooling system prior to packaging (after cooling). These inspections are commonly carried out through human visual assessment, which demands substantial human resources and time. This paper presents an automated system for detecting and classifying defective eggs, such as those with cracks, dents, rough shells, and other surface anomalies using image processing techniques. The system is designed to enhance the visibility of such defects while minimizing the impact of production-related surface disturbances, including water stains, reflections from the cooling process, and red stamps from imported eggs. The proposed system comprises two main approaches: (1) Defects Detection Method, which classifies eggs into two categories: intact and defective; and (2) Pixel Counting and Comparison Method, which classifies eggs into three categories: intact, cracked or dented, and exploded eggs. This system offers a practical and efficient solution for the egg processing industry, reducing reliance on human labor, minimizing inspection time, and lowering hardware requirements for industrial implementation.

Keywords: Robot Vision, Artificial Intelligence Systems, Robotic Applications
Abstract: Object detection in disaster images is often difficult because of class imbalance and a lack of labeled samples. To solve this problem, we introduce Semantic-Guided Generative Augmentation (SGGA), a new method that uses semantic masks to generate more samples for the rare classes. SGGA creates new images by changing clean road areas into texttt{small Road-Blocked} areas using mask-based sampling and prompt-guided inpainting, making sure the new objects appear in the right places. We filter the new images using CLIP similarity and LPIPS distance, ensuring high semantic and visual quality. Experiments on the RescueNet dataset show that SGGA improves texttt{small Road-Blocked} detection by +26.2% mAP@0.5 and +29.7% recall, beating other augmentation methods. Furthermore, t-SNE analysis confirms strong semantic alignment between real and SGGA-generated images. SGGA offers significant advantages, including spatial precision, contextual realism, and low annotation overhead, making it particularly suitable for practical deployment in disaster scenarios and other domains where spatial priors are available.

Keywords: Robot Vision, Artificial Intelligence Systems, Robotic Applications
Abstract: Large language models (LLMs) trained on code-completion can create basic Python programs from docstrings. Given commands in normal language, we discover that these code-writing LLMs may be repurposed to produce robot policy code. For semantic grounding, we propose a modular framework that combines a vision language model(CLIP) with real time object detection (YOLOv8).Our method creates practical action plans that direct a robotic arm to do the requested task and incorporates a pretrained language model for path planning. The framework pipeline in both simulated and real-world settings. Our approach has a 100% success rate in cube interaction and is good generalized to new object categories like rubber duck and marker with success rates between 85% and 50%. On actual hardware environment, the system consistently completes task with an average success rate of 70%. These outcomes show how our pipeline bridges perception, language and control for general robot task and how robust and practically applicable it is.

Keywords: Robot Vision, Sensors and Signal Processing
Abstract: We propose a method to extract objects embedded in flexible materials, such as cushions, using solely 3D point‑cloud data. Conventional plane detection based on RANSAC has difficulty accurately isolating objects that contact curved cushion surfaces. In our approach, we first apply eigenvalue analysis on the covariance matrix to remove edge points, which is followed by DBSCAN clustering. Specifically, a 3×3 covariance matrix is computed from the k‑nearest neighbors of each point, and three eigenvalues are used to calculate a surface variation. We then compute the variance of local curvature, which allows us to distinguish edges from flat, curved, and rough surfaces. Removing these edges creates a clear boundary between the object and the flexible material, which facilitates effective cluster separation. We evaluated our method with experiments that used multiple objects and one participant. Measurements were taken from three viewing angles and averaged to assess robustness. The method achieved high Precision, Recall, and F1‑score.

Keywords: Navigation, Guidance and Control, Civil and Urban Control Systems, Information and Networking
Abstract: Navigation of UAVs in unknown environments with obstacles is essential for applications in disaster response and infrastructure monitoring. However, existing obstacle avoidance algorithms, such as Artificial Potential Field (APF), are unable to generalize across environments with different obstacle configurations. Furthermore, the precise location of the final target may not be available in applications such as search and rescue, in which case approaches such as RF source seeking can be used to align towards the target location. This paper proposes a real-time trajectory planning method, which involves real-time adaptation of APF through a sampling-based approach. The proposed approach utilizes only the bearing angle of the target without its precise location, and adjusts the potential field parameters according to the environment with new obstacle configurations in real time. The main contributions of the article are i) RF source seeking algorithm to provide a bearing angle estimate using AoA based on arbitrary antenna placement which can be generalized to arbitrary number of antennas, and ii) modified APF for adaptable collision avoidance in changing environments, which are evaluated separately in the simulation software Gazebo, using ROS2 for communication. Simulation results show that the RF source seeking algorithm achieves high accuracy, with an average angular error of just 1.48 degrees, and with this estimate, the proposed navigation algorithm improves the success rate of reaching the target by 46% and reduces the trajectory length by 1.2% compared to standard potential fields. However, the computation time is increased significantly, which can be improved with more efficient code implementations.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: End-to-end navigation methods have emerged as promising alternatives to traditional approaches in mobile robot navigation, thanks to their ability to handle obstacles of arbitrary shapes and unbounded categories. However, these methods often require a large number of training steps due to the complexity of interpreting high-dimensional sensor inputs. To address this limitation, we propose a shape-aware end-to-end navigation system equipped with a neural encoder specifically designed for 3D point clouds. By incorporating batch normalization and sparse convolution, the encoder enhances both sample efficiency and training speed. Our system achieves superior navigation performance compared to existing methods, while also significantly accelerating the training process.

Keywords: Navigation, Guidance and Control, Control Theory and Applications, Robotic Applications
Abstract: Classical map-based approaches to autonomous racing are known for their high performance, as they can compute globally optimal reference trajectories for the vehicle to follow. However, these methods are limited to pre-mapped environments, which restricts their applicability in fully autonomous systems since it must operate in inherently unknown or unmapped settings. In contrast, mapless racing approaches eliminate the need for a global map by relying solely on sensor data to construct a local representation of the environment. Despite this advantage, existing mapless methods often fall short in terms of performance and struggle to maintain high speeds. To overcome this limitation, this paper propose a receding horizon minimum-curvature based trajectory generation and tracking that operates within local map framework, enabling high performance racing using only local observations and removing the reliance on global reference. Both Gazebo simulations and real-world experiments demonstrate that the proposed mapless racing method achieves better performance compared to the other two mapless methods. Racing videos in gazebo and real world experiment can be found in Project page.

Keywords: Navigation, Guidance and Control, Robotic Applications, Autonomous Vehicle Systems
Abstract: Sidewalks present various obstacles beyond pedestrians, such as trees and curbs, making it challenging for users to maneuver a personal mobility vehicle (PMV) toward a destination while avoiding these obstacles. This paper focuses on driving assistance provided by the PMV itself. Safe driving assistance requires the PMV to generate control commands, such as velocity and angular velocity, while considering its surrounding environment. To achieve this, we propose a shared controller that integrates user input with PMV's autonomy. Equipped with an RGB camera, the PMV employs semantic segmentation to identify drivable road surfaces free of obstacles between curbs. Within the shared controller, attractive and repulsive forces are generated based on the user's joystick input and the road segmentation results. The resultant force is then applied to the PMV to assist in navigation. Through driving assistance experiments, we demonstrate that a PMV utilizing the proposed shared controller can successfully navigate toward its destination while avoiding collisions with obstacles and curbs. The results confirm the effectiveness of shared-control-based driving assistance with semantic road segmentation.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems, Robotic Applications
Abstract: Achieving fully autonomous exploration and navigation remains a critical challenge in robotics, requiring integrated solutions for localisation, mapping, decision-making and motion planning. Existing approaches either rely on strict navigation rules lacking adaptability or on pre-training, which requires large datasets. These AI methods are often computationally intensive or based on static assumptions, limiting their adaptability in dynamic or unknown environments. This paper introduces a bio-inspired agent based on the Active Inference Framework (AIF), which unifies mapping, localisation, and adaptive decision-making for autonomous navigation, including exploration and goal-reaching. Our model creates and updates a topological map of the environment in real-time and plans goal-directed trajectories to explore or reach objectives without pre-training. Key contributions include a probabilistic reasoning framework for interpretable navigation, robust adaptability to dynamic changes, and a modular ROS2 architecture compatible with existing navigation systems. Our method was tested in simulated and real-world environments. The agent successfully explores large-scale simulated environments and adapts to dynamic obstacles and drift, proving to be comparable to other exploration strategies such as Gbplanner, FAEL and Frontiers. This approach offers a scalable and transparent approach for navigating complex, unstructured environments.

Keywords: Navigation, Guidance and Control
Abstract: Path planning is a fundamental and challenging problem in mobile robotics. Although sampling-based and graph-based algorithms have been widely adopted, they often rely on dense environmental representations. In this paper, we propose the Polygonal Proximity Map which is a pose-graph map that implicitly stores convex free space, and define a geometric condition for path validation using this map. Based on this condition, we develop path planning algorithms without explicit polygonal representations. In the experiments, we demonstrate global path planning by comparing the proposed algorithms with Navigation2 package of ROS2. Experimental results show that our approach achieves superior performance in terms of path length, planning time, and success rate.

Keywords: Robot Mechanism and Control, Robotic Applications, Control Theory and Applications
Abstract: Mobile robots performing outdoor last-mile delivery experience multi-degrees-of-freedom (MDOF) vibrations that cause payload damage. Conventional chassis suspensions cannot effectively isolate such MDOF vibrations. This paper proposes a dual-stage MDOF vibration isolator (VI) system that utilizes only three actuators to isolate five-DOF vibrations. Although three actuators cannot fully isolate all five-DOF vibrations, the proposed design still reduces the magnitude in each DOF. Dynamic analysis shows that payload vibrations partition into three nearly decoupled groups, each controlled independently by the isolator output forces and torques. The upper stage uses one actuator to produce torque for roll isolation; the lower stage uses two actuators to provide force for vertical isolation and torque for pitch isolation. A disturbance observer (DOB)-based feedback-feedforward controller independently isolates pitch and roll vibrations, and a skyhook controller isolates vertical vibrations. Experimental results under various driving conditions demonstrate root mean square vibration reductions of about 79% vertically, 28% longitudinally, and 57% in roll compared to conventional chassis suspension. Furthermore, open-beverage payload tests confirmed liquid-loss reductions of at least 70% (peaking at 85.6%), validating improved payload stability and operational reliability for mobile robots.

Keywords: Robot Mechanism and Control, Robotic Applications
Abstract: This study proposes a global path guided model predictive path integral (MPPI) control for manipulator trajectory tracking. MPPI offers robust and sample-efficient performance, but it faces local minima issues and high computational demands due to obstacles and constraints. To address these problems, we first apply the global path planning algorithm to generate a trajectory from the current location to the target location. Next, a tunnel is generated around the path. The size of the tunnel is based on the safety of the distance from obstacles. By pre-configuring the global path, this method mitigates the local minima problem for MPPI. Furthermore, by utilizing the tunnel to define the search space, the computational efficiency for local path generation is enhanced. MPPI performs calculations on samples corresponding to the tunnel interior, thus deriving the optimal local path. Through simulation, we validate the performance of the proposed algorithm, demonstrating its efficiency in addressing both local minima and computational load issues.

Keywords: Robot Mechanism and Control, Robotic Applications, Human-Robot Interaction
Abstract: Due to natural disasters and industrial accidents, there is currently a critical need for innovative technologies in disaster response and monitoring. Existing technologies in disaster robotics show limitations in navigating debris piles, uneven terrain, and high steps. In order to address this, a new approach in the design and development of a tracked robot is proposed. The flippers, designed to be at the center of its body, also function as sub-crawlers to provide additional support in navigating obstacles. A center of mass (CoM) analysis validated this mechanical design, indicating a balanced and stable robot. The tracked robot, measuring 704 × 400 × 106 mm and weighing 6.2 kg, demonstrated strong stability and traction in climbing a 5-step staircase with a 45◦ slope. Moreover, a comparative analysis was done against existing tracked robot designs. It ranked second in mass density (208 kg/m3 ) and achieved a competitive maximum step of 30 cm, while having the lowest minimum clearance of 11 cm among all other models. Thus, the tracked robot shows potential for exploration in post-disaster environments, capable of navigating through unpredictable terrain.

Keywords: Robot Mechanism and Control, Robotic Applications
Abstract: This paper presents bilinear force control scheme with time-delayed control method to guarantee desired force tracking control under uncertainties. Although the bilinear force control scheme guarantees the desired force tracking, it fails under uncertainties presence. Thus the time-delayed control scheme is introduced to compensate for uncertainties. Combining two control schemes guarantee the force tracking when uncertainties are present. Simulation results of force tracking control performance for three link robot manipulators are provided to confirm the proposal.

Keywords: Robot Mechanism and Control, Biomedical Instruments and Systems, Robotic Applications
Abstract: In recent years, magnetic soft continuum robots (MSCRs) that can be remotely manipulated using external magnetic fields and have high flexibility have been attracting attention as a promising technology in the field of minimally invasive surgery (MIS), and the Asymmetric Magnetized Soft Continuum Robot (AMSCR) has been studied as one of them. AMSCRs have a relatively large workspace compared to conventional MSCRs, but there is a lack of methodology for the magnetization of magnetic materials in AMSCRs and quantitative analysis of the workspace. In this study, we aim to overcome these limitations and propose an optimal magnetization pattern design method through energy-based kinematic analysis of linearly asymmetrically magnetized AMSCRs.

Keywords: Control Theory and Applications, Industrial Applications of Control, Sensors and Signal Processing
Abstract: This paper presents an optimal proportional-integral (PI) disturbance observer design for relative degree-one systems, formulated as a quadratically constrained convex optimization problem (QCQP), and demonstrates its application to a surface-mounted permanent magnet synchronous motor (SPMSM) back-electromotive force (EMF) estimation system. By deriving the transfer function between actual and estimated back-EMF, explicit pole-zero relationships can be expressed in terms of observer gains. Performance requirements, such as bandwidth and attenuation levels of the observer system, are translated into convex equality and inequality constraints. A phase-delay-minimizing cost function is introduced to select the gain that minimizes phase delay within the feasible solution set.

Keywords: Control Theory and Applications, Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: This paper presents a reachability-guided control framework for nonlinear systems that synthesizes pseudo-optimal policies using only local linear models. At each step, a forward reachable tube (FRT) is computed via zonotope-based set propagation; the closest point in the FRT to the target is chosen as an intermediate waypoint, around which a backward reachable tube (BRT) is solved using Hamilton–Jacobi reachability. The resulting value function yields a locally optimal control action. This process is repeated iteratively to steer the system toward the target without requiring global nonlinear dynamics. We evaluate the method on the double integrator, inverted pendulum, and Dubins car, benchmarking against model predictive control baselines. For the double integrator, we additionally benchmark against its ground-truth analytical time-optimal bang–bang solution. Our proposed ZonoReach controller achieves successful setpoint tracking and near time-optimal performance. Results highlight the influence of planning and control horizons, while limitations include reliance on local linear approximations and grid-based solvers for BRT computation. We conclude with directions for improving scalability toward real-world systems.

Keywords: Control Theory and Applications
Abstract: This paper investigates the exponential stability problem for a class of switched systems defined on time scales. By developing a discretized Lyapunov function approach integrated with minimum dwell-time switching strategy, we derive a novel sufficient condition to ensure exponential stability of the considered systems. In contrast to conventional Lyapunov function methods, the proposed approach significantly reduces conservatism while offering improved simplicity and intuitive appeal in stability analysis. To validate the theoretical findings, a representative numerical example with comparative analysis is provided to illustrate the effectiveness and advantages of the proposed method.

Keywords: Control Theory and Applications
Abstract: This paper presents an investigation into the exponential stability of switched positive impulsive systems under mode-dependent dwell time (MDT) switching constraints. By constructing a discretized quadratic copositive Lyapunov function, we derive a novel exponential stability criterion for such systems. The validity of our main results is rigorously verified through numerical examples and simulation studies.

Keywords: Control Theory and Applications, Artificial Intelligence Systems
Abstract: Recent studies have highlighted a structural connection between dynamical systems-based generative models, such as normalizing flows, diffusion models, and flow matching, and the optimal control of probability distributions. While this perspective offers a principled framework, practical implementations typically approximate intractable integrals using finite samples, resulting in a gap between theoretical predictions and empirical outcomes. To quantify this discrepancy, we derive an upper bound on the Wasserstein distance between the theoretically optimal terminal distribution and that obtained via a sample-based algorithm. Furthermore, we extend existing formulations by introducing the parametrizations of controls with neural networks, aiming for better approximations of complex target distributions.

Keywords: Control Theory and Applications
Abstract: This paper investigates stability of switched generalized homogeneous positive systems (SGHPSs), which consist of both stable and unstable subsystems. By leveraging the concept of generalized homogeneity, we develop a novel analytical framework to study the behavior of the system under bounded disturbances. The core of our approach lies in the construction of a max-separable Lyapunov function (MSLF) and the implementation of specially designed switching sequences. These techniques are specifically employed to ensure that the system state trajectories converge to a finite region within the continuous-time domain. Our primary results elucidate the precise conditions under which the system can achieve practical exponential stability. Moreover, the validity and effectiveness of our theoretical findings are demonstrated through a numerical example.

Keywords: Robotic Applications, Robot Vision, Artificial Intelligence Systems
Abstract: Adapting deep learning models to neural processing units (NPUs) poses significant challenges due to limited operator support and hardware-specific constraints. In this paper, we present the deployment of two representative models—PVN3D for object 6D pose estimation and HGGD for 6-DoF grasp detection—on Mobilint's MLA100 Low Profile equipped with the Aries2 NPU. To ensure compatibility, we replace unsupported operations and simplify architectural components while preserving task-specific performance. With these modifications, our NPU-deployed models retain over 96% of their original accuracy while achieving significant power savings. We further analyze the impact of calibration dataset composition on quantization performance. Our results show that the relevance of calibration samples has a greater effect on accuracy than their quantity, highlighting the importance of using calibration data that are semantically consistent with the target domain. These findings offer practical insights into adapting complex models for deployment under NPU constraints.

Keywords: Artificial Intelligence Systems, Navigation, Guidance and Control, Sensors and Signal Processing
Abstract: AI techniques enable control of complex and reconfigurable robot mechanisms. In such systems, a robot may switch among multiple configurations to adapt to varying tasks and environments. We focus on a six-wheeled mobile robot equipped with a bogie suspension system, and propose a mode prediction approach that determines the robot's configuration based on environmental perception. Our previous method utilized front-view camera video as input and employed a 3D ResNet combined with Bayesian temporal fusion to infer the appropriate mode. Although longer video sequences generally improve prediction performance, the limited onboard computational resources of the robot constrain the temporal input size. In this paper, we integrate a Neural Processing Unit (NPU) into the system to alleviate these computational bottlenecks, thereby enabling extended video input. We demonstrate that this enhancement improves prediction accuracy and discuss its implications for real-time deployment.

Keywords: Artificial Intelligence Systems, Robot Vision, Sensors and Signal Processing
Abstract: This paper presents an efficient post-training quantization (PTQ) method tailored for LiDAR-based 3D object detection models targeting deployment on resource-constrained edge hardware. Traditional PTQ methods often degrade performance significantly due to the sparse and spatially distributed nature of LiDAR data. To address this, we propose an iterative clipping threshold search strategy that minimizes activation quantization error by exploiting the unique distribution characteristics of LiDAR features—without requiring retraining. Additionally, we introduce a fully integer quantization pipeline for the PointPillar architecture using INT8 and INT16 representations for both weights and activations. Experimental results on the KITTI dataset demonstrate that the proposed approach maintains accuracy comparable to the full-precision FP32 model and even outperforms quantization-aware training (QAT) in some cases, while preserving the deployment efficiency of PTQ. The proposed method enables low-latency, energy-efficient inference for real-time 3D object detection on edge AI accelerators.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Robotic Applications
Abstract: Efficient deployment of deep learning models on mobile robots requires careful optimization to balance real-time performance with high accuracy on resource-constrained edge platforms. YOLOv8, widely adopted for object detection, and its segmentation model, YOLOv8-seg, enable instance-level mask prediction for richer spatial understanding. In this study, YOLOv8-seg was implemented on two hardware platforms, an NPU (MXQ) and a GPU (TensorRT on Jetson AGX Orin), and their suitability for indoor perception tasks was evaluated. The analysis highlights the advantages and potential challenges of NPU-based deployment, providing insights for the design of efficient and scalable perception systems in autonomous mobile robots.

Keywords: Robotic Applications, Robot Vision, Artificial Intelligence Systems
Abstract: This paper presents an embedded system for efficient 3D object detection in robotics and autonomous systems using point cloud data acquired from 3D LiDAR sensors. Due to the large data volume and high computational complexity of point cloud processing, real-time object detection on embedded platforms remains challenging. To address this, the proposed system integrates a voxel feature encoder and a 3D convolution accelerator on an FPGA, combined with an ARM-based embedded module for post-processing and system control. The point cloud data are voxelized and converted into voxel-wise features, which are processed by the 3D convolution accelerator and further refined through a region proposal network (RPN) and post-processing to generate 3D bounding boxes in both bird’s eye view (BEV) and image space. The proposed system successfully detects objects in categories such as cars, cyclists, and pedestrians, achieving inference times comparable to those of conventional high-performance CPU/GPU platforms.

Keywords: Artificial Intelligence Systems, Control Theory and Applications, Information and Networking
Abstract: In response to the growing demand for eco-friendly and efficient VRF technologies, this study presents an automatic indoor unit location classification algorithm to improve operational efficiency and energy savings. The core idea employs Q-learning to model the influence of actions on each indoor unit’s behavior, enabling direct comparison via Q-tables. The proposed method utilizes temperature, refrigerant pipe temperature, operation rate, and activation status data collected from indoor unit sensors to construct a Q-learning–based reinforcement learning model. Cosine similarity between Q-tables evaluates spatial similarity, with a max-margin method automatically determining the optimal clustering threshold. A maximum margin method from machine learning is then applied to automatically determine the threshold of similarity between indoor units, enabling high-precision clustering of units located in the same space. Compared to conventional linear correlation-based methods, the proposed algorithm improves classification accuracy relying solely on temperature-related data. Experiments on diverse VRF setups with 8 to 15 indoor units demonstrate over 92% accuracy and training completion within five days. These findings support enhanced cooperative control and energy-efficient VRF operation in smart building systems.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Robot Vision
Abstract: Underwater robots equipped with optical cameras play a crucial role in supporting diver activities and performing autonomous underwater tasks. However, because of optical phenomena in underwater environments, captured images often contain a large amount of unstructured glare and scattering effects. These artifacts provide valuable information, such as breathing bubbles or material surface properties, depending on the application. However, their characteristics vary significantly depending on the relative positioning of the light source and objects, affecting overall brightness and luminance. Furthermore, their irregular nature makes it difficult to establish a consistent labeling standard, as these artifacts exhibit abrupt and ambiguous boundaries. To address this issue, we propose a pseudo ground truth (pseudo GT) generation algorithm based on pixel clustering to segment unstructured glare areas. Since the generated glare regions are difficult to distinguish from the surrounding background, we design a region-based loss function instead of a pixel-wise loss function. Specifically, we utilize color distribution histograms as an objective function and loss function to improve segmentation accuracy. The effectiveness of the proposed pseudo GT was validated by comparing it with hand-crafted results using mean Intersection over Union (mIoU) scores. Furthermore, we trained a ResNet50-based model using the generated Pseudo GT and conducted qualitative evaluations. Through data augmentation, we trained the model using an indoor water tank dataset and tested it on real underwater environments, as well as different indoor tank settings. The results confirm that our method effectively detects glare regions in various underwater conditions.

Keywords: Artificial Intelligence Systems, Control Theory and Applications
Abstract: This paper introduces an effective idea which can easily extend filter design methods of recurrent neural networks (RNNs) to gated recurrent units (GRUs). Although the GRU is widely used in the many applications, its filter design method has not been studied yet. Because the gating mechanism makes the dynamics of the RNNs more complex, existing filter design methods of the RNNs cannot be directly applied to the GRUs. In the proposed method, the Takagi-Sugeno (T-S) fuzzy model is used to simply reformulate the dynamics of the GRU as the weighted sum of sub-recurrent neural networks, and a filter design method is presented for the fuzzified GRU model. In addition, the mismatched membership function problem of the fuzzy filtering error system is resolved by adopting the transformed membership functions. It has an advantage that the existing relaxation methods of the stabilization criterion can be directly applied to the presented method. The proposed method is verified through a GRU model, and the average root‑mean‑squared errors for each of the three hidden states are close to zero (0.1130, 0.1796, 0.1106) for a given three‑dimensional numerical problem.

Keywords: Artificial Intelligence Systems, Robot Mechanism and Control, Robotic Applications
Abstract: Brachiation, inspired by ape locomotion, involves swinging from one substrate to another. Existing approaches typically rely on simple grippers and computationally expensive optimal control to compute feasible states and control trajectories. In contrast, learning-based methods often lack physical modeling and require extensive training data. We present a brachiating system using high degree-of-freedom anthropomorphic hands to generate swing trajectories and perform stable grasps in a physics-based simulation environment (MuJoCo). An optimal open-loop trajectory is first generated via trajectory optimization based on a desired grasp location. A tracking controller follows this reference, while a grasping controller activates upon proximity to ensure secure contact. To reduce computational cost, we train a Generative Adversarial Imitation Learning (GAIL) policy using expert trajectories from the optimization framework. The GAIL-based controller generalizes to perturbed conditions and eliminates the need for repeated re-optimization, significantly lowering computation time. It also adapts to varying initial configurations, removing the requirement to rerun optimization for each case. We compare the learned model with a traditional optimal controller and demonstrate marked improvements in both computational efficiency and versatility.

Keywords: Artificial Intelligence Systems, Robot Vision
Abstract: River quarrying plays an important role in sourcing aggregates like sand and gravel, supporting local economies and infrastructure development. However, unregulated quarry operations can cause severe environmental degradation, including riverbank erosion and habitat loss. Existing raft monitoring studies have largely relied on satellite imagery, which is effective for regional assessments but less suited to capturing the fine-scale, rapidly changing conditions of quarry activities. Aerial-based monitoring provides a practical alternative by offering high-resolution observations at the site level. To address this gap, this study introduces aerial-based detection models for bamboo rafts and tiny persons operating in river quarry sites. Two custom datasets were developed to train and evaluate object detection models under real-world conditions. YOLOv5l6 for tiny person detection achieved an F1-score of 93.47%, and mAP@0.5 of 90.32%. In contrast, YOLOv11 for raft detection demonstrated high sensitivity achieving F1-score of 82.26%, and mAP@0.5 of 85.80%. Overall, this study demonstrates the potential of YOLO-based models for aerial-based river quarry monitoring. The proposed system strengthens oversight of quarrying activities and supports regulators and communities in promoting environmental compliance in remote river systems.

Keywords: Artificial Intelligence Systems, Robot Vision, Autonomous Vehicle Systems
Abstract: Accurate centerline estimation is important for autonomous ground vehicle (AGV) navigation on unstructured roads. This study presents a line marker detection system that combines YOLOv11n-seg instance segmentation with rule-based decision logic to identify unstructured road features that guide centerline estimation. The models were trained on a custom dataset collected using a GoPro camera mounted at a 13° downward angle on a ground vehicle, capturing asphalt and concrete road surfaces. Performance evaluation yielded 95.8% Recall, 50.8% Precision, 66.1% F1-Score, and 80.9% mAP@0.5. Solid white and yellow lines offered strong visual hints, while concrete slab joints aided in centerline identification. A rule-based strategy selected the most reliable line(s) per frame, achieving 89.58% decision-making accuracy and a 62.48% line marker detection accuracy. Results demonstrate the algorithm’s potential for robust centerline estimation in AGV applications.

Keywords: Sensors and Signal Processing, Navigation, Guidance and Control, Control Theory and Applications
Abstract: In this paper, we consider situation where the number of weighted particles in particle filter becomes too small. Usual approach for avoiding this is the resampling, but there are situations where this may still happen. It is possible that the realized measurement is unlikely and there are not enough particles to handle this situation. We propose to use a Kalman type filter that uses particles from the last time step as fallback method in this kind of situation.

Keywords: Sensors and Signal Processing, Navigation, Guidance and Control, Control Theory and Applications
Abstract: In this paper, we propose a method for state estimation that uses data samples that may be irregularly distributed as learning data. We assume that there is no known mathematical measurement model that could be fitted to these samples. The proposed algorithm fits a local linear model that uses sample information within the prior. In linearisation, we fit according to the prior an affine model and also compute its residual covariance. The resulting linearisation can be used directly in the Kalman Filter for state estimation. We show in examples how the proposed method can be used for creating local linearisations based on non-uniform sample points within prior.

Keywords: Sensors and Signal Processing, Artificial Intelligence Systems, Control Theory and Applications
Abstract: This study introduces a fault diagnosis model for wind turbine blade fatigue testing, based on an improved Teager Energy Operator (TEO) integrated with HDC-CNN-LSTM. By combining theoretical derivation and blade testing characteristics, a TEO response baseline for healthy states was established. Exponential Weighted Moving Average (EWMA) was employed to enhance damage features, while Short-Time Fourier Transform (STFT) was used to refine TEO features. Time-frequency domain statistical features from both strain and TEO response signals were combined to create a comprehensive blade damage detection dataset. The model leverages Convolutional Neural Network (CNN) for adaptive feature extraction and incorporates Hybrid Dilated Convolutions (HDC) to capture multi-scale temporal characteristics. A multi-head attention mechanism was introduced to capture data diversity from various perspectives, improving the model's representation capabilities. A two-layer Long Short-Term Memory (LSTM) network was employed for temporal modeling and classification. Experimental results demonstrate that this approach achieves high accuracy in identifying blade faults, outperforming traditional LSTM neural network architectures.

Keywords: Rehabilitation Robot, Human-Robot Interaction, Sensors and Signal Processing
Abstract: The rising demand for advanced control of prosthetic devices underscores the need for robust grip force estimation methods at various arm postures. This study presents a domain adversarial method for precise estimation of grip force with high-density electromyography (HD-EMG) under varying arm postures. Using adversarial training, the proposed approach extracts domain-invariant features that are beneficial for force estimation across different arm positions. The experimental results demonstrate a 49.6% improvement in the accuracy of the grip force estimation compared to the CNN model without domain adversarial training, with a reduction in the normalized root mean square error (NRMSE) from 0.125±0.011 to 0.063±0.004. The proposed method has great potential to ensure accurate grip force control of prosthetic hands by addressing posture-induced variability in HD-EMG signals.

Keywords: Sensors and Signal Processing, Robotic Applications, Navigation, Guidance and Control
Abstract: Accurate state estimation is essential for quadruped robots, particularly when relying solely on onboard proprioceptive sensors. However, conventional IMU-based estimators often suffer from drift due to unconsidered accelerometer bias. In this paper, we propose a force observer-based method for IMU bias compensation that enhances base position and velocity estimation (i.e., of the main body) without requiring any exteroceptive sensing. The approach estimates base acceleration from ground reaction forces using a model-based force observer and compares it with IMU measurements. The resulting discrepancy is incorporated into a Kalman filter as a correction term. To evaluate the robustness of the method, we conduct simulation experiments using a quadruped robot and two types of IMU sensors with different bias characteristics. The proposed filter consistently outperforms a baseline estimator, achieving notably lower RMSE in both position and velocity. In particular, the method significantly reduces long-term position drift even with the high-bias sensor, demonstrating its effectiveness and robustness under varying sensor conditions.

Keywords: Sensors and Signal Processing, Autonomous Vehicle Systems, Robot Vision
Abstract: We propose a drift-aware estimation framework for roll and pitch using an Extended Kalman Filter (EKF) that fuses high-frequency inertial data from an IMU with drift-compensated observations from a visual odometry (VO) system. The EKF state vector explicitly models angle errors, angular velocities, and gyroscope biases, allowing for the long-term correction of integration drift. In the prediction step, IMU angular velocity is propagated through a kinematic model. In contrast, the update step incorporates roll and pitch estimates from VO to correct accumulated drift and refine bias estimates. The fusion process leverages the complementary properties of the two sensors' high-rate IMU samples for responsiveness and low-drift VO samples for stability. We test our method using real-world driving data collected on various terrains, including smooth surfaces, around corners, and on rough terrain. Quantitative results show that our EKF-based approach achieves improved estimation accuracy compared to either IMU or VO alone (up to 45% improvement in both RMSE and MAE). The proposed method is low-complexity, robust to motion dynamics, and suitable for real-time implementation in fully autonomous navigation systems that require stable orientation across extended timescales.

Keywords: Robot Vision, Robotic Applications, Industrial Applications of Control
Abstract: Optimizing drill and blast processes in surface mining enhances material fragmentation, leading to millions of dollars of cost and significant energy savings every year. Understanding the structural properties of blastholes can improve blast design and drilling quality. Visual inspection of blastholes, especially their slurry-covered walls, offers a fast and cost-effective method for analysis. To explore correlations between slurry texture and hole or mineral properties, the first step is to classify slurry textures. This paper introduces new techniques for that classification. To avoid the high cost of matching observations with expert-labelled datasets and chemical assays in downhole image analysis, the authors propose an unsupervised deep learning approach using a beta variational autoencoder (beta-VAE). This model learns textural features from unlabeled downhole video data. However, transferring the beta-VAE to this new domain causes information loss in deeper layers. The authors address this by modifying the architecture, significantly improving its performance for geological applications. The learned feature space supports tasks like unsupervised clustering and query-by-image. Their method works on new datasets without manual labelling and can be adapted to focus on different features. A demonstrated clustering pipeline achieved 59% accuracy on novice-labelled data.

Keywords: Robot Vision, Autonomous Vehicle Systems, Human-Robot Interaction
Abstract: Scene Coordinate Regression (SCR) has become a prominent approach in visual localization for predicting 2D-3D correspondences. While many SCR methods either treat all image regions as equally important or focus on semantically prominent areas like buildings, in reality, only specific regions provide reliable cues for correspondence learning. Our analysis reveals a mismatch between human-defined semantic assumptions and the regions that are empirically beneficial for SCR. By projecting reconstructed 3D points onto training images and evaluating their alignment with text embeddings, we find that less intuitive concept words often correlate more strongly with triangulated points than expected categories like buildings. Although using these better-aligned prompts for text-guided masking can improve performance in certain scenes, it may degrade results elsewhere, exposing the approach’s sensitivity to scene characteristics. This highlights the limitations of fixed prompt-based strategies and motivates the need for adaptive, data-driven region selection in SCR.

Keywords: Robot Vision, Human-Robot Interaction, Robotic Applications
Abstract: This paper presents a comparative analysis of three widely-used JavaScript frameworks, ReactJS, VueJS, and AngularJS, for real-time image streaming in Robot Operating System 2 (ROS2) environments. Using the rclnodejs library for ROS2-WebSocket integration, the frameworks were tested in two setups: an isolated system using an NVIDIA Jetson Xavier NX, and a multi platform setup with a remote field PC. Each framework's performance was evaluated in terms of latency, frame rate, and memory usage, with a Python-based OpenCV node serving as a reference benchmark. Results show that VueJS had the best overall performance, overcoming the benchmark for the multi platform setup, with the lowest latency and memory usage, while ReactJS provided similar results with slightly higher resource consumption. AngularJS, even though offers a structured development approach, exhibited higher latency. All the frameworks showed a very low memory usage when compared with the Python code. The findings shows VueJS as the most efficient choice for real-time image visualization in ROS2 applications, however all the JavaScripts frameworks tested appear to perform better than the Python.

Keywords: Robot Vision, Robotic Applications, Sensors and Signal Processing
Abstract: Landmine detection, especially of PFM-1 “butterfly’’ mines, remains a critical challenge in post-conflict environments due to their small size and plastic construction. Modern demining operations increasingly incorporate advanced sensing modalities, including multi-spectral imaging. Long-Wave Infrared (LWIR), in particular, offers potential for detecting PFM-1 mines under certain environmental conditions. However, the limited availability of annotated thermal data restricts the generalization capabilities of deep learning approaches. To address these limitations, we propose a multi-stage, feature-based detection algorithm tailored for PFM-1 mines in LWIR thermal imagery. The method is trained on Track 1 and Track 2 of the MineInsight dataset, using F1 and F2 score-weighted loss functions to prioritize recall and reduce false negatives. Across ten independent runs, the proposed method exhibits comparable performance and complementary robustness relative to YOLOv8, especially under conditions of limited training data or partial occlusion.

Keywords: Robot Vision, Human-Robot Interaction, Control Theory and Applications
Abstract: We present a hybrid vision-based tracking framework for robust image-based visual servoing (IBVS) under severe occlusions. Traditional methods like Lucas–Kanade (LK) are lightweight but fragile to drift and occlusion, while deep learning models demand continuous visibility and high computation. Our system integrates a fast global template matcher, a deep-feature-enhanced LK module for sub-pixel alignment, and a lightweight residual regressor to correct local misalignments from blur or occlusion. When visual confidence degrades, a GRU-based predictor uses motion history to extrapolate pose updates. The final output, translation, rotation, and scale deltas are directly formatted for 30 Hz control loops, enabling seamless robot actuation. Evaluated on handheld video sequences with up to 90% occlusion, our method achieves <2 px tracking error while maintaining real-time performance on CPU. This approach bridges perception and control, offering a low-latency, occlusion-resilient solution for real-world robotic applications such as manipulation, inspection, and visual alignment.

Keywords: Sensors and Signal Processing, Navigation, Guidance and Control, Human-Robot Interaction
Abstract: This paper presents a hybrid approach that combines a Kalman filter with cubic spline interpolation to estimate human walking trajectories using a 2D LiDAR sensor. A Kalman filter based on a constant velocity motion model is employed to track human position using leg scan data. When two leg clusters are detected, the filter receives an observation update. However, during walking, self-occlusion between the legs can occur, resulting in missing data. To address this issue, the cubic spline algorithm is applied to estimate the human position and velocity during occlusion periods. The estimated trajectory is then used to classify the left and right legs. An experiment is conducted to validate the effectiveness of the proposed method.

Keywords: Robot Vision, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: This study proposes a method for real-time and reliable drivable area estimation by fusing drivable area masks extracted from camera images with interpolated ground point clouds obtained from LiDAR. First, traversable regions are detected from RGB images using object detection and instance segmentation techniques, and then converted into binary mask representations. Simultaneously, ground points are extracted from LiDAR point clouds and converted into range images, which are then interpolated to generate dense point cloud representations. These interpolated ground point clouds are projected into the image coordinate system and spatially aligned with the segmentation mask. By filtering only the overlapping regions between the point cloud and the segmentation mask, the system generates a final drivable area point cloud with reduced false positives. The proposed method operates effectively in real time even with low-resolution LiDAR sensors or in environments with limited visibility. By leveraging the complementary strengths of visual and LiDAR-based perception, it successfully addresses the limitations of single-sensor approaches. Experimental validation in real-world environments demonstrates the robustness and real-time applicability of the proposed system.

Keywords: Navigation, Guidance and Control, Robotic Applications, Autonomous Vehicle Systems
Abstract: This paper presents a novel approach for vehicle position tracking using an interacting multiple model (IMM) framework grounded in Lie theory on the 3-sphere. The proposed method addresses attitude increments and errors through Lie algebra, employing exponential and logarithmic mappings for precise correction and evaluation. Attitude covariances are consistently represented in the Lie algebra. To validate its effectiveness, the approach is benchmarked against two alternative filters: (1) a unit quaternion-based unscented Kalman filter (UKF) that does not incorporate Lie theory, and (2) a single-model UKF that integrates both unit quaternions and Lie algebraic operations. Simulation results show that the proposed IMM-Lie theory framework achieves superior tracking accuracy and reduced estimation errors. Additionally, it dynamically adapts to varying vehicle maneuvers by updating model probabilities, enabling timely and appropriate responses. These results underscore the robustness and practical utility of the method in complex, maneuver-rich tracking scenarios.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Control Theory and Applications
Abstract: This paper presents a Sequential Convex Programming (SCP) framework for solving nonconvex optimal control problems with improved numerical efficiency and robustness. The proposed algorithm iteratively linearizes the system dynamics and constraints, formulating a sequence of convex subproblems that are solved using a high-performance conic solver. Key innovations include the use of of diagonal scaling matrices and centering vectors to transform state, control, and parameter variables into scaled quantities, ensuring better numerical conditioning and solver performance. The framework also incorporates trust regions and virtual states to handle artificial infeasibility and ensure convergence. The effectiveness of the approach is demonstrated through a multi-phase rocket landing guidance problem, where the algorithm achieves real-time performance while maintaining high accuracy. The results highlight the scalability and robustness of the proposed method, making it suitable for complex nonlinear control applications in aerospace and beyond.

Keywords: Navigation, Guidance and Control, Robotic Applications, Information and Networking
Abstract: This paper proposes a weight-based graph model to measure traffic flow in a multi-robot system. The proposed method calculates the weight of each edge by combining distance, robot capacity, bottleneck risk, and section status, which enables the quantitative evaluation of real-time traffic flow and congestion within the collaborative AMR environment. Unlike the existing distance-based static method, this method dynamically adjusts the weight in real-time, reflecting the robot's current state, parameters, and environmental information. Additionally, it is possible to avoid bottleneck situations by dynamically adjusting the weight. Through experiments on multi-path distribution, bottleneck response, and environmental error scenarios conducted in the Gazebo simulation environment based on ROS2, the effectiveness of the proposed method was demonstrated, showing improved performance in terms of congestion avoidance and flow efficiency compared to the existing distance-based method.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: Post-disaster environments often present confined, unstable, and hazardous terrain that challenges human responders. Tracked ground robots offer a safer alternative, particularly in GPS-denied indoor settings. This study presents a 3D LiDAR-based approach for autonomous flipper control in tracked robots, enhancing their capability to navigate obstacles without relying on additional sensors. By directly processing LiDAR point cloud data, the system estimates obstacle height, distance, and tilt angle to dynamically adjust flipper positions in real time. Experimental results show reliable performance, achieving over 94% average accuracy in distance detection and up to 92% accuracy in obstacle height estimation, with improved performance at farther sensor distances. These capabilities contribute to enhanced terrain adaptability, improved obstacle-climbing performance, and overall stability which is a key factors for effective deployment in disaster response operations. Unlike sensor fusion methods, the proposed solution simplifies the control architecture, reduces system complexity, and enables efficient, autonomous operation, minimizing reliance on human intervention and improving mission effectiveness.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems, Robotic Applications
Abstract: This study proposes an effective registration method for precise alignment and mapping of underwater terrain using three-dimensional multibeam echo sounder (3D-MBES) data. The proposed approach accumulates temporally collected MBES point clouds using a sliding window technique to construct stable terrain data, which are then projected onto a two-dimensional (2D) depth image on the XY plane. A deep learning-based feature extraction method employing the HardNet descriptor is applied to the generated image, and image-based registration is performed to estimate relative transformations between adjacent scans. The estimated transformation is subsequently extended to a 3D transformation matrix and applied to the original point cloud, ensuring both computational efficiency and registration accuracy. The registered point clouds are then accumulated to form locally aligned terrain blocks, referred to as Map-tiles, which are further integrated to construct a global map. Experimental results using real underwater data demonstrate the accuracy and applicability of the proposed method in underwater environments.

Keywords: Robot Mechanism and Control, Robotic Applications, Exoskeleton Robot
Abstract: Possessing resilience against tabletop collisions is crucial for grippers. Conventional passive bending grippers require a recovery phase after impact, reducing grasping efficiency. This paper proposes a novel mechanism: using downward pressing against the tabletop to trigger passive finger bending along an inclined path, enabling enveloping grasp. We design a gripper with passively bending fingers following an inclined trajectory, improving efficiency. It also incorporates active locking to maintain grip force—named the PIMAL gripper. A Chebyshev linkage guides the fingertip along an inward-inclined, near-linear path. After pressing triggers contact, the finger tightens passively to grasp. A motor then rotates a limit block until it locks the active link, maintaining the grasp. A parallelogram mechanism stabilizes the fingertip. This dual locking allows direct lifting without recovery, boosting efficiency. Theoretical and experimental results confirm highly efficient grasping. Applicable in industrial sorting, logistics, and service robotics, this design offers a novel “Passive Adaptation-Active Locking” solution for desktop grasping triggered by contact pressing.

Keywords: Robot Mechanism and Control, Robotic Applications, Exoskeleton Robot
Abstract: Adaptive multi-mode grasping is currently a key research area for robot grippers. The existing parallel grippers cannot touch the desktop while grasping, making it difficult to grasp thin objects on the desktop. This paper proposes a novel underactuated robotic gripper, the gear-linkage gripper (GL gripper). By integrating a gear-linkage mechanism and a parallelogram linkage, the design enables pinching, scooping. The GL gripper requires only a single actuator to enable coordinated multi-finger motion. The gear-linkage mechanism enables linear motion and passive mode switching of the device, demonstrating its adaptability to tables of different heights. In contrast, the parallelogram linkage ensures that the fingertips maintain a fixed orientation relative to the base. The incorporation of elastic elements enhances the fingers' flexibility for effective environmental interaction. The GL gripper with two fingers was developed and tested in grasping objects of varying sizes and thicknesses. Experimental results show that the GL gripper can achieve reliable objects grasping and maintain a stable performance for desktop grasp.

Keywords: Robot Mechanism and Control, Control Theory and Applications, Robotic Applications
Abstract: This research considers the stabilization problem of the highly nonlinear model of a spherical inverted pendulum with two degrees of freedom (that is, the pivot point of the inverted pendulum is fixed in the apace and its end point moves on a sphere-like surface). The research focuses specifically on presenting relatively simple controllers that operate even in the case of uncertainty in the parameters of the mechanical model (by implementing state and output controllers under partial knowledge of the gravitational factor) and when state variables whose measurement induces noisy signals are not implemented in the control feedback (by using an output-based controller). The control problems of inverted pendulum models appear in many physical systems, so we will discuss some possible relevant applications of the research results. Numerical and graphical results demonstrate the performance of the proposed control algorithms.

Keywords: Robotic Applications, Sensors and Signal Processing
Abstract: When operating autonomous mobile robots indoors, a system is required to recognize which floor they are on and move between floors. Using a camera to read a dial and recognize the current floor is not possible when there are objects or people in front of the dial. Communicating with an elevator to recognize the current floor requires time and money to install the necessary equipment. Therefore, in this study, we propose a system that uses a barometric pressure sensor to determine the floor-level and move autonomously between floors. The proposed method performs floor-level determination in response to barometric pressure changes caused by time variations. Under the condition that the button is pressed by a person, experiments confirmed that the robot could enter the elevator from its current floor and exit at the specified floor.

Keywords: Robotic Applications, Sensors and Signal Processing, Navigation, Guidance and Control
Abstract: We propose RA-LLO, a real-time odometry framework tailored for highly dynamic legged robots, combining inertial sensors, joint kinematics, foot-contact sensing, and LiDAR data into a unified probabilistic approach. At its core, the method uses an error state Kalman filter with a Gaussian process-based continuous-time motion model, precisely discretized for accurate state prediction. Foot contacts are handled smoothly through confidence-weighted sensor inputs from pneumatic force sensors, ensuring stable stance estimation without sudden corrections. LiDAR scans are deskewed using a cubic B-spline fitted to predicted motion, aligning data accurately in time. Experiments with a Unitree Go2 robot demonstrated significant improvements in vertical drift and overall odometry accuracy compared to existing methods, running efficiently with low-drift state estimation suitable for challenging, unstructured terrains.

Keywords: Robot Mechanism and Control
Abstract: This paper proposes an adaptive finger mechanism for environmental constraints. The mechanism is based on the coupling of Scott linkages and passive elastic elements and is controlled by a single linear driving device to achieve composite grasping functions such as pinching, bimodal scooping (symmetrical scooping and asymmetrical scooping) and adaptive enveloping. The proposed finger mechanism is more concise compared to previously developed mechanisms, can be installed on existing commercial robotic arms and has lower production costs. This mechanism utilizes the linear motion characteristics and geometric constraints of the Scott linkage, combined with spring limit, to achieve vertical flexibility in the face of environmental constraints and can adaptively grasp objects of different shapes, materials and positions. Especially in the symmetrical scooping configuration, it can effectively grasp large thin objects such as cards; Under asymmetric configuration, the stability of sheet grasping is improved by tilting the gripper. This study provides new ideas for the design of low-cost, high-precision environment adaptive grippers and expands the potential application of the underactuated robotic gripper in complex environments.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems, Robotic Applications
Abstract: Where many tasks are entrusted to autonomous dynamic systems, it is important to provide reliable solutions able to tackle the increasing complexity of such systems and their tasks. Conventional control methods require modeling and stability analysis of these dynamic systems as a whole. This means that with increases in the complexity of the dynamic system, controller design and stability analysis become more difficult. In mobile robot control, in order to handle the complexity of the combined mobile platform and the tools, a modular control approach can be utilized that breaks the complex system down into less complex components and proves the stability of the controlled dynamic system through the stability of its sub systems. In this study, a general skid-steered mobile platform is considered as a single sub system, and the controller for its motion trajectory tracking is designed. Then, a stability analysis of the controller is conducted, and its performance is assessed in a commonly used physics engine. The results show promising outcomes, and the framework presented in this paper can facilitate the use of mobile robots.

Keywords: Control Theory and Applications, Industrial Applications of Control, Autonomous Vehicle Systems
Abstract: This work addresses a extended class of optimal control problems where the target for the system state has the form of an ellipsoid rather than a fixed, single point. As a computationally affordable method for resolving the extended problem, we present a revised version of the differential dynamic programming (DDP), termed the differential dynamic programming with ellipsoidal target set (ETS-DDP). To this end, the problem with an ellipsoidal target set is reformulated into an equivalent form with the orthogonal projection operator, yielding that the resulting cost functions turn out to be discontinuous at some points. As the DDP usually requires the differentiability of cost functions, in the ETS-DDP formulation we locally approximate the (nonsmooth) cost functions to smoothed ones near the path generated at the previous iteration, by utilizing the explicit form of the orthogonal projection operator. Moreover, a statistical inference is also presented for a design method of the ellipsoidal target set, based on data on admissible target points collected by expect demonstration. Via a simulation on autonomous parking of a vehicle, it is seen that the proposed ETS-DDP efficiently derives an admissible state trajectory while running much faster than the point-targeted DDP, at the expense of optimality.

Keywords: Control Theory and Applications, Industrial Applications of Control, Robot Mechanism and Control
Abstract: This study proposes an integral error-based adaptive law for neural identifiers, aimed at enhancing the performance of online system identification for nonlinear systems. Unlike conventional adaptive laws that update the neural network based on instantaneous errors, the proposed approach performs updates using identification errors accumulated over time. This mechanism enables the neural network to achieve more consistent and accurate function approximation over the entire time interval, ensuring stable online learning of unknown nonlinear dynamics. A Lyapunov-based theoretical analysis guarantees the uniform ultimate boundedness of the neural identifier. Simulation results on a nonlinear robot manipulator system demonstrate the effectiveness and improved convergence properties of the proposed method compared to a conventional instantaneous error-based approach.

Keywords: Control Theory and Applications
Abstract: This paper proposes data collection conditions for data-driven system identification of nonlinear systems that admit a Koopman linear embedding. Koopman operator-based system identification does not provide conditions on how data should be collected or how much data is required to guarantee identifiability. To address this issue, we integrate the identifiability condition from behavioral systems theory into a Koopman-based system identification framework. Since nonlinear systems that admit a linear embedding are generally not controllable, we adopt a relaxed version of the identification condition proposed by the fundamental lemma. This relaxed condition provides a criterion for data informativeness by imposing a rank condition on the Hankel matrix formed from the collected data. This ensures that the true system aligns with the most powerful unfalsified model, thereby guaranteeing identifiability. We extend this condition to nonlinear systems that admit a linear embedding, and establish a theoretical lower bound on the number of data points necessary for identifying nonlinear systems. Simulation results demonstrate that accurate system modeling and controller design are achievable with the minimal data satisfying this condition.

Keywords: Control Theory and Applications, Industrial Applications of Control
Abstract: Grid-forming inverters (GFM) integrated with the capabilities of traditional synchronous generators (SGs) are emerging as a promising solution for the grids dominated by inverter-based resources (IBR). However, the GFM inverters require several cascaded control loops to effectively emulate the qualities and functionalities of the SGs. Since GFMs lack inherent inertia to support the grid, the entire process of frequency and voltage regulation relies substantially on the control structure implemented. Therefore, it is crucial to develop advanced control methods that are both simple and capable of providing better dynamic response. At present, the inner control loops are predominantly dual-loop configurations, where interaction between the control loop bandwidths leads to instability. To address this, this paper proposes a single loop Model Predictive based voltage control for the inner loop, combined with an outer primary Droop controller. Additionally, a modified adaptive virtual impedance scheme is incorporated to enable current limiting during symmetrical fault-induced voltage sags. The effectiveness of the proposed control is evaluated across a range of short circuit ratio (SCR) values using PLECS simulations.

Keywords: Control Theory and Applications
Abstract: Q-filter-based disturbance observer (DOB) is one of the most widely used robust controller due to its design simplicity. Such simplicity arises from that reducing time constant of low pass filters, not only ensures robust stability but also enhances nominal performance recovery---ability to recover the trajectory of nominal closed-loop system. However, in contrast to noise-free setup, excessively small time constant can rather damage the nominal performance recovery under measurement noise. That is, minimizing time constant is no longer immediately guaranteeing nominal performance recovery. Motivated by this observation, this paper concentrates on determination of time constant to guarantee transient and steady state performance. This analysis uses Lyapunov method utilizing the coordinate transformation inspired by the analysis based on singular perturbation theory. As a result, we present an affordable noise level and open interval for time constant that guarantees both the required performances. The analysis can also lead to theoretical demonstration on that excessively reducing time constant gives assurance to achieve target performance only for noise-free case.

Keywords: Control Theory and Applications, Robot Mechanism and Control
Abstract: Accurate identification of inertial parameters that reflect the hardware of the physical-world is essential for model-based control and building high-fidelity simulations. Software Defined Robot (SDR) frameworks, where functionality is primarily defined and reconfigured through software, require adaptability to changes in physical properties. To ensure that SDR can autonomously adapt to changes in hardware configuration, such as adding sensors and changing hardware components, it is essential to continuously update the inertial parameters of the robots. In this paper, we propose an optimization-based real-time identification method that estimates inertial parameters using sensor data, including joint positions, velocities, and currents/torques. We validate the proposed method in both simulation and hardware on our 3 Degrees of Freedom (DoF) robot, demonstrating its effectiveness in identification of inertial parameters.

Keywords: Information and Networking, Robotic Applications
Abstract: Despite continuous efforts to deploy robots in markets and daily life, widespread adoption has yet to be achieved. Recently, opinion leaders in the robotics field have advocated for a shift from hardware-driven to software-driven markets, anticipating that this transition will enhance the versatility and scalability of robotic systems. A key concept enabling this shift is the Software-Defined Robot (SDR). Among various SDR features, Over-the-Air (OTA) software updates are particularly critical, as they directly impact the lifecycle and maintainability of robotic software. In this paper, we propose an OTA update framework tailored for SDR systems. To facilitate practical deployment, the framework integrates ROS 2 middleware and Docker Engine, enabling modular and containerized software management. As a proof of concept, we implement and validate two Linux shell scripts that perform OTA updates within the proposed architecture.

Keywords: Robotic Applications, Robot Mechanism and Control, Robot Vision
Abstract: This paper presents DORI, a real-world manipulator dataset designed for learning language-conditioned manipulation policies. The dataset includes multi-view RGB images from top-view and wrist-view cameras, end-effector poses, joint states, and frame-aligned natural language instructions. It consists of 52 manipulation tasks and 294 demonstrations, and is well-suited for training and evaluating language conditioned manipulation policies across diverse perspectives and expressions.

Keywords: Robotic Applications, Artificial Intelligence Systems
Abstract: As robotic manipulation continues to advance, a major challenge is the absence of a common framework for teleoperation and data collection across various robots. Without such a foundation, each new robot platform requires its own interface and data pipeline, and behavior cloning algorithms like ACT must be redesigned whenever the target robot changes. To address these issues, we present a framework that combines a ROS2/MoveIt2-based teleoperation system with transformer-based behavior cloning. Using a haptic device (Desktop 6D), we collected demonstration data by teleoperating different robot arms, including the UR10e and Kinova Gen3. These demonstrations from various robots were then used to train the ACT algorithm, allowing policy learning that can be applied across different robot platforms. The proposed approach connects hardware diversity with learning-based policy development and supports the vision of software-defined robotics, offering a scalable path toward more adaptable and intelligent robotic systems.

Keywords: Robotic Applications, Industrial Applications of Control
Abstract: This paper presents a method for resolving interface incompatibilities between software modules in robotic systems by introducing a bridge module. In practice, modular upgrades and vendor-specific designs often lead to inconsistencies in the interface signatures of software modules—particularly when integrating modules from different sources or versions. The proposed solution leverages the standardized information models defined in ISO 22166-201 and ISO 22166-202 to describe required and provided interfaces, and uses a GUI-based tool to manually map mismatched interface signatures. Once the mapping is completed, conversion code for a bridge module can be automatically generated for either Linux or Windows environments. Furthermore, an extended tool supporting ROS 2 nodes demonstrates the practical applicability of the approach in real-world robotic systems. The bridge module enables legacy or heterogeneous software modules to interoperate without requiring direct modification, thereby enhancing maintainability and modularity.

Keywords: Robotic Applications, Control Theory and Applications, Industrial Applications of Control
Abstract: Manipulators mounted on an unmanned ground vehicle should be operated very accurately with various missions such as bomb detection and disposal. The manipulator is sometimes situated in harsh outdoor conditions like wind, snow and shock blow occurring sinusoidal disturbance torques on the joint motors which are strong relation with manipulator link structure. In this study, sliding mode control (SMC) combined with disturbance observer based (DOB) and repetitive disturbance observer (RDO) were considered at addressing the tracking control issue for a simplified three -DOF robot manipulator. The manipulator dynamic equation was derived and Lyapunov stability is investigated to analyze the control stability and robustness of SMC and disturbance estimators. Several simulations were implemented, compared among traditional proportional, integral and differential (PID) control, SMC and combined SMC. It was verified that this combined SMC method was suitable for minimizing tracking errors and fast tracking.

Keywords: Robot Vision, Robotic Applications, Control Devices and Instruments
Abstract: The quality assessment of hole-making in carbon fiber reinforced polymer (CFRP) composites is a key technical challenge in high-end equipment fields such as aerospace and automotive manufacturing. To address the problems of fiber texture interference and feature coupling in hole defect detection, this study proposes a dynamic region-growing detection framework based on breadth-first search (BFS). This framework innovatively integrates Hough circle detection and hierarchical traversal mechanisms, effectively overcoming the application limitations of traditional threshold segmentation methods in the context of fiber texture backgrounds. Firstly, a preprocessing procedure based on multi-scale morphological filtering is adopted to significantly suppress the surface texture noise of carbon fibers. Additionally, an adaptive BFS defect search algorithm is designed, taking the hole center coordinates located by Hough circle as the growth starting point, and achieving precise traversal of defect regions through dynamic radius constraints. In tests on 80 typical CFRP drilling samples, this method achieved a detection accuracy of 93.46%, an improvement of 16.73 percentage points over traditional morphological methods. Processing time per image was controlled within 1600 ms, providing a quantitative decision-making basis for real-time quality inspection in industrial settings. The research results have significant engineering application value for the intelligent manufacturing of high-precision composite components.

Keywords: Control Theory and Applications, Process Control Systems
Abstract: This article presents a homothetic tube-based adaptive model predictive control (MPC) strategy to handle discrete-time linear time-invariant systems with parametric uncertainties and hard constraints imposed on the states and the control inputs. The proposed solution systematically fuses a gradient descent-based adaptive parameter identification strategy with a suitably designed tube-based MPC. An estimated model is utilized in the MPC for the purpose of state predictions. The parameters of the estimated plant model are updated at every time instant through an adaptive update law by utilizing the measured states and inputs from the uncertain plant. The task of satisfying the hard constraints in the presence of errors in state predictions, arising due to model mismatch between the estimated model and the uncertain plant, is accounted for by suitably tightening the constraints within the MPC optimization routine. It is guaranteed that the proposed tube-based adaptive MPC is recursively feasible if initially feasible, and the closed-loop states are bounded and asymptotically converging to the origin; the claimed properties are further validated through a simulation example.

Keywords: Robot Mechanism and Control
Abstract: This paper presents a multifunctional modular robotic platform based on Mori, enhanced through the integration of novel Two-DoF units with Linear-to-Rotary mechanism (TDLR-units). While the Mori platform features high reconfigurability through its origami-inspired triangular lattice structure, its ability to interact with external objects has been limited. To address this, a novel TDLR-unit is introduced, enabling interaction by converting linear actuation into rotational motion via a linkage mechanism. Three TDLR-units were integrated into Mori, allowing for 6-DoF across the system. Experimental validation demonstrated successful execution of various tasks including adaptive object gripping and crawling locomotion. The system effectively handled irregular and multiple objects, and performed coordinated surface movements. These results confirm that the TDLR-units significantly enhance Mori’s end-effector-level interaction capabilities and expand its functional range. This study establishes a foundation for modular robots with physical interaction capabilities. Future work will focus on extending functionality through integration with various module types for broader application scenarios.

Keywords: Robot Mechanism and Control, Robotic Applications, Control Theory and Applications
Abstract: Folding mechanisms, represented by origami structures, serve as alternatives to conventional rotational components by providing passive compliance and reducing the need for complex joint assemblies, thus offering compactness and a high degree of rotational freedom. Due to these structural advantages, folding mechanisms are widely used in modern robotics. However, compared to conventional systems, they are not commonly employed in applications requiring precise control, primarily because analytical modeling is more difficult. MOZART is a multi-functional robotic module that incorporates three parallel folding mechanisms, but it currently lacks an analytical model and therefore relies on limited control methods. In this study, we present a dynamic model of MOZART that includes parallel folding systems and attempt to validate it through trajectory tracking experiments. The results show that model-based control using the proposed model demonstrates superior tracking performance compared to conventional control methods.

Keywords: Autonomous Vehicle Systems, Robot Vision, Robotic Applications
Abstract: Global localization is a critical capability for autonomous driving systems. LiDAR-based methods are particularly effective in GNSS-denied environments. In order to support the evaluation of global localization performance in diverse scenarios, publicly available multi-session datasets are widely used. However, the ground truth poses provided by the datasets may contain inaccuracies or inter-session misalignments. Inaccurate poses can degrade both the reliability of benchmarking results and the quality of training data used in learning-based localization methods. In this work, we propose a factor graph-based method for refining ground truth poses in multi-session datasets. Our approach integrates both intra-session and inter-session constraints into a unified graph optimization framework. We evaluate the proposed method on the NCLT and Mulran datasets. The refined poses significantly improve the consistency of the resulting 3D point cloud maps. The improved ground truth enables more accurate benchmarking and supports the generation of higher-quality training data for learning-based global localization methods.

Keywords: Artificial Intelligence Systems, Robot Vision
Abstract: In this paper we implemented an elemental technology for robotic arm based On-Orbit Service that generates 3D shape information from images captured by a 2D camera during rendezvous and infers pose data to enable the manipulator to grasp diverse and complex structures of a target satellite. On-Orbit Service requires prior generation of 3D shape information of the target structure. We simulated the rendezvous process with the target satellite using Ansys STK (Systems Tool Kit) to acquire image data and generated 3D shape information from the captured images using NeRF(Neural Radiance Fields). The results confirmed that 3D shapes can be reconstructed from images obtained during the rendezvous process. This study is expected to provide startups and researchers planning future On-Orbit Service missions with payload grasping information for robotic arms using neural radiance fields on irregular objects.

Keywords: Control Theory and Applications
Abstract: In this paper, the global output feedback control scheme for a class of nonlinear systems with unmeasured state dependent growth and saturated input is considered. The nonlinearities rely on both the unmeasured state and the unknown constant. To cope with the unmeasured state, a high-gain observer is brought in. In order to solve the problem of saturated input, a new auxiliary system with constant gain is introduced, and the linear growth condition with unknown growth rate is eliminated. Significantly different from closely related literatures, although the growth rate is an unknown constant, global stability can be still obtained by introducing the constant not dynamic high gain, which reduces the dimension of the closed-loop system. In addition, the backstepping approach is used to present a systematic output feedback control scheme. An observer-controller design framework is put forward, ensuring the closed-loop system remains globally bounded. In the end, a practical example is provided to demonstrate the effectiveness of the proposed method.

Keywords: Control Theory and Applications
Abstract: This paper investigates the adaptive output-feedback tracking control problem of a robot arm system with asymmetric dead-zone input. In order to address the unknown dead-zone parameters, unmeasurable states, and nonlinear characteristics in the system, rather than constructing the dead-zone inverse, this paper proposes an adaptive control method based on a dynamic high gain observer. By transforming the dynamic model of the robotic arm into a nonlinear system with uncertain terms and analyzing the nonsymmetric properties of dead-zone input, the study designs an adaptive controller integrated with dynamic gain adjustment and estimation mechanisms. By constructing Lyapunov functions, it is theoretically proven that all signals in the closed-loop system are globally uniformly bounded, and the tracking error can converge to any predetermined neighborhood. This study expands the application scope of adaptive control in the field of robotics, which is of significance for promoting the development of robot control technology. Finally, we provide simulations to demonstrate the effectiveness of our controller.

Keywords: Robot Mechanism and Control, Control Theory and Applications, Robotic Applications
Abstract: This paper proposes a collision detection algorithm of a rectangular mobile robot by representing the robot using angles and distances, similar to a circular robot. The proposed algorithm detects collisions by evaluating the distances to the obstacle and robot’s surfaces. The proposed method consists of representation of the rectangular robot, and compensating for angle distortion at a point on the surfaces. By employing this novel representation method, the rectangular robots with various lengths and edge shapes can be accurately represented. This enhances path planning and obstacle avoidance performance. The proposed algorithm improves the possibility of passing through narrow environment compared to circular representation, and reduces computation power during obstacle detection than polygonal description.

Keywords: Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: In LiDAR-based SLAM, scene recognition algorithms are essential for tasks such as loop closure detection and robot localization within a constructed 3D map. Unlike camera-based methods that rely on texture features extracted from images, LiDAR sensors perform scene recognition by matching structural features of the environment based on distance measurements. This paper presents a method for scene recognition that extracts structural features from the spatial relationships between the LiDAR sensor and its surroundings, and utilizes the DBoW (Discrete Bag of Words) algorithm for recognizing previously visited scenes. The proposed method employs a solid-state LiDAR sensor with a horizontal field of view of 120 degrees, a vertical field of view of 35 degrees, and a resolution of 192×56. The effectiveness of the method is validated using data collected from various environments.

Keywords: Artificial Intelligence Systems, Robotic Applications
Abstract: The development of robust perception systems for surgical robots requires large-scale surgical environment datasets. However, real-world medical data collection faces significant privacy and safety restrictions. This results in scarce training data for developing robotic perception systems. To address this data scarcity, we propose a diffusion-based pipeline that generates realistic surgical scenes through two key steps. First, we preprocess limited real images by automatically generating text prompts using an image captioning model. This creates a text-image paired dataset without any manual annotation. Second, we fine-tune a latent diffusion model with Low-Rank Adaptation (LoRA) to efficiently capture surgical domain features. Our experimental results demonstrate high-fidelity synthetic images that replicate surgical components such as MRI machines and surgical monitors. The generated images successfully replicate authentic operation room styles and environments. These realistic surgical scene syntheses provide a practical data augmentation framework for training perception systems in next-generation surgical robotics.

Keywords: Artificial Intelligence Systems, Robot Mechanism and Control
Abstract: Model free deep reinforcement learning algorithms, especially Soft Actor-Critic (SAC), have achieved excel lent sample efficiency and stability in high-dimensional continuous control tasks, but their traditional Gaussian strategies have limitations in expressing multi-modal action distributions. Meanwhile, Diffusion Model, a highly expressive gen erative model, has been used for offline reinforcement learning to accurately fit complex behavioral distributions. To address the above challenges, this paper proposes Denoising Diffusion-AC, an online learning algorithm that integrates conditional diffusion policies into the Actor-Critic framework. Specifically, we use the denoising diffusion probabilistic model (DDPM) implemented by multilayer perceptual machine (MLP) as a policy network to model the action distribu tion through forward noise diffusion and backward denoising process; at each step of the backward process, we inject the gradient of the value function estimated by the dual Q-network to achieve noise- guided Q-learning, and balancing strategy exploration and exploitation by jointly optimizing the noise prediction loss and the value improvement loss. We conduct large-scale experiments in classical continuous control environments such as Ant,Hopper and Walker2d, and the results show that Denoising Diffusion-AC outperforms the baseline of several algorithms of the traditional AC algorithm in terms of convergence speed, final performance and strategy robustness, validating the diffusion strategy in the effectiveness and advantages of online Actor-Critic learning.

Keywords: Sensors and Signal Processing, Artificial Intelligence Systems
Abstract: This paper presents a method for estimating golf club motion during a golf swing. An inertial sensor and a ball marker are mounted on the golf club shaft. The sensor provides inertial data, while a depth camera captures both the ball’s position and the orientation of the golf shaft. By fusing these data, the trajectory of the golf club is accurately tracked throughout the swing. This approach also lays the foundation for incorporating human pose estimation in future work.

Keywords: Robot Vision, Artificial Intelligence Systems
Abstract: Active Label Correction (ALC) is a method for identifying incorrect labels in a dataset and refining them via human review. Recently, ALC has been applied to semantic segmentation using predictions from foundation models as pseudo-labels to reduce annotation cost. However, these pseudo-labels contain substantially inaccurate boundaries and numerous unlabeled pixels. We propose a three-stage pseudo-label upgrade pipeline consisting of label map generation, prediction map inference, and pseudo-label construction, which effectively addresses the aforementioned limitations. Qualitative results demonstrate the superiority of our method in producing accurate and complete annotations.

Keywords: Artificial Intelligence Systems, Multimedia Systems, Industrial Applications of Control
Abstract: Hand-drawn sketches offer an intuitive interface for 3D content creation, yet real-time conversion to textured meshes remains challenging due to depth ambiguity and sparse geometric cues. This work outlines a zero-shot sketch-to-3D pipeline that progressively refines geometric and appearance estimates to produce rich 3D models in under ten seconds. Starting from a sketch, the method derives multi-view imagery, infers depth and coarse point distributions, and regresses high-fidelity mesh surfaces. By enforcing consistency across multiple views and exploiting rapid, feed-forward inference, the pipeline avoids per-instance training and heavy optimization. Empirical evaluation on varied hand-drawn inputs demonstrates accurate geometry, faithful texture reproduction, and stable topology across object categories. Qualitative and quantitative comparisons against baseline methods demonstrate the pipeline’s efficiency and superior output quality. The plug-and-play design requires no fine-tuning or category-specific templates, supporting interactive sketch-based workflows in VR/AR prototyping, metaverse integration, and beyond. These findings suggest that runtime-efficient sketch-driven 3D modeling is practical, paving the way for broader accessibility in 3D content creation.

Keywords: Artificial Intelligence Systems
Abstract: We propose a pipeline for detecting and removing harmful objects from multi-view images and reconstructing 3D point clouds. Harmful content, such as weapons, blood, and cigarettes, presents challenges for deploying 3D vision systems in safety-critical or ethical contexts. Existing methods assume clean inputs and lack mechanisms for content-aware filtering. Our approach addresses this gap by integrating four key components: harmful object detection, multi-view object tracking, view-consistent object removal, and 3D reconstruction. We train the detection module on the HOD dataset and apply the full pipeline to real-world multi-view scene from Mip-NeRF 360. Experimental results show that our method can successfully remove harmful objects while preserving spatial and geometric consistency across views. This work offers a practical solution for content-aware 3D reconstruction and highlights its potential for applications in robotics, AR/VR, and digital content generation.

Keywords: Robot Vision, Sensors and Signal Processing, Human-Robot Interaction
Abstract: Active stereo vision is one of the most effective ways to obtain 3D information for robotic applications. Currently, it is still very difficult to reconstruct 3D shape of a continuous dynamic object in real time. An effective solution to cope with this problem is to design the structured light pattern with wide applicability and to develop the stereo vision matching algorithm with high accuracy. In this article, we propose a novel stereo vision matching method based on the structured light line pattern that is designed with four colors in the following cycle: cyan, red, light-green, red, purple, red, light-green, red, purple and red. The lines in different color in both camera views are extracted based on the HSV color model and threshold selection. After the center alignment, the cyan lines are matched based on a line indexing algorithm. Then, the light green lines and the purple lines are matched based on the matched cyan lines respectively. Finally, the red lines are matched based on the matched cyan, light green and purple lines. Experimental results indicated that the proposed four-color-line structured light pattern and its stereo vision matching method are robust in reconstructing the dynamic objects.

Keywords: Robot Vision, Artificial Intelligence Systems, Autonomous Vehicle Systems
Abstract: To effectively apply artificial intelligence in real-world environments, it is essential for models to recognize unfamiliar classes and handle unforeseen situations. Generalized Category Discovery in Semantic Segmentation (GCDSS) addresses this need by requiring models to cluster unlabeled classes at test time. Despite its potential, previous work on GCDSS has lacked realistic and convincing scenario settings. To bring this task closer to practical application, we shift the focus from urban outdoor scenes to indoor environments. Indoor spaces contain scene-specific objects. For instance, classrooms and living rooms do not contain beds, whereas bedrooms do. By identifying such space-specific classes and defining them as novel classes, we propose a more realistic formulation of the GCDSS task. Furthermore, our experiments and analysis reveal critical limitations of existing GCDSS methods in practical settings. These findings underscore the challenges of real-world deployment and suggest important directions for future work.

Keywords: Robot Vision, Artificial Intelligence Systems, Robotic Applications
Abstract: Visual Place Recognition (VPR) is a crucial technique used to locate mobile robots in large-scale environments. Recent VPR methods only focus on achieving high performance, which requires a large amount of computational resources. However, in real-worlds, due to the limited amount of resources, deploying off-the-shelf large models on mobile robots is challenging. To address this problem, we propose a simple yet effective distillation method called Structural Consistency Knowledge Distillation (SCKD), which transfers batch similarity from a large teacher network to a lightweight student network. By aligning the batch similarity, SCKD ensures the student model effectively captures the structural relationship within the teacher’s feature space. We conduct extensive experiments on 8 large scale benchmarks. Our experimental results demonstrate SCKD significantly outperforms previous state of-the-art methods.

Keywords: Industrial Applications of Control, Artificial Intelligence Systems
Abstract: We propose a pipeline to generate high-quality, diverse anomaly datasets for training anomaly detection models using only few-shot anomaly data (three samples) from the MVTec AD dataset. The approach leverages the FLUX.1-dev diffusion model to restore damaged anomalous images, fine-tunes the model with restored and damaged image pairs, and synthesizes new anomalous samples during inference by incorporating a normal (good) image as a contextual input. Experiments on the MVTec AD dataset show that the pipeline produces effective datasets for most anomaly classes.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems
Abstract: In real-world scenarios, obtaining complete 3D shapes as ground truth for point cloud completion is inherently challenging due to sensor limitations and occlusions. As a result, recent research has focused on self-supervised learning approaches that do not rely on explicit ground-truth supervision. However, many of these methods depend on generating supervision cues using multi-view images or artificially synthesized partial-complete pairs, which may not generalize well to arbitrary real-world cases. To address this challenge, we introduce a novel two-stage self-supervised approach leveraging structural priors represented by a learnable discrete codebook. In the first stage, our network reconstructs full 3D shapes from partial observations while distilling and storing salient local geometry into a learnable codebook, thus accumulating a repository of class-level structural priors. In the second stage, a Transformer-based attention module dynamically queries this codebook to retrieve and integrate pertinent structural cues, refining local features for more coherent and accurate completions. Our experiments show that incorporating structural priors via the proposed codebook improves the model’s ability to infer missing regions, yielding consistent gains on benchmark datasets.

Keywords: Autonomous Vehicle Systems
Abstract: This paper presents the development of a remote driving interface system for real full-scale autonomous vehicles. The proposed system integrates multi-camera video feeds and physical input devices to enable intuitive and responsive remote control. A Hyundai Santa Fe DM vehicle, modified for autonomous driving, captures its surroundings using six high-resolution cameras, and the video feeds are displayed through a unified user interface. Remote control commands comprising steering and pedal data generated from input devices are transmitted via a Zao On-Premise Server to the vehicle. These commands are then delivered to the motors through an I/O terminal installed on the vehicle, enabling actual vehicle movement.

Keywords: Human-Robot Interaction
Abstract: This study explores the effectiveness of robot-assisted intervention for children with ASD, focusing on the relationship between emotional expression and clinical severity. Based on the Theory of Mind framework, occupational therapists conducted emotion-focused therapeutic sessions using the humanoid robot NAO. During the sessions, children’s facial expressions were recorded and analyzed as time-series data through an automated emotion recognition system using DeepFace. The relationship between emotional indicators and autism severity was examined, revealing a trend in which children with higher severity levels exhibited reduced expressions of sadness and a greater proportion of positive affect. An N-of-1 analysis further highlighted visual differences in emotional expression patterns between children with moderate and severe ASD. These findings inform the design of personalized robot-assisted interventions.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Robot Vision
Abstract: Knowledge distillation for vision-language models such as CLIP is essential for efficient deployment in resource-constrained environments. However, existing methods often neglect explicit alignment of cross-modal relationships. We propose a novel multi-level distillation framework employing embedding-level cross-attention and affinity-level relational distillation, aggregating multiple image-text affinity matrices for richer cross-modal guidance during student training. Extensive experiments demonstrate that our approach effectively preserves cross-modal relational structures, significantly improving zero-shot classification and cross-modal retrieval performance.

Keywords: Robot Vision, Robotic Applications, Autonomous Vehicle Systems
Abstract: Accurate large-scale maps are essential for reliable navigation of autonomous vehicles and mobile robots in outdoor environments. However, SLAM approaches suffer from drift and high computational costs over extended areas, while pure 3D reconstruction methods incur prohibitive processing times and similarly degrade at scale. This paper introduces a fully automated map-merging framework that fuses multiple vision-based 3D point cloud submaps into a single, large-scale map. Each submap is generated by applying ORB-SLAM3 for RGB-D pose estimation followed by 3D reconstruction via COLMAP. Overlapping regions are estimated automatically by extracting ORB descriptors, matching via DBoW3, and back-projecting correspondences into 3D space; DBSCAN clustering then removes outliers to yield robust overlap subsets. These subsets drive FPFH and TEASER++ registration, whose resulting transformations align all submaps in a common coordinate frame. The proposed method significantly enhances automation and robustness in large-scale map construction.

Keywords: Artificial Intelligence Systems, Robotic Applications, Robot Vision
Abstract: With the rapid development of intelligent industrial inspection robots, enterprises are increasingly adopting robotic solutions for industrial production. Industrial defect detection plays an important role in quality assurance. While machine vision-based automatic optical inspection(AOI) technology has achieved high-precision detection for flat surfaces, significant challenges remain in curved surface inspection. To address this gap, we propose an Intelligent Vision-Based Industrial Inspection Robot capable of performing comprehensive scanning and image acquisition of curved workpieces. By analyzing collected images, we have established both simulated and real-world datasets and developed a novel high-resolution industrial defect detection method, PLPatch. Our approach employs detection unit pre-positioning and localized texture feature memory bank to achieve complete image defect detection. Experimental results demonstrate that this method outperforms existing state-of-the-art techniques on both simulated and real datasets.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: In head-to-head autonomous racing, minimizing lap time while overtaking opponent vehicle presents a crucial control challenge. In this work, we propose a switching MPC framework that selects between two distinct MPC formulations depending on the racing context: one for pure time-optimal racing and another for overtaking situations. The framework classifies scenarios based on the relative position of the opponent and dynamically switches the cost function and constraints to suit each case. The proposed approach is evaluated in a 1:10 scale head-to-head autonomous racing simulator environment. Experimental results demonstrate that the switching MPC achieves consistently lower lap times and performs more successful overtaking maneuvers compared to a single MPC.

Keywords: Sensors and Signal Processing, Navigation, Guidance and Control, Artificial Intelligence Systems
Abstract: This paper introduces a lightweight deep learning framework to improve the reliability of inertial sensors in high-speed autonomous racing. Traditional IMU-based Dead-Reckoning suffers from drift and noise under dynamic conditions, while external sensors like GPS and cameras can be unreliable due to environmental disturbances. To overcome these issues, we propose the De-Noise IMU Net—a dual-path convolutional model that denoises IMU signals by extracting both temporal and pointwise features. The fused signal is then processed by an Unscented Kalman Filter for accurate state estimation. Experiments using the F1TENTH platform demonstrate that our method reduces the Absolute Trajectory Error(ATE) by 79% and Absolute Rotation Error(ARE) by 85%, while maintaining real-time performance with minimal computational overhead, demonstrating its effectiveness in enhancing accuracy and robustness for autonomous racing.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: End-to-End driving systems which map raw sensor input to control actions via deep neural networks have been studied. However, they are vulnerable to domain shifts (e.g., weather changes), often leading to collisions in different domains. In this paper, we propose Spatio-Temporal Fusion Module for End-to-End driving network to mitigate the impact of domain shifts and reduce collision scenarios. We evaluate our method in the CARLA[3] simulator and demonstrate performance in nighttime driving scenes, achieving a 35.9% improvement in driving score over baseline model.

Keywords: Robot Vision, Artificial Intelligence Systems, Robotic Applications
Abstract: Understanding dynamic 3D scenes at both geometric and semantic levels is increasingly important in applications such as robot perception, autonomous navigation, and AR/VR systems. Recent advances in 3D Gaussian Splatting have demonstrated impressive performance in photorealistic rendering and dynamic scene reconstruction. However, existing dynamic splatting methods are limited in their ability to incorporate semantic understanding, while approaches like Feature 3DGS enable semantic feature representation but are restricted to static scenes. In this work, we propose a unified framework that empowers 4D Gaussian Splatting with per-Gaussian semantic feature embeddings alongside deformation embeddings, enabling semantically-aware modeling of dynamic scenes. Experimental results demonstrate the effectiveness of our approach in novel view synthesis of dynamic environments, while also producing high-quality feature renderings that reveal scene semantics across time.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Robotic Applications
Abstract: The applications of continuum robots are widespread, including use in industrial automation, aerospace systems, and medical procedures. Unlike conventional rigid-link robots, continuum robots possess a continuously flexible structure, enabling them to operate effectively in narrow or curved pathways. Among various types of continuum robots, tendon-driven designs offer several key advantages, such as high force transmission, high flexibility, and lightweight construction. To accurately control the position of the robot tip, a control framework is essential. This paper presents a control framework specifically designed for tip control of a tendon-driven continuum robot. The proposed framework utilizes force-based control, as force modeling provides a more accurate representation of the robot’s behavior compared to constant curvature models. A proportional control strategy is employed, based on position feedback obtained from the model. Simulations are conducted under two conditions: without external load and with external load, to evaluate the performance of the controller. The results demonstrate the controller’s effectiveness in enabling the continuum robot tip to follow the desired trajectory, even in the presence of external load.

Keywords: Artificial Intelligence Systems, Robot Vision, Industrial Applications of Control
Abstract: In manufacturing industries, the accurate detection of steel surface defects is critical for ensuring product quality and safety. However, supervised learning-based defect detection systems face significant challenges due to data scarcity and class imbalance in industrial environments. In this paper, we propose a pipeline that combines diffusion model-based data augmentation with pseudo-labeling and a dual-decoder architecture. Our approach first utilizes a fine-tuned conditional diffusion model to generate synthetic defect images with diverse characteristics and backgrounds. To obtain supervision for these unlabeled synthetic images, they are processed through a pretrained segmentation model to create pseudo-labels. These pseudo-labels are then combined with original labeled data to form an augmented training dataset. A dual-decoder segmentation network is trained on this augmented dataset, performing both multi-class and binary segmentation tasks simultaneously. Experimental results on the Magnetic Tile dataset demonstrate the effectiveness of our method, achieving significant improvements of 2.46 % in mIoU compared to baseline approaches, validating the practical applicability in real industrial environments.

Keywords: Biomedical Instruments and Systems
Abstract: Living organisms maintain internal stability through a process called homeostasis, which is essential for health. The disruption of homeostasis can lead to diseases such as metabolic syndrome. Recent studies have suggested that homeostasis is governed by feedback control, and biomolecular Proportional-Integral-Derivative (PID) control mechanisms are gaining attention as biologically implementable systems. This study focused on glucose metabolism, assuming it is regulated by biomolecular PID control, and aimed to estimate its control parameters from time-series data of humoral factors.

Keywords: Industrial Applications of Control, Human-Robot Interaction, Robotic Applications
Abstract: In recent industrial environments, the demand for collaborative robots is rising. However, current robots face high costs due to expensive sensors and task-specific grippers that vary with the environment. This study proposes a novel system consisting of two 2-DOF robots equipped with high-friction ball-type grippers and no force/torque sensors. The goal of this system is to lift objects of varying sizes to a target height. Force control is achieved through reinforcement learning using the Proximal Policy Optimization (PPO) algorithm. Training was performed on boxes with widths of 10 cm and 15 cm. Both cases successfully reached the target height of 15 cm. During the motion, the applied torque on each actuator was recorded, with a maximum torque of 4 Nm, confirming compatibility with the real robot’s torque limits. This approach demonstrates that effective cooperative manipulation is possible even with minimal hardware. Future work will implement this trained model on a real robot and add cameras to detect box positions. The final goal is to develop a fully integrated sensorless cooperative force control system suitable for diverse industrial manipulation tasks.

Keywords: Sensors and Signal Processing
Abstract: In this study, we propose a novel wearable device to detect the motion of the lower limb using piezoelectric sensors. The wearable device comprised twelve piezoelectric sensors and body supports to fix the sensors at the lower limb joints. Specifically, we detected six reference movements per leg and estimated each lower limb joint angle using the measured data from the attached sensors at the joints. By comparing the estimated angles with actual angles measured by infrared trackers, we have a good agreement between the sensing values and the joint angles.

Keywords: Robotic Applications
Abstract: Multi-sensor localization is one of the core components in the robotics and autonous driving. In particular, LiDAR-Visual-Inertial (LVI) localization enables highly accurate pose estimation, but remains sensitive to initial pose estimation and is prone to registration failures caused by errors in the rotational component. To address this issue, this paper proposes a robust initialization method that leverages the matching between 3D line features extracted from a LiDAR prior map and 2D line features detected in camera images. The proposed method adopts a two-stage strategy to optimize the initial pose, enabling stable pose estimation even in environments with large rotational errors. Experimental results in real factory environments demonstrate that the proposed approach achieves improved accuracy and robustness compared to conventional localization methods.

Keywords: Sensors and Signal Processing, Process Control Systems, Industrial Applications of Control
Abstract: We present a 20 mm diameter, low-cost six-axis force/torque fingertip sensor that uses MEMS barometers embedded in a compliant spring–polymer architecture. Each module contains two closely spaced barometers on a rigid PCB, encapsulated in vacuum-molded Vyta 20 polymer. A 0.5 mm thick SUS304 leaf spring separates shear and normal forces, raising the stiffness to 50 kN/m under 100 N loads. Simulations and experiments confirm linearity and a usable response range into the tens of newtons. Ongoing work focuses on complete six-axis calibration and integration into the LEAP Hand platform.

Keywords: Robot Mechanism and Control, Human-Robot Interaction, Control Theory and Applications
Abstract: Strawberry harvesting remains a major challenge in agricultural automation because of the fruit’s soft texture, irregular shape, and densely clustered growth. This paper presents a mechanically actuated, three-fingered robotic gripper specifically designed for strawberry harvesting. The proposed gripper establishes four stable contact points and features a geometry optimized for reaching small fruit that grow in clusters. To ensure effective and adaptive grasping, a hybrid control strategy is adopted: the metacarpophalangeal (MCP) and proximal interphalangeal (PIP) joints are position-controlled for accurate posture, while the DIP joints employ mass–spring–damper (MCK) control to provide compliance with the fruit’s surface. Grasp postures are generated by applying inverse kinematics to the estimated size and centroid of the strawberry obtained through object recognition. As a potential improvement, we also propose enhancing the gripper's contact surface by applying high-friction, deformable materials such as silicone to increase stability and reduce fruit damage in real harvesting environments.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Navigation, Guidance and Control
Abstract: In real-world environments, wheel-based mobile robots often encounter non-traversable terrain, such as stairs or steep slopes, which can result in navigation failures or mechanical hazards. However, conventional exploration methods fail to evaluate terrain traversability, resulting in unsafe trajectories. To address these limitations, we propose a traversability-aware exploration framework that enhances safety by integrating terrain analysis into the path planning process. First, traversable regions are detected based on multiple terrain conditions, while accounting for hazardous areas such as stair edges that may be misclassified due to sensor blind spots. Second, traversability-aware path filtering is applied, where candidate paths are evaluated using PCA-based terrain features during A* search to exclude unsafe paths. Based on FAEL's framework, the proposed method is applied, and an experiment is conducted in a simulation environment that includes a non-traversable area. As a result of the experiment, the proposed method achieves a higher success rate compared to the existing method.

Keywords: Autonomous Vehicle Systems, Robot Vision, Sensors and Signal Processing
Abstract: Pedestrian detection is a fundamental component of autonomous mobility systems. However, among all pedestrians, those unaware of an approaching vehicle require particular attention. This paper proposes a Diffusion-based Pedestrian Attention Classification (DPAC) module that determines whether a pedestrian is aware of the ego vehicle in real-world driving scenarios. The proposed module takes bounding boxes from a detector as input and classifies the pedestrian’s attention state. Leveraging a diffusion model, the DPAC module maintains robust performance even under low-resolution inputs. It is lightweight in both computation time and memory usage, enabling real-time deployment directly following detection modules. The module is validated on the Joint Attention for Autonomous Driving (JAAD) dataset, which includes annotations for pedestrian attention. Experimental results demonstrate superior performance over existing methods. When attached to a detector, the complete system processes a 1920 times 1080 image—including detection and attention classification—within 33 ms. The results confirm both the effectiveness and scalability of the proposed DPAC module for integration into diverse perception pipelines.

Keywords: Rehabilitation Robot, Human-Robot Interaction, Exoskeleton Robot
Abstract: This paper proposes a textile-based artificial muscle system for wearable assistive robotics, integrating shape memory alloy (SMA) actuators with embedded temperature, displacement, and force sensors. A silver-coated thread temperature sensor is woven perpendicularly to the SMA wire for localized thermal monitoring, while a EGaIn displacement sensor is directly coupled to the actuator structure for precise motion measurement. A compact load cell is incorporated to provide real-time monitoring of contraction force. All sensors are seamlessly embedded in the textile, maintaining softness, breathability, and comfort without bulky external modules. The actuator module is designed to support the back, assisting spinal motion and enabling real-time feedback and adaptive control. This integrated design manages both motion tracking and thermal regulation internally, enhancing system efficiency and user comfort. Experimental validation demonstrates accurate detection of temperature, displacement, and force changes, with stable and repeatable performance. The proposed system establishes a promising platform for next-generation smart textile actuators and offers broad applicability in wearable robotics, rehabilitation, and assistive technologies.

Keywords: Robot Vision, Artificial Intelligence Systems
Abstract: Deformable Linear Objects (DLOs) are widely used in various industrial applications, including medicine, manufacturing, and agriculture. Accurate shape estimation is essential for control tasks; however, occlusion hinders reliable tracking, especially in the absence of markers. This paper introduces a real-time DLO position estimation method that is robust to occlusions, using B-splines and the robot's Tool Center Point(TCP) information. We generate initial 2D estimates for the DLO nodes using RT-DLO and fit the points to a B-spline curve. A new set of nodes is then created in each camera image and used for stereo matching. The resulting 3D nodes are validated against marker-based ground truth. Experimental results demonstrate that with 100 generated nodes, the proposed method achieves a mean position error of 6.92 mm without occlusion, and 10.55 mm with occlusion.

Keywords: Industrial Applications of Control, Process Control Systems, Control Theory and Applications
Abstract: This paper proposes a smart charging strategy for lithium-ion batteries based on quadratic programming (QP) under temperature and aging constraints. The method dynamically adjusts the charging current by solving a QP problem at each time step to achieve target SOC while minimizing capacity loss. The proposed strategy integrates an aging model based on solid electrolyte interphase (SEI) growth and active material (AM) degradation and enforces voltage and core temperature limits using linearized inequality constraints. Simulation results under varying ambient temperatures demonstrate the strategy’s ability to limit thermal stress and reduce degradation risk during fast and prolonged charging.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: Autonomous drones are being researched in various fields such as exploration, rescue, fault detection, and delivery, and are drones that perform their assigned tasks without direct human control. These drones are powered by lithium polymer batteries to accomplish their various missions. However, these LiPo batteries have a limited capacity, which limits the amount of time they can operate. Therefore, predicting the future flight time of an autonomous drone is crucial for reliable mission performance. In this study, we propose a method to predict the operational time of an autonomous drone flight using an LSTM-based model. The proposed method uses voltage and current data to predict the future operational time. Experimental results show that the proposed method can reliably predict the flight time.

Keywords: Robotic Applications, Biomedical Instruments and Systems, Human-Robot Interaction
Abstract: Maintaining a constant contact force between a robot end-effector and a target surface is critical in both industrial and medical applications. Conventional approaches rely on force sensors and position control algorithms; in this study, we propose a low-cost visual-feedback method that uses an RGB camera to adjust control parameters in real time by recognizing the contact state between the end-effector and the surface. Experiments show that, after fine-tuning a pretrained neural network, the proposed scheme accurately identifies contact conditions and rapidly adapts motion parameters, thereby improving response speed.

Keywords: Robotic Applications, Biomedical Instruments and Systems
Abstract: Vitreoretinal surgery is an exceptionally delicate procedure that demands high precision throughout its duration. Vision-based assistance plays a crucial role in trocar positioning and the safety of autonomous ophthalmic interventions. Here, we introduce an attention-augmented ResNet-backbone U-Net for accurate trocar segmentation in intraocular images. Evaluated on a custom dataset comprising both eyeball phantoms and real trocars, our model achieves a Dice coefficient of 98.53%, outperforming the conventional U-Net’s 94.25% on the same data. This enhanced segmentation capability enables reliable, real-time tool localization and trajectory planning, demonstrating significant potential to improve the safety and efficiency of next-generation autonomous eye surgeries.

Keywords: Robotic Applications, Artificial Intelligence Systems
Abstract: We present a real-time ball balancing system built on a 6-DoF UR5e robotic arm, using a resistive touch panel for tactile feedback and a Proximal Policy Optimization (PPO) agent for adaptive control. The system operates entirely with velocity commands and leverages low-latency (x, y) feedback from the panel to stabilize a free-moving ball on a flat surface. Control trajectories are generated via cubic spline interpolation through sampled Cartesian waypoints, converted into joint-space commands using inverse kinematics, and filtered to ensure physical feasibility. The entire policy is trained in simulation and transferred directly to the physical UR5e platform. In contrast to prior works that rely on high-DoF or torque-controlled manipulators, this setup utilizes standard industrial hardware and low-cost sensors. Experiments demonstrate reliable stabilization under various conditions, validating the feasibility of reinforcement learning for real-time dynamic control tasks in practical, resource-constrained environments.

Keywords: Rehabilitation Robot, Robot Mechanism and Control, Human-Robot Interaction
Abstract: The increasing proportion of lower-limb amputees, particularly below-knee amputees, highlights the need for more efficient ankle-foot prosthesis designs. This study proposes a novel 2-DOF prosthesis that integrates a Unidirectional Parallel Elastic Actuator (UPEA) structure with a cable-driven actuation system, in which the actuators are relocated to the upper body (waist) region. Based on prior research, the metabolic cost model indicates that for added masses exceeding approximately 0.645 kg, positioning them at the waist is more energy-efficient than at the shank. Accordingly, the proposed design reduces the distal mass by relocating all components—except the UPEA—from the shank to the waist, thereby improving gait efficiency while maintaining propulsion capability. By combining the dorsiflexion energy storage feature of the UPEA with the metabolic advantages of proximal mass relocation, the proposed configuration is expected to enhance walking efficiency and contribute to long-term energy savings for below-knee amputees.

Keywords: Rehabilitation Robot, Robot Mechanism and Control, Human-Robot Interaction
Abstract: Prosthetic technology is a crucial solution for helping lower-limb amputees return to daily life. This study explores a control strategy for prosthetics that considers scenarios of balance loss. Traditional control methods rely on explicit modeling of balance recovery steps, which are difficult to optimize for individual users. To overcome this, a reinforcement learning-based approach is proposed. A bipedal robot model was developed using MATLAB Simscape, and the right leg was replaced with a prosthesis model. Initially, the model learned to walk on flat terrain using the TD3 algorithm. Then, obstacle-induced imbalance scenarios were introduced to train the prosthetic control in recovering from falls. The simulation results show that 1,000 training episodes were insufficient for stable gait, and at least 4,000 episodes are required for meaningful performance. This study confirms the feasibility of prosthetic control via reinforcement learning and suggests directions for future development in more complex environments and using various learning strategies.

Keywords: Rehabilitation Robot, Robot Mechanism and Control, Human-Robot Interaction
Abstract: This study proposes a powered ankle-foot prosthesis that replaces conventional motor-driven actuators with shape memory alloy (SMA) spring-based artificial muscles to reduce device weight and operating noise. We introduce a portable water circulation system that adjusts water temperature between 28°C and 83°C for efficient heating and cooling of the SMA springs. The system utilizes cartridge heaters and Peltier modules for thermal control, and a DC motor controlled mixing valve regulates the output water temperature. Compared to conventional prostheses, our design reduces component weight by approximately 236g and eliminates 42-58 dB of actuator noise. While physical constraints limited full scale SMA spring implementation, this research presents a promising step toward quieter, lighter, and more natural ankle prosthesis.

Keywords: Robot Mechanism and Control
Abstract: In this study, a novel leg-and-wheel transformable mechanism using origami cylinders was introduced. The mechanism demonstrated a bidirectional mode conversion between leg and wheel modes. In particular, each leg-andwheel transformable origami mechanism unit was based on Kresling pattern origami cylinders, tendon-driven operation methods, and a hybrid design structure. The Kresling pattern origami cylinders were deployed to deform the legs through extension, contraction, and bending with the tendon-driven methods. In addition, the hybrid design of soft and rigid components enabled the transformation for the mode conversion.

Keywords: Human-Robot Interaction, Biomedical Instruments and Systems, Robotic Applications
Abstract: Accurate and realistic transmission of haptic patterns is a crucial factor in advancing haptic devices. The challenge arises from the presence of both high-frequency haptic patterns, like textures, and low-frequency haptic patterns, such as shapes or curvatures, when interacting with real objects. This limitation becomes apparent when attempting to convey such diverse tactile signals through a single actuator. In this paper, we present a novel solution in the form of a pneumatic-electromagnetic integrated fingertip haptic device, designed to enhance the fidelity of sensory transmission. The pneumatic actuator is utilized for implementing the low-frequency component of any given haptic pattern. In contrast, the electromagnetic actuator takes on the high-frequency component, aligning with their respective mechanical characteristics. A series of experiments were conducted to validate the mechanical properties of our proposed haptic device. In conclusion, our innovative approach holds promise in significantly improving the accuracy and realism of haptic patterns transmission.

Keywords: Robot Mechanism and Control, Control Devices and Instruments, Robotic Applications
Abstract: As a linear actuator, the twisted string actuator (TSA) offers ease of miniaturization, flexibility, and the capability to exert a strong actuation force, due to its large gear ratio. However, the operating range relative to its size has been reported as limited. This study aims to overcome this limitation by extending the operating range of TSA using the coiling phase, following the twisting phase. To achieve this, we analyze the training process to effectively mitigate the irregular overlapping phenomena in the coiling stage of TSA, considering the hysteresis and the training effect of TSA under different loads. Additionally, through various experiments, we validate the proposed training process for extended TSA operation. It is expected that the proposed training process for TSA will enable precise actuation within the extended operating range, facilitating a wider array of applications.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems, Multimedia Systems
Abstract: We address coverage path planning on grid maps that contain thin barriers—local, direction-free edges that block movement across selected cell boundaries. Our hybrid framework first builds a TB-Spanning Tree linking barrier regions without creating cycles, then attaches Spiral-Spanning Trees to every remaining empty area. The union is acyclic, guarantees complete coverage, and can be generated in linear time. Formal analysis shows the resulting path length differs from an optimal Hamiltonian traversal by at most a small, leaf-dependent additive constant.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Control Theory and Applications
Abstract: In this paper, a hybrid learning scheme that combines data-driven and physics-based models is proposed to enable real-time estimation of vehicle cornering stiffness. Existing data-driven based cornering stiffness estimation methods require ground truth stiffness data as labels during training, which limits their applicability. To overcome this limitation, a loss structure is designed that allows training without any ground truth stiffness data. The proposed network is trained by minimizing residuals derived from physical dynamics equations, and the estimated stiffness values are utilized as parameters in a linear parameter-varying model predictive control (LPV-MPC) controller. Simulation results validate the improved driving stability under various road friction conditions.

Keywords: Control Devices and Instruments, Robot Mechanism and Control, Control Theory and Applications
Abstract: Tensegrity structures are unique structures formed by the balance of tension and compression members, and are attracting attention in various fields such as robots and aerospace due to their characteristics such as lightness, flexibility, shock absorption, and energy efficiency. In particular, geometric design is a key factor that determines the mechanical performance and range of motion of the structure. In this study, we analyze how the effective workspace that the top plate of a tensegrity prism structure can secure changes according to the change in the number of struts (3 to 6) by arranging the same struts with the same base radius. We mathematically define the reachable range according to the translational and rotational motion of the top plate, and obtain the effective range by reflecting the collision conditions between the struts. The analysis results show that as the number of struts increases, the stiffness and stability of the structure improve, but the range of motion tends to be limited, which provides useful design guidelines for the design of tensegrity-based robots and various devices.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Sensors and Signal Processing
Abstract: This paper investigates the impact of phase delays introduced by the Kalman Filter (KF) and Low-Pass Filter (LPF) on the string stability of Cooperative Adaptive Cruise Control (CACC) systems. We systematically analyze how variations in the KF noise covariance and LPF cutoff frequency affect the overall phase delay and, in turn, the minimum headway time required to maintain string stability. Using simulation-based analysis with realistic vehicle dynamics and estimation models, we quantitatively examine the trade-off between estimation accuracy and string stability, and provide phase-aware design guidelines to enable safe platooning without relying on inter-vehicle communication.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Control Theory and Applications
Abstract: This paper presents a Deep Neural Network-based Disturbance Observer (DNN-DOB) for lateral vehicle control, leveraging data generated from the CarSim simulation environment. The built-in CarSim controller is considered as a human driver, and its steering commands and vehicle states are used as reference data. The DNN-DOB is trained to estimate and compensate for the difference between the steering input of the CarSim controller and the output of a baseline Linear Quadratic Regulator (LQR) controller, thereby addressing unmodeled dynamics and external disturbances. A feedforward multilayer perceptron architecture is employed, with normalization applied to both inputs and outputs to ensure robust performance when deployed in Simulink via ONNX integration. The experimental workflow includes data preprocessing, normalization, neural network training, and real-time inference within Simulink. The DNN-DOB output is combined with the LQR command to generate the final steering input. Comparative analysis across various driving scenarios demonstrates that the proposed DNN-DOB-LQR significantly improves lane-keeping performance compared to the conventional LQR controller. This study highlights the effectiveness and practical feasibility of integrating data-driven disturbance compensation into traditional vehicle control frameworks using simulation-based reference data.

Keywords: Control Theory and Applications
Abstract: Generalized and combination synchronization have been well studied in real-valued nonlinear systems but are less explored in complex-variable dynamical systems. Identifying complex parameters on the synchronization manifold presents challenges due to the absence of persistent excitation (PE) conditions in the complex domain. This paper investigates complex generalized combination synchronization (CGCS) and parameter identification among three heterodimensional complex-variable chaotic systems (CVCSs) (or hyper-chaotic systems). Based on Lyapunov stability theory and employing adaptive control in the complex field, some sufficient conditions are derived to achieve CGCS for complex-variable chaotic systems (CVCSs). Furthermore, when all or some system parameters are unknown, synchronization and parameter identification criteria are rigorously formulated. The case where all parameters are known is also addressed within the proposed framework. These results are supported by theoretical analysis grounded in stability theory and complex analysis. Finally, numerical simulations are presented to validate the theoretical findings.

Keywords: Industrial Applications of Control, Control Theory and Applications, Robotic Applications
Abstract: We propose a novel high-force/high-precision interaction control framework of a dual-arm robot system on a flexible base, with one arm holding, or making contact with, a supporting surface, while the other arm can exert any arbitrary wrench in a certain polytope through a desired pose against environments or objects. Our proposed framework can achieve high-force/precision tasks by utilizing the supporting surface just as we humans do while taking into account various important constraints (e.g., system stability, joint angle/torque limits, friction-cone constraint, etc.) and the passive compliance of the flexible base. We first design the control as a combination of: 1) nominal control; 2) active stiffness control; and 3) feedback wrench control. We then sequentially perform optimizations of the nominal configuration (and its related wrenches) and the active stiffness control gain. We also design the proportional–integral type feedback wrench control to improve the robustness and precision of the control. The key theoretical enabler for our framework is a novel stiffness analysis of the dualarm system with flexibility, which, when combined with certain constraints, provides some peculiar relations, that can effectively be used to significantly simplify the optimization problem-solving and to facilitate the feedback wrench control design by manifesting the compliance relation at the interaction port. The efficacy of the theory is then validated and demonstrated through simulations and experiments.

Keywords: Exoskeleton Robot, Robot Mechanism and Control, Human-Robot Interaction
Abstract: This paper proposes a novel elastic element-based hip assistive device designed to support swing motion. The device functions as a passive hip assitive mechanism that provides assitance to the wearer during gait by utilizing the elongation and contraction of an elastic element. To facilitate the design of the device, a kinematic analysis is conducted to evaluate the elongation and contraction length of the elastic element according to the hip joint range of motion. In addition, a static analysis is performed to investigate the assistive force profile that can be generated at the hip joint for a given stiffness of the elastic element.

Keywords: Autonomous Vehicle Systems, Robot Vision, Robot Mechanism and Control
Abstract: LiDAR sensors have been widely used in autonomous navigation due to their ability to accurately perceive 3D spatial information. However, in heavy snowfall environments, point cloud data is severely degraded by snowfall noise, which significantly reduces the performance of autonomous navigation systems. To address this challenge, we propose a voxel-based hash-map Dual-Density desnowing Filter(DDF). DDF partitions space using an efficient voxel hash-map structure that stores only occupied cells, thereby minimizing memory usage and accelerating neighbor queries. The filter applies a dual-density criterion by combining two types of density: (1) Neighborhood density, which evaluates the local quantity of points by counting the number of neighboring points within a certain range, and (2) Structural Coherence density, which assesses the structural coherence of neighboring points by calculating the Gaussian-weighted density based on the spatial distribution of points. Simulation results demonstrate that DDF effectively removes snowfall noise while preserving true structural points, producing cleaner and more reliable point clouds for downstream tasks in adverse weather conditions.

Keywords: Rehabilitation Robot, Human-Robot Interaction, Biomedical Instruments and Systems
Abstract: Focal muscle vibration (FMV) can activate skeletal muscle via the tonic-vibration reflex while avoiding several drawbacks of functional electrical stimulation (FES). This study first mapped frequency–response curves of a Type I-dominant and Type II-dominant muscles with 60–100 Hz bursts; peak displacement occurred at 75 Hz and 100 Hz respectively, confirming fibre-specific optima. These bands were then combined into a dual-frequency sweep (75→100 Hz) and applied to the forearm flexor–pronator mass in three adults. Compared with fixed 75 Hz, fixed 100 Hz, or a low-band sweep (50↔75 Hz), the dual sweep increased grip force by 15 % immediately and sustained a 10 % gain 120 min later. The protocol therefore restores the natural slow-first / fast-second recruitment sequence lost in FES and delivers durable strength gains with a simple 100 Hz actuator. Despite limits in sample size, bandwidth, and fatigue metrics, frequency-swept FMV emerges as a low-fatigue adjunct for strength rehabilitation and athletic conditioning.

Keywords: Artificial Intelligence Systems, Process Control Systems, Control Devices and Instruments
Abstract: Despite the growing importance of intelligent HVAC systems, there has been limited research on cooperative control strategies for multiple indoor units to achieve both occupant comfort and energy savings. Modeling the effects of multi-unit control and dynamic thermal loads on indoor temperature poses significant challenges due to complex and variable system dynamics. In this study, we develop a robust temperature prediction model by combining physics-based differential equations and validate it with real operation logs. This model accurately reflects the influence of temperature setting and thermal loads. Leveraging this model as a simulation environment, we apply the Deep Reinforcement Learning (DRL) algorithm to optimize the control policy of each indoor unit. The RL agent is trained with a reward that balances energy penalty and tracking of a comfort temperature trajectory. We compare Deep Q-Network (DQN) and Proximal Policy Optimization (PPO). In simulations, PPO learned slowly, while DQN consistently achieved stable rewards, better temperature tracking, and lower energy consumption.

Keywords: Robot Mechanism and Control, Human-Robot Interaction
Abstract: This paper presents a novel leader system designed to provide intuitive and precise control of robots used in nuclear accidents and other disaster response scenarios. Traditional joystick-based interfaces and sensor-only robotic arms often suffer from limited precision, operator fatigue, and poor adaptability. To address these challenges, the proposed leader system incorporates actuated joints with gravity compensation, reducing the physical burden on the user. It mirrors the follower robot, ARMstrong, with a 3:1 scale and 16 degrees of freedom, enabling accurate motion mapping. An Adaptive Wheel-Mode Gripper with 2 degrees of freedom allows flexible operation of both dual-arm and mobile robots through a simple mode-switching mechanism. The system is controlled by a Raspberry Pi and communicates with the follower system via Wi-Fi, enabling real-time position exchange and feedback. Experimental results demonstrate suc cessful synchronization between leader and follower, confirming the system's effectiveness in enhancing control accuracy and operability. Future research will focus on optimizing gravity compensation and evaluating long-term usability in extended operations.

Keywords: Robot Mechanism and Control, Robotic Applications, Robot Vision
Abstract: Robot precision manipulation needs high-quality human demonstration data, especially with rich tactile feedback for fine force control and alignment. To effectively bridge the gap between human dexterity and robotic capabilities, we've developed a novel wearable optical tactile data collection system for human-robot teaching. This system centers on a wearable optical tactile gripper designed to naturally capture human manipulation. Mimicking industrial robot grippers like the Robotiq 2F-85 and integrating a high-resolution Meta DIGIT optical tactile sensor, our device allows human operators to perform intricate tasks while precisely recording their tactile interactions. A dedicated calibration module ensures accurate force and position data, minimizing the sim-to-real gap and making the collected teaching data highly effective for robot learning. The acquired tactile imagery, offering sub-millimeter position resolution and detailed force distribution, is crucial for enabling robots to achieve precision manipulation and perform tasks with human-like dexterity. We anticipate this system will significantly contribute to building large-scale, high-fidelity tactile datasets, advancing future robotic learning and automation.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Navigation, Guidance and Control
Abstract: Map maintenance is essential for the long-term navigation of robots in changing environments. There are transient objects such as parked cars and standing signs in urban environments. As a result, some structures on the previous map may not be observed when revisiting the environment. Determining whether these structures have truly disappeared or are simply not visible at the time of mapping is important for maintaining the map during updates. In this paper, we propose a method to remove points that are no longer existing but unintentionally remain on the updated map. Our approach analyzes the boundaries of disappeared clusters to find nearby points that likely belong to the same object and should be removed together. The proposed method improves the reliability of the map and supports more reliable lifelong mapping and localization in changing environments.

Keywords: Control Theory and Applications, Artificial Intelligence Systems, Robot Mechanism and Control
Abstract: Robotic manipulation is crucial for executing diverse object interaction tasks, and in particular, it should be capable of performing pick-and-place operations to desired locations under payload variations. Model-based approaches require accurate system dynamics, while learning-based approaches suffer from low sampling efficiency due to the need for extensive exploration in high-dimensional action spaces. To address these limitations, we propose a hybrid control framework that combines inverse kinematics (IK) with a reinforcement learning (RL)-based policy that compensates for residual errors. This design enables robust control even under unknown and varying payloads, allowing the robot to reach suboptimal positions when the target is kinematically unreachable. Moreover, leveraging IK guidance improves sampling efficiency and significantly accelerates the learning process compared with pure RL-based approaches.

Keywords: Human-Robot Interaction, Robotic Applications, Information and Networking
Abstract: Human-robot interaction has evolved into a multimodal interaction that actively incorporates non-verbal behaviors. This progression is particularly significant for endowing humanoid robots with natural and socially acceptable behaviors. The aim of this study is to enable more natural and effective interactions between humans and humanoid robots by interpreting human speech, adaptively controlling various context-appropriate gestures, and synchronizing multimodal expressions in real-time. To achieve this, we propose an empathy-enhanced multimodal interaction control architecture that integrates large language models (LLMs) to generate semantically appropriate verbal responses, emotional expressions, and synchronized semantic gestures tailored to the conversational context. The proposed system was implemented on a humanoid robot platform and its operational feasibility was experimentally validated. We expect that this research will contribute to enhancing the social capabilities of humanoid robots and promote more intuitive and human-like interaction in future human-robot communication systems.

Keywords: Control Theory and Applications
Abstract: This paper proposes a new dynamic-integral event-triggered control mechanism to solve the consensus problem of general linear multi-agent systems. Based on the dynamic-integral event-triggered control protocol, the leaderless consensus problem is investigated. Under the novel strategy, it is shown that leaderless consensus is guaranteed with undirected communication topology, and Zeno behavior can be excluded by finding a positive lower bound of the inter-event intervals. Furthermore, we extend the dynamic-integral event-triggered control protocol to discuss the applicability of formation control. The effectiveness of the obtained results is validated by numerical simulations.

Keywords: Control Theory and Applications, Industrial Applications of Control, Control Devices and Instruments
Abstract: In hydraulically actuated mobile cranes, actuator-induced force variations directly affect internal stress levels. The motion under dynamic loading can lead to progressive stress build-up over time, increasing the risk of material degradation and structural failure. Conventional control approaches typically omit structural stress feedback, focusing on performance objectives and overlooking internal force effects. This study introduces a stress-responsive valve control framework for the hydraulic crane systems. The proposed method uses simulation data to create a stress profile that follows the same pattern as the applied loads. A control coefficient K is computed in real time based on normalized stress and stress rate, using a polynomial formulation. This coefficient adjusts the actuator valve signals to reduce control intensity when stress levels increase or fluctuate. The implementation is carried out in a rigid multibody simulation environment with integrated sensor emulation and actuator logic. At each simulation step, valve signals are updated according to the stress levels therefore adjusting the dynamics of crane, adaptively. Comparative analysis between the stress-aware control system and an uncontrolled baseline is performed under identical actuation conditions. Results indicate a reduction in stress peaks while maintaining smooth and desired movements of crane. The analysis showed a 62.71% reduction in the effective damping ratio. These results demonstrate that the proposed method reduces actuator aggressiveness in high-stress regions and limits structural loading without compromising actuation objectives.

Keywords: Control Devices and Instruments, Robotic Applications
Abstract: This paper proposes a novel magnetic actuation system (MAS) utilizing Halbach arrays to expand the traditionally two-dimensional workspace into three dimensions. While conventional electromagnetic systems face limitations due to their bulky size and power consumption, typical Halbach-based permanent magnet systems offer a restricted two-dimensional workspace confined within their array. To overcome these constraints, we introduce an optimized dual array design based on dipole and quadrupole configurations. By optimizing the magnet orientations, the proposed system can achieve uniform magnetic fields and gradients in an external 3D workspace. Simulation results demonstrate that the dipole array effectively generates strong, uniform magnetic fields suitable for applying torque, while the quadrupole array forms a consistent gradient ideal for translational control. These advancements enhance the controllability of magnetic medical robots, such as capsule endoscopes and micro-nano robots, in clinical settings. Future work aims to integrate both arrays into a unified system and verify its efficacy in real-world applications.

Keywords: Sensors and Signal Processing, Autonomous Vehicle Systems
Abstract: Generating realistic night-time traffic scenes requires precise control over global color tone and spatial structure, as such scenes involve complex layouts under challenging lighting. We present a conditional image generation framework that combines color distribution guidance, semantic layout control, and sampling refinement for night-time image synthesis. Specifically, we align intermediate denoising outputs with the color distribution of a reference image using Sliced Wasserstein distance. To ensure spatial controllability, we incorporate a segmentation map via ControlNet. Additionally, we employ Adaptive Projective Guidance (APG) to improve the stability and sharpness of the generated results. Our approach produces visually plausible and semantically accurate night-time traffic scenes.

Keywords: Artificial Intelligence Systems, Robot Mechanism and Control, Human-Robot Interaction
Abstract: Physical AI systems such as robots, drones, and autonomous vehicles are now regulated as high‑risk AI systems by the European Union Artificial Intelligence Act (EU AI Act). Articles 14, 15, and 16 require such systems to provide human oversight, accuracy and cybersecurity, and recovery‑ready resilience [1][2][3]. This paper proposes the Enhanced Guardian Integrated System (EGIS): a dual‑brain controller whose hardware‑isolated Guardian monitors the Host, enforces safety, accepts human overrides, and restores corrupted storage. We compare EGIS with recent fault-tolerant and secure control approaches in literature, including redundant hardware architectures, adaptive fault-tolerant controllers, and safety supervisory systems. Unlike conventional fault-tolerant control which often assumes the primary controller is trustworthy, EGIS provides an independent safety layer compliant with emerging AI safety regulations. Case study scenario analyses demonstrate EGIS halting a hacked robot and rolling back its file‑system state, thereby meeting the EU AI Act’s requirements for resilience, recovery and explicit human oversight.

Keywords: Artificial Intelligence Systems, Control Devices and Instruments, Robotic Applications
Abstract: Accurate external force estimation is essential in robotic systems, and the Kalman Filter is a widely used method for this purpose. However, the filter’s performance heavily depends on properly tuned noise covariances, which are often set manually through trial and error. In this work, a deep reinforcement learning approach based on Proximal Policy Optimization (PPO) is used to automatically optimize the Kalman Filter’s parameters. The learning agent interacts with a custom environment, adjusting the noise covariances to minimize the estimation error of joint velocity. Compared to conventional manual tuning, the PPO-optimized filter yields smoother and more accurate estimations. The results demonstrate the potential of reinforcement learning for adaptive Kalman Filter tuning in dynamic systems.

Keywords: Navigation, Guidance and Control, Sensors and Signal Processing, Artificial Intelligence Systems
Abstract: This paper compares traditional wavelet denoising methods with deep learning-based approaches across multiple signal types and noise conditions. We evaluated five methods on sine waves, chirp signals, and block signals with signal-to-noise ratio levels from 10 − 20 dB. The multi-stage wavelet denoising convolutional neural network achieved the highest performance, demonstrating 15 – 20 % improvements over traditional methods with peak signal-to-noise ratio values reaching 40.43 dB. Deep learning methods showed superior performance, particularly for complex signals, establishing their effectiveness for modern signal denoising applications.

Keywords: Multimedia Systems, Sensors and Signal Processing, Artificial Intelligence Systems
Abstract: Non-verbal information should be utilized in the daily communication process of the hearing-impaired. For effective interaction, it is necessary to recognize the other person's mouth shape visually. Thus, it is essential to indicate the exact direction of speech generation to the hearing-impaired. To address this issue, this study proposes a micarray-based system that provides directional information based on accurately extracting human voices in a noisy environment. To detect human voice direction even in noisy environments, RNNoise-based noise suppression and MSFF (Multi-Stage Frequency Filtering) are combined to effectively remove non-voice signals. In addition, this system compared the TDOA (time difference of arrival) and RMS-based energy vector methods as direction estimation algorithms. The results of the experiment showed that the RMS-based method achieved better accuracy over short distances, while the TDOA method performed better over long distances. Therefore, we confirmed that the two methods can be used in conjunction with each other.

Keywords: Autonomous Vehicle Systems, Robotic Applications
Abstract: PX4, an open-source autopilot for UAVs, includes automated integration testing as part of its development workflow. However, this testing is limited to MAVLink-based communication and lacks support for ROS2, which is increasingly used on companion computers for handling high-level mission tasks. This gap hinders seamless testing and validation in modern ROS2-based unmanned aerial vehicle (UAV) systems. In this paper, we propose, Aurora, an Automated ROS2 Integration and Reliability Assessment testbed for UAVs using the PX4 autopilot as the flight controller. The testbed uses ROS2 action server-client architecture for running and managing individual tests and a test manager to run multiple tests automatically. This framework uses JSON based user friendly template for effortlessly defining complex test cases and also provides a user interface to run and create multiple tests with ease. The Aurora testbed is tested in software-in-the-loop (SITL) as well as hardware-in-the-loop (HITL) settings. The code is available at https://github.com/tiiuae/Aurora.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Robot Vision
Abstract: We present a practical extension to diffusion-based planning by implementing constraint-aware classifier guidance as a constrained inverse problem. Our method injects differentiable gradients of energy functions—representing safety, comfort, and speed constraints—into the denoising process of a pretrained diffusion model, enabling test-time adaptation without retraining. This work operationalizes the classifier guidance component that was proposed but not implemented in prior diffusion planners, offering interpretable and flexible control during inference. Experiments on the nuPlan dataset validate the effectiveness of our approach in generating constraint-compliant and diverse trajectories, while revisiting the role of guidance in diffusion-based planning models.

Keywords: Human-Robot Interaction, Robot Mechanism and Control, Exoskeleton Robot
Abstract: In order to ensure the degrees of freedom (DOFs) of each finger while simultaneously achieving high DOFs for all fingers, the number of actuators increases, which leads to disadvantages in weight and maximum output compared to underactuated methods. However, by using a torque distribution mechanism, the power of a single actuator can be selectively transmitted, allowing the actuator’s output to be concentrated on the fingers requiring movement. This enables high operational efficiency even with a small number of actuators. In this paper, we implemented a mechanism that divides the input torque into transmission, braking, and floating states using an electroadhesive clutch, achieving a total weight of approximately 200 g. It is expected that through further weight reduction, the system weight can be decreased to below 100 g. Therefore, this mechanism is anticipated to serve as a novel alternative for high-DOF wearable gloves and prosthetic hands.

Keywords: Rehabilitation Robot, Human-Robot Interaction, Biomedical Instruments and Systems
Abstract: Functional electrical stimulation (FES) cycling provides therapeutic exercise for individuals with lower limb paralysis, but performance may be limited by friction at the seat–pelvis interface. This study tested whether a passive rotational seat, incorporating a vertical-axis bearing to allow low-resistance pelvic rotation, improves outcomes without altering FES settings. A participant with spinal cord injury (ASIA A, T10) completed cycling trials with both fixed and rotational seats under identical stimulation. The rotational seat increased cycling distance by 37.8% and stabilized crank angular velocity, particularly around dead zones (0° and 250°). These results suggest that reducing seat friction enhances pedaling consistency. The proposed mechanical solution requires no sensors, actuators, or stimulation changes, offering a simple means to improve FES-assisted cycling performance.

Keywords: Control Theory and Applications, Industrial Applications of Control, Control Devices and Instruments
Abstract: For high-performance control of Permanent Magnet Synchronous Motors (PMSMs), it is essential to develop control techniques that are robust against disturbances and parameter variations. In particular, PMSMs are susceptible to instability due to the reduction of permanent magnet flux with temperature rise during operation, which leads to torque degradation. In this paper, we propose a Linear Parameter-Varying (LPV) H-infinity controller that incorporates thermal derating compensation. The proposed controller adapts to the varying parameters according to the rotor position and is designed to guarantee robust performance against various disturbances, such as parameter perturbations and load torque changes. Furthermore, the system enables current state observation using only a position sensor, minimizing the sensor configuration while achieving high control precision. The effectiveness of the proposed approach is validated through both simulations and experiments.

Keywords: Robotic Applications, Control Devices and Instruments, Human-Robot Interaction
Abstract: We propose a soft electromagnetic sliding actuator that provides various planar motions to construct highly compliant actuation systems. The actuator is composed of a fully soft actuation base (stator) for generating electromagnetic and magnetic forces and a rigid neodymium magnet (slider) that slides on the actuation base. A parallel liquid-metal coil array in the stator is designed based on theoretical modeling and an optimization process to maximize the electromagnetic field density. The stretchable magnetic components in the stator allow the slider to retain its position stably without additional constraints. By incorporating an untethered structure in which the slider is mechanically decoupled from the stator, the actuator can be operated with reduced power consumption, attributed to the absence of a restoring force. The trajectory of the slider can be programmed by selectively applying the input current to the liquid-meal coil array, and the location of the slider can be estimated by measuring the change in inductance of each coil. Moreover, the proposed actuator demonstrates the capability of operating on curved surfaces through its physical compliance as well as on inclined surfaces thanks to the holding force generated by the magnetic components of the stator. Taking advantage of the unique characteristics of our actuator, robotic applications, including shape morphing systems and sensor-actuator integrated systems, are demonstrated.

Keywords: Sensors and Signal Processing, Human-Robot Interaction, Robotic Applications
Abstract: We propose a compact wearable glove capable of estimating both the finger bone lengths and the joint angles of the wearer with a simple stretch-based sensing mechanism. The soft sensing glove is designed to easily stretch and to be one-size-fits-all, both measuring the size of the hand and estimating the finger joint motions of the thumb, index, and middle fingers. The system was calibrated and evaluated using comprehensive hand motion data that reflect the extensive range of natural human hand motions and various anatomical structures. The data were collected with a custom motion-capture setup and transformed into the joint angles through our post-processing method. The glove system is capable of reconstructing arbitrary and even unconventional hand poses with accuracy and robustness, confirmed by evaluations on the estimation of bone lengths (mean error: 2.1mm), joint angles (mean error: 4.16°), and fingertip positions (mean 3D error: 4.02mm), and on overall hand pose reconstructions in various applications. The proposed glove allows us to take advantage of the dexterity of the human hand with potential applications, including but not limited to teleoperation of anthropomorphic robot hands or surgical robots, virtual and augmented reality, and collection of human motion data.