Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems
Abstract: This paper proposed a methodology for determining the preferred gimbal angle set for a nonredundant Control Moment Gyroscopes (CMG) cluster and discussed its performance. To determine the preferred gimbal angle set, the gimbal angle trajectory is first defined. The trajectory uses the preferred gimbal angle as an intermediate point to generate the gimbal rate command. The optimal solution of preferred gimbal angle obtained through the Sequential Quadratic Programming (SQP) optimization method and utilize as a dataset label. Instead of using the preferred gimbal angle set, many datasets are generated in a grid format to train a Multi-Layer Perceptron (MLP) in a supervised framework. The trained network estimates the optimal gimbal angle set based on a given target attitude. Compared to computationally intensive optimization methods, the MLP model can quickly estimate an optimal gimbal angle set that achieves the desired attitude. Therefore, in this study, the performance of the proposed method was evaluated by training the MLP with datasets generated using grid intervals and assessing the results based on the estimated preferred gimbal angles.

Keywords: Biomedical Instruments and Systems, Artificial Intelligence Systems, Information and Networking
Abstract: Sustaining engagement in cognitive monitoring for aging populations requires assessment tools that continuously adapt to individual performance. We proposed a dynamic difficulty adjustment (DDA) framework for a tablet-based memory-matching game, leveraging synthetic data generation, swarm intelligence, and reinforcement learning. Addressing typical data limitations, we generated behaviorally consistent synthetic gameplay trajectories via parametric sampling, augmenting an original dataset from healthy cognitive (HC) and possible Mild Cognitive Impairment (pMCI) participants. This augmented data allowed us to robustly train a Q-learning agent. Its reward function integrated performance metrics, error rates, and difficulty changes, with parameter weights optimized using the Artificial Bee Colony (ABC) algorithm. The agent's policies, refined over multiple training episodes to Q-table convergence, closely aligned with expert-defined suitability criteria for difficulty adjustments. Upon evaluation using a test dataset, the system yielded 98.55% positive rewards in the HC group and 94.29% in the pMCI group. This high positive feedback rate across diverse participant groups underscores the DDA framework's success in promoting player success and fostering an encouraging environment, vital for long-term engagement in cognitive monitoring.

Keywords: Navigation, Guidance and Control
Abstract: Track-Before-Detect (TBD) estimates target state directly from raw sensor data and thus remains effective at low SNR. A particle-filter implementation lifts the linear-Gaussian restriction, handling nonlinear motion and non-Gaussian clutter. Bernoulli-TBD goes further by attaching a Bernoulli existence variable, so target birth, death, missed detections, and false alarms are managed in a single recursion. Because particle filters suffer from weight degeneracy, resampling is essential, and its variant can greatly affect accuracy and cost. We therefore benchmark three well-known resampling schemes, each matched to the condition for which it is designed. Experiments confirm that choosing the resampling rule that fits the operating scenario yields the best tracking performance.

Keywords: Robot Mechanism and Control
Abstract: This study proposed and experimentally validated a novel tension control mechanism to suppress slack in open-loop wire rope systems. The proposed mechanism combines a rope-feeding unit, which transports the rope at a faster speed than the drum, with a contact adjustment unit that modulates power transmission, thereby increasing internal tension during rope release. For performance evaluation, slack suppression, positional accuracy, and stability under repetitive operation were established as key metrics, and experiments were conducted under both unidirectional and reciprocating motion scenarios. The results demonstrated that the proposed mechanism effectively eliminated slack and functional failures, reducing positional error from 98.395% to 0.67%. Furthermore, during repeated operations, cumulative error remained below 5%, confirming reliable performance. Previous studies relied on external pre-tension or lacked the ability to detect slack in real time, presenting inherent limitations. By achieving stable rope release without external pre-tension through active control based on internal tension sensing, this study overcomes those limitations and demonstrates clear differentiation. Therefore, the proposed mechanism is expected to contribute to addressing issues of tension maintenance and slack suppression not only in open-loop wire rope systems but also across the broader field of tendon-driven robotics.

Keywords: Robot Mechanism and Control, Robotic Applications, Control Theory and Applications
Abstract: This paper proposes a six-degree-of-freedom (6-DOF) motion platform control system to effectively mitigate shock during the landing of an Unmanned Aerial Vehicle (UAV). To prevent airframe overturning and structural damage caused by disturbances, a sensorless impedance control scheme based on flight data is developed. Since UAV motion is continuous and inherently difficult to predict, designing a robust controller presents additional challenges. To address this, the proposed controller combines a sliding mode controller (SMC), which accounts for system dynamics, with the flight data-based impedance controller, and it operates in coordination with a platform composed of six linear actuators. System stability is verified through Lyapunov-based analysis, and the effectiveness of the method is validated through experimental testing. Results confirm that the proposed approach significantly reduces landing shock, indicating its potential to improve the operational safety of UAV.

Keywords: Artificial Intelligence Systems, Control Theory and Applications
Abstract: Data-driven Koopman operator approximation has gained interest recently for its ability to embed nonlinear systems into a lifted linear state space using only measurements. When control inputs are included, however, the lifted dynamics render a bilinear form, which poses challenges for controller synthesis, such as Model Predictive Control (MPC). This paper proposes an on-policy actor-critic Deep Reinforcement Learning (DRL) framework that simultaneously learns the Koopman bilinear dynamics and an MPC neural cost map. Instead of directly generating control actions, the actor network takes the Koopman-lifted states and produces MPC weight matrices for each prediction step. These state-dependent weight matrices serve as high-level guidance for the control objective, allowing the low-level MPC to run under very short prediction horizon while maintaining stability and enforcing safety constraints. Simulations carried out with the OpenAI Gym library demonstrate that, without requiring explicit knowledge of the dynamics, the proposed Actor-Critic Koopman MPC (ACKMPC) achieves control accuracy and disturbance robustness on par with a model-based ACMPC, and outperforms a pure DRL-learned policy using baseline Proximal Policy Optimization (PPO). It also exceeds standard Koopman MPC (KMPC) in both robustness and computational efficiency.

Keywords: Artificial Intelligence Systems
Abstract: With the rapid expansion of electric vehicle (EV) adoption, improving battery charging efficiency has become a critical challenge. To maximize charging performance, EVs activate a battery thermal management system (BTMS) on the way to charging stations, aiming to ensure that the battery reaches its optimal temperature for efficient charging upon arrival. However, current BTMS activation timing is typically determined using a calibration-mapping approach, which does not account for the dynamic thermal behavior of the battery, leading to limited accuracy. In this study, we propose a Long Short-Term Memory (LSTM)-based model to predict battery temperature behavior during the pre-heating phase prior to charging. To improve prediction accuracy, a Physics-Informed Neural Network (PINN) is employed to integrate governing physical laws into the learning process. In addition, the model excludes initial data with unstable temperature behavior and utilizes variable window sizes to better capture temporal patterns. Real-time input variables received via CAN communication were fed into the model, demonstrating the feasibility of on-device AI in an automotive environment. Experimental results demonstrate that the proposed model outperforms conventional calibration-based methods in accurately estimating the time required to reach the optimal charging temperature. This research offers a promising approach to improving EV charging efficiency and lays the groundwork for intelligent thermal control in next-generation battery management systems.

Keywords: Artificial Intelligence Systems, Navigation, Guidance and Control, Robotic Applications
Abstract: A key challenge in multi-agent systems is developing distributed perimeter defense strategies under sensing and communication constraints. In such scenarios, defenders must coordinate to intercept intruders without relying on centralized control or constant inter-agent communication. We propose a distributed strategy in which each defender adjusts its angular velocity based on local sensing and a probability-guided potential field generated by a central sensing unit. The unit estimates the angular distribution of intruders using a von Mises kernel and broadcasts the resulting field to all defenders as global threat cues. The control framework integrates local interception, collision avoidance, and field guidance. Simulation results show that the proposed method enables conflict-free, adaptive defense and efficient boundary coverage. Further analysis indicates that the potential field is more influential in small-scale defense, and that interception efficiency is maximized when the estimation bias is moderate-avoiding both over-concentration and over-dispersion.

Keywords: Artificial Intelligence Systems, Robotic Applications, Control Theory and Applications
Abstract: Recent advances in imitation learning have enabled robots to perform increasingly complex manipulation tasks in unstructured environments. However, many learned policies rely on discrete action chunking, which often introduces discontinuities at chunk boundaries. These discontinuities degrade motion quality and are particularly problematic in dynamic tasks such as throwing or lifting heavy objects, where smooth trajectories are critical for momentum transfer and system stability. In this work, we present a lightweight post-processing framework for smoothing chunked action sequences. Our method combines three key components: (1) inference-aware chunk scheduling to proactively generate overlapping chunks and avoid pauses from inference delays; (2) linear blending in the overlap region to reduce abrupt transitions; and (3) jerk-minimizing trajectory optimization constrained within a bounded perturbation space. The proposed method was validated on a position-controlled robotic arm performing dynamic manipulation tasks. Experimental results demonstrate that our approach significantly reduces vibration and motion jitter, leading to smoother execution and improved mechanical robustness.

Keywords: Artificial Intelligence Systems, Robot Vision, Human-Robot Interaction
Abstract: This study proposes a target person tracking and following method using image recognition based on a machine learning algorithm, FOMO(Faster objects, More objects), which is suitable for tiny microcomputers. For the purpose we designed and producted an autonomous mobile robot equipped with a small camera including microcomputer-based processing system, thermal array sensors, and a 2D-LiDAR. The image processing system inside the camera module incorporates a lightweight object detection model of a target person's clothing, which is generated by machine learning algorithm FOMO. This state-of-the-art machine learning architecture works well for small object detection in sequential images, reducing computational burden and also contributing to small-size robot design. This paper newly derives an estimation model for calibration between image-pixel difference and walking distance towards depth direction, which is available for explicit control of the traveling distance between a target person to be followed and the mobile robot. This reasonable tracking and following method can be created from the unique mechanical design of the robot, which includes low-angle positioning of the camera and its wide-angle lens. Finally, we show successful following results in experiments, in which unexpected partial occlusions by another non-target person occur frequently.

Keywords: Artificial Intelligence Systems, Robotic Applications
Abstract: Autonomous docking in mobile robots requires precise control under dynamic conditions like sensor noise and surface variation. While PID controllers are simple, fixed gains lack adaptability. Recent LLM-based dynamic PID tuning improves flexibility but suffers from high latency. We propose LoGenE (LoRA-based Genetic Evolution), a gradient-free neuroevolution framework that optimizes lightweight LoRA adapters using control logs and reward-guided evolution. Trained on docking data, LoGenE achieves high success rates and reduced docking time in ROS + Gazebo, with sub-second inference latency suitable for real-time robotic systems.

Keywords: Artificial Intelligence Systems, Robot Vision, Industrial Applications of Control
Abstract: Robust fall detection is critical in safety-sensitive contexts such as elderly care. Recent fall detection methods leverage 2D human pose estimation, which encodes posture as compact skeletal keypoints. However, these systems are vulnerable to performance degradation when body parts are occluded or keypoints are missing. We propose a fall detection framework based on single-frame 2D pose estimation and a Skeleton-aware Graph Attention Network (StoneGAT). StoneGAT enhances GAT by incorporating edge features such as bone lengths, joint angles, and confidence metrics. We also introduce PointOut, a data augmentation strategy that masks keypoints during training to simulate occlusion and improve robustness. Experiments on a combined dataset of AI-hub and in-house samples show that StoneGAT with PointOut outperforms baseline models under severe occlusion.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Artificial Intelligence Systems
Abstract: To reduce the burden of managing electric fences against wild animals, we propose an autonomous patrol robot that performs wild animal detection and electric fence anomaly detection during autonomous driving. The autonomous driving is based on the Real-Time Kinematic Global Navigation Satellite System (RTK-GNSS), while object detection is performed by Convolutional Neural Network (CNN) models and the probabilistic Hough transform. To demonstrate the effectiveness of the proposed robot, we evaluated the detection models and conducted an operation test to patrol an actual electric fence with the evaluation of patrol time, processing time, power consumption, and energy consumption.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Robotic Applications
Abstract: Stable aerial manipulation during tasks such as object catching, perching, or contact with rigid surfaces necessarily requires compliant behavior, often achieved via impedance control. Successful manipulation depends on how the impedance control can tackle the unavoidable coupling forces between the aerial vehicle and the manipulator. However, the existing impedance controllers for aerial manipulator either ignore these coupling forces or require their precise knowledge, which is difficult, if at all possible, to obtain. We introduce an impedance controller for aerial manipulator which, via suitably designed adaptive laws, overcomes the need for a priori knowledge of the system dynamics and of the coupling forces. The closed-loop system stability is proved analytically and experimental results with a payload-catching scenario demonstrate significant improvements in overall stability and tracking over the state-of-the-art impedance controllers.

Keywords: Autonomous Vehicle Systems, Sensors and Signal Processing, Robotic Applications
Abstract: In this study, a lidar sensor is modelled within a physics engine environment to overcome the complexities and uncertainties of off-road autonomous driving and to replicate real-world conditions. The modelling methodology and parametric model parameters are presented. Under predefined test conditions, data acquired from an actual lidar sensor was analyzed and compared with the output of the lidar sensor model to evaluate its consistency and reliability. Additionally, the generated point cloud is visualized over the terrain features, demonstrating the sensor’s capability to represent environmental structures. The synthetic data produced by the modelled lidar sensor is intended to support the development of autonomous vehicle algorithms in off-road conditions

Keywords: Autonomous Vehicle Systems, Control Theory and Applications
Abstract: In this paper, we propose a lateral control of the autonomous vehicle with lateral offset constraint. To satisfy the lateral offset constraint, we developed a novel lateral model. The proposed lateral model of this paper consists of the lateral offset(LO) model and look-ahead lateral offset(LALO) model. The desired look-ahead lateral offset is designed using the LO model and disturbance observer-based time-varying asymmetric barrier lyapunov function control. For tracking desired look-ahead lateral offset, the extended state observer based dynamic surface control is designed using the LALO model. The performance of the proposed method is evaluated using cosimulation between CarMaker and MATLAB/Simulink.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Robotic Applications
Abstract: This study aims to compare the performance of two local path planning algorithms—Timed Elastic Band (TEB) and Dynamic Window Approach (DWA)—for a three-wheeled omnidirectional robot, evaluating their performances in parking maneuvers, rectangular trajectories, circular trajectories and static obstacle avoidance scenarios. Experimental results show that performance of the TEB consistently outperforms ones of the DWA, especially in terms of time efficiency and trajectory accuracy. In parking maneuvers, TEB completes tasks swiftly and precisely, maintaining stable velocity and posture control, whereas DWA exhibits more pronounced heading oscillations during turns, which reduces accuracy. In static obstacle avoidance tests, TEB demonstrates superior path planning capabilities, navigating around obstacles smoothly and maintaining optimal distances, while DWA tends to generate more conservative and less efficient paths. Wheel velocity analysis further indicates that TEB achieves more stable wheel velocity distribution, avoiding the velocity instabilities seen in DWA. In summary, TEB is undoubtedly the more suitable choice for applications requiring rapid, high-precision navigation and robust obstacle avoidance, while DWA remains a feasible option for simpler implementation needs.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Sensors and Signal Processing
Abstract: Automation in the construction industry has the potential to improve productivity and speed of production. This entails the transport of modules and materials in different parts of the building process. This work proposes a concept for the use of a large-scale autonomous mobile robot (AMR) in construction prefabrication that requires only minor changes to existing infrastructure. The AMR is used to transport loads indoors to manufacturing platforms, as well as storage facilities. In addition, the concept includes outdoor navigation, so that loads can be delivered from and to a construction crane. To make this possible, a general navigation mode is defined, which focuses on favorable properties on trips to a target location using various sensor configurations. Furthermore, to achieve higher positioning and localization accuracy at target destinations, a second localization mode is defined. The validation is done in simulation, as well as on the timber prefabrication system of the cluster for Integrative Computational Design and Construction for Architecture.

Keywords: Artificial Intelligence Systems, Robotic Applications, Industrial Applications of Control
Abstract: Grasping is one of the fundamental skills for multi-fingered robotic grippers in diverse manipulation tasks. We propose a method for Blind grasping (i.e., without use any vision sensor) based on reinforcement learning using only proprioceptive information (i.e., joint angles) and a low-cost 1-axis force sensor. Instead of directly training the grasp policy from limited sensor data, we first pre-train a teacher policy in simulation with full observation, including object pose and velocity, 3-axis contact forces, joint states, and fingertips pose. After successful pre-training, this teacher policy generates trajectories for effective grasps, which are then used to train a student policy through behavioral cloning. The student policy learns to replicate the actions of the fully observed policy using only the limited sensors available in the real gripper. Our approach shows that a blind grasping policy can be developed without relying on complex vision system or expensive tactile sensors, using only 1-axis force sensors and proprioceptive information.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Robot Vision
Abstract: Railway is a complex system, embedding hundreds of thousands of parts and objects. They constitute tracks, trains, signaling, etc. Their huge diversity makes it impossible for human agents to know every one of them, even less so their technical specificity or utility. Computer Vision (CV) tools offer elevated performance solutions that could help human agents identify parts and adapt their actions regarding reuse and recycling. Those CV models require high quality annotated data. They serve as the essential ground truth that guides the learning process for robust performances. However, railway datasets are scarce, acquisition is laborious and dangerous, and annotation is time-consuming and expensive. Therefore, providing high quality labeled railway data is still a challenge. To address this issue, our paper introduces a complete framework for railway object detection and focuses on the training and validation dataset creation. The proposed approach integrates photorealistic viewpoints’ generation of physical objects, from a single 3D representation. This pipeline significantly reduces the time and resources needed to produce high-quality training data. Then, the database is used to train a classifier to recognize those parts in various situations. Experimental results show that models trained with those generated images achieve competitive performance compared with classical real-life datasets. Both offline and real-time object recognition achieve an average of 98,3% precision. This framework offers a scalable, cost-effective solution for rapid CV model deployment. It can be extended to any industrial setting but we illustrate our solution on railway signaling object identification.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Robot Vision
Abstract: Robust autonomous navigation in leisure environments like country clubs demands precise yet efficient visual perception. This paper introduces Gram Matrix Guided Relational Alignment, a novel knowledge distillation framework that balances segmentation accuracy with computational efficiency for on-device deployment. This method captures high-level structural knowledge by computing Gram matrices from channel attention maps, modeling global context and inter-channel correlations. This information is transferred from a complex teacher to a lightweight student network using Centered Kernel Alignment (CKA) for robust alignment. On a real-world country club dataset, our method significantly outperforms existing distillation techniques. Notably, the resulting lightweight student model achieves superior performance to the original teacher network, greatly enhancing the reliability of autonomous collision avoidance systems.

Keywords: Artificial Intelligence Systems, Control Theory and Applications, Industrial Applications of Control
Abstract: In this study, we introduce two novel two-stage detection approaches to enhance the performance of block diagram recognition using large language models (LLMs), by integrating visual inference and text generation capabilities. The first approach, termed Two-Stage by Double LLM (TSD), employs one LLM to generate LaTeX code from a block diagram image and a second LLM to interpret the code. The second approach, termed Two-Stage by Single LLM (TSS), similarly generates LaTeX code from the image but uses the same LLM for both generation and interpretation. We evaluate these methods on the dataset of block diagram images. Experimental results confirm that while ChatGPT o4-mini attains the accuracy of 35%, the TSD and TSS methods achieve significantly higher accuracies of 87% and 96%, respectively.

Keywords: Artificial Intelligence Systems
Abstract: Irrigation ponds play a vital role in rice cultivation in Japan. However, in recent years, the aging population of agricultural workers and the shortage of successors have made it increasingly difficult to manage irrigation ponds in Japan. To address this issue and revolutionize biological surveys through new technologies, this study aims to develop a system that detects and recognizes rare freshwater fish using images captured by underwater drones and foundation models in deep learning. We evaluated the performance of YOLO-World and Grounded SAM 2 as foundation models for object detection and segmentation. While YOLO-World required annotated data for training, Grounded SAM 2 showed potential for annotation-free processing. However, the latter exhibited a high rate of false detections, indicating that improving accuracy remains a key challenge for future development.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control
Abstract: In recent years, integrating renewable energy sources has increased the complexity and uncertainty of power systems, resulting in frequent line overload issues and greater challenges for automated control and management. To address this problem, this paper proposes the OC-LSTM algorithm, a deep reinforcement learning (DRL) approach that combines the option-critic framework with long short-term memory (LSTM) neural networks to achieve efficient power system management. The algorithm leverages LSTM networks to extract temporal features from the power system. It employs the option-critic framework to learn topology adjustment policies, effectively mitigating overload risks and maintaining system stability. Experimental results demonstrate that the OC-LSTM algorithm outperforms standard DRL algorithms during training. Notably, OC-LSTM achieves continuous stable operation for 60 hours without manual intervention in test systems such as IEEE 5-Bus, IEEE 14-Bus, and L2RPN WCCI 2020.

Keywords: Artificial Intelligence Systems, Control Theory and Applications, Industrial Applications of Control
Abstract: With the increasing complexity of modern power systems due to the access of large-scale renewable energy sources, minimizing operational costs while achieving stable grid operation has become a core challenge in power scheduling and optimization. Energy dispatch and topology control are key measures to improve power system stability and flexibility. However, the outputs of their traditional control policies rely on predefined rules or mathematical optimization models, which are prone to computational bottlenecks and response lags in high-dimensional dynamic environments, making it difficult to meet the demands of smart grids. In recent years, deep reinforcement learning (DRL) has gradually become a cutting-edge technology for power system scheduling and control by virtue of its powerful adaptive learning and decision optimization capabilities. According to the existing research, DRL can improve the flexibility and anti-interference ability of the power grid by learning the optimal policies through autonomous interaction, surpassing the real-time decision-making ability of traditional optimization methods in high-dimensional state space. In this paper, we systematically review the applications of DRL in energy dispatch and topology control, focusing on its optimization policies, technological breakthroughs and applicability, and analyze the current challenges and future research directions.

Keywords: Control Theory and Applications, Process Control Systems, Industrial Applications of Control
Abstract: In this paper, event-triggered adaptive critic based optimal tracking control is proposed. We approximate the unknown terms in the optimal control formulation using a neural-network (NN) to solve the Hamilton–Jacobi–Bellman equation. We explicitly enhance closed-loop robustness by adding a Lyapunov-based stabilizing term to the adaptive critic weight update law. Finally, adopting an event-triggered mechanism, we eliminate unnecessary updates and significantly reduce computational load. RL provides a useful tool because it is difficult to calculate the unknown term when RL is not in used. Therefore, we apply a NN–based adaptive critic to estimate the unknown term in the cost function.

Keywords: Control Theory and Applications, Artificial Intelligence Systems, Industrial Applications of Control
Abstract: This paper proposes a reinforcement learning-based adaptive critic controller to find the optimal solution for the nonlinear multi-agent systems. To simplify the process of proving the uniform ultimate boundedness (UUB) of the closed-loop system, an additional stabilizing term is complemented in the adaptive critic weight update law. An asymmetric barrier Lyapunov function (ABLF) also utilized to provide an optimal solution for nonlinear multi-agent systems under state constraints. The simulation results are provided to validate the proposed method.

Keywords: Industrial Applications of Control, Control Devices and Instruments
Abstract: It is crucial to accurately estimate the state of lithium-ion batteries to improve the energy efficiency of applications. such as electric vehicles and energy storage systems. The performance of the Extended Kalman Filter (EKF), used to estimate the state of charge (SOC) in batteries, is determined by its parameters. However, the parameters of the EKF, selected on the basis of cell characteristics, can result in significant estimation errors, potentially reducing the energy efficiency of the battery. Therefore, this paper proposes an EKF framework capable of achieving an estimation accuracy applicable to real-world applications at the battery pack level. The selected EKF parameters at the battery pack level are validated using a real-time simulator, ensuring the reliability necessary for operation within the battery management system (BMS). The proposed algorithm demonstrates stable performance with an average estimation error of less than 2% with real-world applicability.

Keywords: Robotic Applications, Human-Robot Interaction
Abstract: Since the elderly and the disabled are limited in their activities and cannot get baggage at a distance, robots have been proposed to assist them in their daily lives. Previous research has focused on collaborative work between robots and humans, and has investigated about digital twin operations and remote manipulation of complex tasks. In contrast, a method to support picking by manipulating objects using a Mixed Reality (MR) has attracted attention. Although this method is intuitive and easy to instruct the robot, it is difficult for the elderly because it requires precise manipulation in the MR space. In this study, a picking system using multimodal information of the MR operations by the gaze and voice activation is proposed for a mobile robot with a hand. The MR operation by the eye gaze is intuitive and allows a fast selection and an operation, while the speech operation is superior in recognition speed and flexibility. The effectiveness of this method is demonstrated through experiments of automatic robot driving and remote grasping.

Keywords: Robotic Applications, Robot Mechanism and Control
Abstract: This paper presents a novel design methodology for structurally reinforcing soft fingers with passive support structures, aiming to enhance mechanical capabilities without compromising compliance. We propose and evaluate multiple reinforcement strategies focusing on continuous sheets made from aluminum and a serial chain hinge support. We justify our design choices through both mathematical modeling and experimental validation. The approach emphasizes rapid prototyping using off-the-shelf components such as aluminum sheets and door hinges to ensure practicality and reproducibility. Our findings show that the hinge finger configuration significantly improves the horizontal (HA) and vertical (VA) grasping capabilities of the SofIA gripper. Comprehensive performance evaluation includes payload capacity, lateral deflection, grasping force, and object range. The best evaluated finger configuration has maximum grasping range coverage and achieved a maximum payload of 18.1 N. We further demonstrate the effectiveness of our gripper on the EGAD dataset, showcasing its robustness and applicability. Design files and implementation details are released under an open-source license to foster wider adoption and collaboration in soft robotics.

Keywords: Robotic Applications, Robot Mechanism and Control
Abstract: This study proposes a new automatic cleaning system for cat litter boxes that combines robotic arm technology with computer vision. Unlike traditional automatic cleaning systems that require special housing, our approach uses a general-purpose robotic arm and web camera, making it possible to retrofit existing litter boxes. We implemented high-precision waste detection technology using YOLO, achieving a mean Average Precision (mAP) value of 97.8%. Combined with coordinate transformation technology, our experimental evaluation achieved grasping success rates of 82.5% for standard objects and 89.6% for cylindrical objects. The waste detection technology through image recognition has the potential to provide new added value in monitoring cat health conditions. The results of this research present a new approach to automation in pet care products and are expected to serve as foundational technology for future pet-sitting robots.

Keywords: Robotic Applications, Artificial Intelligence Systems, Robot Vision
Abstract: This paper presents an object-centric semantic navigation (OCNav) framework for mobile robots based on visual-language navigation (VLN). With the advancement of large language models (LLMs), human-robot interaction has emerged as a crucial aspect of autonomous navigation in real-world environments. In particular, object semantics and spatial relations are essential for mobile robots to interpret and operate within home settings. Therefore, we propose a novel semantic topological graph that integrates semantic and spatial information into topological nodes in textual format. Each node additionally encodes a static level that reflects the positional permanence of objects to enhance the robustness of spatial semantic representation. Given a user instruction in natural language, the mobile robot interprets the command based on the semantic attributes encoded in the graph. Then, the target location is selected based on the highest confidence score, computed through the cosine similarity of sentence-BERT embeddings. Experimental results from both Habitat simulation and real-world deployment on the LG Q9 robot validate the effectiveness of the proposed OCNav framework.

Keywords: Robotic Applications, Industrial Applications of Control, Information and Networking
Abstract: This paper aims to establish a quantitative foundation for applying real time operating systems to robotics systems, where general purpose operating systems like Linux are commonly used but often fail to meet strict timing requirements, especially in time critical domains such as autonomous vehicles, missile guidance, surgical robotics, and nuclear systems. Although ROS2 has become the standard middleware in modern robotics, its deployment on real time operating systems remains rare due to lack of empirical evidence and practical adoption experience in the robotics domain. To address this gap, we quantitatively benchmark ROS2 communication latency across three platforms: standard Linux, soft real time Linux with a PREEMPT RT patch, and the commercial RTOS QNX. Experiments are conducted on an embedded board using identical publisher subscriber logic, under varying message sizes, publication frequencies, intra and inter process configurations, and stress conditions. The results show that QNX consistently achieves lower and more stable latency for small high frequency messages, but suffers degradation under multi node stress scenarios. Based on these findings, we propose a layered control structure in which RTOS handles low level real time tasks such as motor control and sensor feedback, while high level planning and computation remain on GPOS. This study provides a practical reference for robotics system designers considering real time architectures and outlines the trade offs and deployment strategies for adopting RTOS in embedded robotic control.

Keywords: Robotic Applications, Autonomous Vehicle Systems, Information and Networking
Abstract: This paper proposes a distributed asynchronous consensus ADMM (C-ADMM) algorithm for multi-agent task assignment. The strict global synchronization requirements of the conventional C-ADMM are relaxed through the introduction of partial barriers and bounded delays. The algorithm is implemented in two variants: a Ground Control Station (GCS)-based Distributed Asynchronous C-ADMM and a Decentralized Asynchronous C-ADMM that exchanges state information directly among agents. Both variants discard delay information that exceeds a bounded delay threshold and improve computational efficiency by allowing each agent to update without waiting for simultaneous responses of neighbors. MATLAB simulation results confirm that the proposed Distributed Asynchronous C-ADMM reduces computation time by 40–45% with only a 4–10% loss in optimality compared to the original C-ADMM-based MUR-TAP. The Decentralized Asynchronous C-ADMM reduces computation time by 46–52% at the expense of a 12–37% loss in optimality. These results demonstrate a flexible trade-off between convergence speed and solution accuracy in realistic distributed environments subject to network delays and information loss.

Keywords: Sensors and Signal Processing, Information and Networking
Abstract: This paper proposes a variable step-size diffusion least-mean-square algorithm based on an event-triggered combination method. While much of the existing research in distributed system estimation focuses on improving the convergence rate and accuracy, relatively little attention has been paid to energy efficiency in communications. Thus, this paper proposes an algorithm that aims to reduce communication cost with minimal performance degradation. A variable step size is proposed to achieve stable convergence and accurate estimation, and an event-triggered strategy is designed based on the derived mean square deviation. The dynamic triggering threshold is set to respond adaptively to abrupt decreases in the mean square deviation. The simulation results demonstrate that the proposed algorithm maintains desirable performance with significantly reduced communication cost.

Keywords: Robotic Applications, Artificial Intelligence Systems, Robot Mechanism and Control
Abstract: In real-world systems, reinforcement learning policies often exhibit a lack of action smoothness, making it difficult to ensure safe and reliable control. While several methods have been proposed to improve action smoothness, they often suffer from high computational cost and exhibit limited adaptability to complex dynamics. We propose Self-Distilled Conditioning for Action Policy Smoothing (SCAPS), a memory-efficient and adaptive framework that promotes smooth actions through self-distillation by aligning the current policy with an exponential moving average of its past behaviors in a scalable manner. We evaluate the proposed approach on a bipedal wheeled robot in Isaac Sim and perform sim-to-sim transfer to MuJoCo to assess generalization. Compared to CAPS and Vanilla PPO, SCAPS achieves smoother actions and lower memory usage, confirming its effectiveness in simulated control environments.

Keywords: Artificial Intelligence Systems, Robot Vision
Abstract: 3D range–inertial Simultaneous Localization and Mapping (SLAM) has driven significant advances in robot navigation across diverse environments. Although LiDAR sensors deliver highly accurate depth measurements, stereo camera–based methods often suffer from degraded performance due to distortion and sparsity in the reconstructed point clouds. To overcome these limitations, we propose a generative sparse network specifically designed for range–inertial SLAM with stereo cameras. Our approach leverages a 4D fully convolutional network to denoise of 3D point cloud maps, enabling more reliable pose estimation and mapping.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Control Theory and Applications
Abstract: Accurate and sample-efficient learning of mobile robot dynamics is crucial for high-speed model-predictive control. This paper introduces a novel approach that leverages both SE(2) equivariance and morphological symmetry for mobile robot dynamics learning. We propose a lightweight neural architecture that simultaneously incorporates these symmetries while maintaining computational efficiency. The model's performance is evaluated against traditional neural networks, demonstrating superior accuracy in predicting robot dynamics. Our architecture achieves this through a canonicalization layer for SE(2) transformations and an equivariant multi-layer perceptron that preserves morphological symmetry. Compared with a non-equivariant four-layer MLP, our network reduces positional mean-square error by 52% while maintaining robustness under adversarial SE(2) transformations. The proposed approach offers an effective solution for dynamics learning in resource-constrained mobile robots, bridging the gap between physics-based models and data-driven approaches.

Keywords: Information and Networking, Rehabilitation Robot, Artificial Intelligence Systems
Abstract: Real-world reinforcement learning applications often assume that there is no loss of data from the environment—an assumption that is highly unrealistic, as exemplified by networked systems that are prone to packet loss. One simple approach for such scenarios is a memoryless method that takes no action when timely data is missing. However, this approach exhibits highly task-dependent performance as the agent ignores environmental changes during the inactive periods. In this paper, we propose an imputation-based algorithm that predicts missing data and makes decisions based on the imputed ones, enabling robust performance even in environments with frequent data loss. The proposed algorithm is evaluated in an environment characterized by frequent information loss, and empirical results demonstrate that it outperforms the memoryless approach in terms of asymptotic performance.

Keywords: Artificial Intelligence Systems, Information and Networking
Abstract: Reliable estimation of the state of health (SOH) of lithium-ion batteries is critical for maintaining safe and efficient operation in real-world applications. However, dynamic operating conditions, such as tem- perature fluctuations and various charge-discharge patterns, induce non-stationary data distributions, which degrade the generalization capability of conventional SOH estimation models. To address this challenge, we propose a ConvLSTM with Continual Learning (CCL) framework that integrates a ConvLSTM-based spatiotemporal encoder with Elastic Weight Consolidation (EWC). The proposed method enables sequen- tial learning across different data distributions while retaining previously acquired knowledge, effectively mitigating catastrophic forgetting. We validate the effectiveness of CCL through two key experiments: (1) robustness evaluation under temperature varying environments, and (2) stability analysis under sequential data distribution shift scenarios, with a focus on balanced representation preservation. Experimental results show that CCL consistently outperforms baseline transfer learning approaches in both estimation accuracy and generalization performance. Moreover, CCL significantly reduces task-wise performance variance, high- lighting its robustness and adaptability under non-stationary conditions.

Keywords: Robotic Applications, Robot Mechanism and Control, Artificial Intelligence Systems
Abstract: Theorizing a chemical phenomenon requires numerous experiments. By allowing AI to perform these experiments, we can minimize human fatigue and produce consistent results. This paper presented an open-sourced method for building an experiment-specialized bi-manual robot platform, ALOHA-EA. We build ALOHA-EA locally, sourcing and simplifying it from the original ALOHA. And we also provided a detailed manual for recreating ALOHA-EA. Additionally, we demonstrated how to design episodes for the task by breaking down a big task into smaller ones. We demonstrated how to create our software for teleoperation and recording raw data from the platform. By not recording data in the RLDS format, we can adjust or modify our raw data into various models, such as ACT. Finally, we evaluated our proposed platform by brewing hand-drip coffee, which is safer but more challenging than real chemical experiments. We achieve ** percentage success rate in hand-drip coffee brewing with our proposed platform. Ultimately, we developed and published ALOHA-EA's manual, along with the *K-sized RLDS dataset, for open-source use.

Keywords: Artificial Intelligence Systems, Process Control Systems, Industrial Applications of Control
Abstract: This study presents a comparative evaluation of seven time-series deep learning models—CNN, LSTM, CNN-LSTM, Transformer, Conformer, Reformer, and Informer—for surrogate modeling of the Boil-Off Gas (BOG) reliquefaction process. To assess model performance under varying conditions, input sequence lengths (8, 16, 32, 64) and forecasting strategies (single-point and 32-step direct multi-step) were applied. Evaluation was conducted using accuracy metrics (sMAPE, MAE, IOA) and efficiency indicators (parameter count, inference time). In short-term prediction, all models demonstrated similar accuracy, with CNN exhibiting the highest efficiency due to its lightweight architecture and fast inference speed. In contrast, during long-term forecasting, CNN’s accuracy significantly deteriorated as the input sequence length increased, whereas most Transformer-based models maintained stable and high performance. Among them, the Conformer outperformed LSTM by approximately 40% in MAE and 26% in sMAPE, and surpassed the Transformer by about 28% in MAE, achieving the most consistent and accurate predictions. Originally developed for speech recognition, the Conformer has rarely been applied to process surrogate modeling.

Keywords: Artificial Intelligence Systems, Process Control Systems
Abstract: Hydrogen-based steelmaking is essential for steel industrial decarbonization, but its deployment requires stable renewable energy supply. Lithium-ion battery (LiBs) based energy storage systems (ESS) is a promising solution to this challenge. However, LiBs inevitably degrade due to material and electrochemical aging, leading to reduced storage capacity over time. This degradation limits the amount of usable green energy, making accurate prediction of battery health essential for reliable system operation. This study presents a data-driven model to estimate the State of Health (SOH) of LiBs using features derived from incremental capacity (IC) curves. From the Oxford dataset, we extract and compare three feature types: peak-based, statistical, and a combined hybrid set. A Gated Recurrent Unit (GRU) model is trained and validated through five rounds of Leave-One-Out (LOO) cross-validation. The model achieves consistent accuracy across various degradation patterns, effectively handling both peak- and statistics-dominant trends. Even without extensive tuning, the hybrid feature set enables high performance, highlighting the value of combining interpretable physical features with a lightweight neural network. These results emphasize the importance of thoughtful feature engineering and model design in building robust SOH prediction frameworks for renewable energy storage systems.

Keywords: Process Control Systems
Abstract: With the rapid growth of electric vehicles (EVs), efficient battery thermal management (BTM) is essential for ensuring battery performance, safety, and longevity. This paper proposes an integrated three-step optimization framework for energy-efficient BTM using nonlinear model predictive control (NMPC). The method combines a control-oriented electro-thermal battery model, static and dynamic modeling of the coolant and refrigerant circuits, and an extended Kalman filter (EKF) for state estimation. The three-step process involves: (1) real-time estimation of the maximum cooling capacity of the battery chiller, (2) NMPC-based optimization of coolant flow and temperature under physical constraints, and (3) optimization of compressor and fan power to achieve cooling demand. Simulation results demonstrate that the proposed approach reduces energy consumption by up to 7.9% compared to rule-based control, while maintaining precise temperature regulation.

Keywords: Process Control Systems, Artificial Intelligence Systems, Industrial Applications of Control
Abstract: To extend the driving range of electric vehicles (EVs), optimal operation of the heat pump system is essential. Since real-time optimal control requires a dynamic model of the EV heat pump system, a control-oriented model is necessary. However, a trade-off exists between model complexity and computational efficiency; conventional control-oriented models often lack the accuracy required for advanced control strategies such as model predictive control (MPC). This study proposes a novel control-oriented modeling approach for an EV heat pump system integrated with waste heat recovery via an oil cooler. The model adopts a hybrid framework combining physics-based and data-driven components. A pseudo-steady-state approximation (PSSA) constructs the white-box structure, while a deep neural network (DNN) estimates key parameters as a black-box model. The proposed model is validated using MATLAB Simulink simulation data and compared with a long short-term memory (LSTM) model. Evaluation in both training and extended domains shows high predictive accuracy with low computational cost, supporting its suitability for real-time control.

Keywords: Artificial Intelligence Systems, Robotic Applications
Abstract: In this work, we use a CVAE-based imitation learning approach to train task-specific robot behaviors from limited data without reward functions. To improve robustness, failure cases are included in training. However, in multi-task settings, overlapping demonstrations can cause ambiguity during inference. Experiments show that structured, clearly separated demos help resolve this issue and improve task distinction.

Keywords: Robotic Applications, Robot Vision, Robot Mechanism and Control
Abstract: As robotics technology advances, robotic arms are increasingly being used to provide automation services across a wide range of fields. This paper proposes an automated buffet table cleaning system utilizing an RGB-D camera and a 6-axis robotic arm. The system is designed to automatically remove empty plates and bowls from tables in buffet restaurants while customers leave their seats to get more food. If a mobile manipulator arrives at the table, a camera first captures an image of the tabletop. Then, the widely used YOLACT model is used to segment and classify the tableware in the 2D image, from which the 2D coordinates of their bottom edges are extracted. These 2D coordinates are then converted into 3D positions using the RGB-D camera. The 6-axis manipulator moves to calculated positions and performs pick-and-place operations. In addition, the system includes a custom gripper that rotates its fingers according to the inclination of its side surfaces, allowing for stable and hygienic grasping. This process is repeated until all tableware items on the table are successfully relocated.

Keywords: Robot Mechanism and Control, Autonomous Vehicle Systems, Control Theory and Applications
Abstract: Holonomic multi-steerable-wheel-drive (MSWD) platforms enable true omnidirectional motion but suffer kinematic and numerical challenges near zero yaw rate. We first identify that the classical instantaneous center of rotation (ICR) diverges and that Jacobian pseudo-inverse–based velocity calculations become ill-conditioned under low yaw. To address these issues, we introduce a bounded, continuously differentiable spherical-coordinate ICR reformulation and a direct screw-theory projection method that is algebraically equivalent yet avoids pseudo-inverse instability. In over-actuated MSWD systems, this projection also leverages redundant degrees of freedom for natural wheel-speed distribution. MATLAB simulations across multiple MSWD layouts confirm that our methods eliminate singularities, ensure smooth, stable steering and speed control, and generalize to any steerable-wheel configuration.

Keywords: Robot Vision
Abstract: Accurate point cloud data registration is crucial for various applications, particularly in automated manufacturing processes. In the case of automatic shoe buffing, the generation of buffing paths is crucial for the buffing quality, which is heavily based on precisely registered shoe point cloud data. The Iterative Closest Point (ICP) algorithm, widely used for point cloud registration, exhibits limited performance in multi-view scenarios where the overlap between point sets is insufficient. This limitation arises from its reliance on local geometric correspondences, which often leads to suboptimal alignment or convergence to incorrect local minima when the shared point cloud regions are small. In this paper, we propose an Indirect-Iterative Closest Point(I-ICP) registration method, adapting ICP on an auxiliary surrogate rather than using it on target shoe for precise data matching. The auxiliary surrogate is a simple, small tool that enables ICP to be performed on more easily registered point cloud data. We discuss the enhanced accuracy of data registration achieved through the proposed method, demonstrating its potential to improve buffing path generation in automatic shoe buffing.

Keywords: Artificial Intelligence Systems, Robotic Applications, Process Control Systems
Abstract: Imitation learning has been widely adopted as an effective tool for enabling robots to perform complex manipulation tasks. However, in real-world assembly scenarios that demand high precision, imitation policies trained solely on offline demonstrations often struggle with generalization and fail to recover from errors. In this paper, we propose a novel framework that augments a pre-trained imitation policy with a residual policy trained through reinforcement learning using human feedback. The residual policy learns corrective actions based on real-time human interventions, and is guided by a reward structure that accounts for the frequency, context, and timing of each intervention. The corrective actions generated by the learned residual policy are combined with the actions from the original imitation learning policy to produce a unified command for robot execution. This work emphasizes the methodological contribution, providing a detailed system design, reward shaping strategy, and data collection pipeline. Such a framework is expected to be applicable to a wide range of learning-based algorithms, serving as a general strategy for enhancing the robustness and adaptability of robot control policies.

Keywords: Artificial Intelligence Systems, Robotic Applications, Industrial Applications of Control
Abstract: This paper presents a robotic assembly framework that combines Vision-Language Models (VLMs) with imitation learning for assembly manipulation tasks. Our system employs a gripper-equipped robot that moves in 3D space to perform assembly operations. The framework integrates visual perception, natural language understanding, and learned primitive skills to enable flexible and adaptive robotic manipulation. Experimental results demonstrate the effectiveness of our approach in assembly scenarios, achieving high success rates while maintaining interpretability through the structured primitive skill decomposition.

Keywords: Artificial Intelligence Systems, Robot Vision, Sensors and Signal Processing
Abstract: Temporal action segmentation is essential for learning robotic skills from demonstration, particularly in contact-rich manipulation tasks composed of multiple sequential sub-actions. While prior methods use visual and proprioceptive signals, they typically encode motion data implicitly. In this paper, we propose a constraint-informed temporal action segmentation framework that explicitly integrates motion constraint information derived from end-effector velocity and force/torque profiles into a multistage prediction model. The key idea is to use constrained motion directions as signals for sub-task transitions and to jointly predict both action labels and constraint states at each stage. We validate our approach on a simulated peg-in-hole insertion task and compare it with several baselines. Results show that incorporating constraint cues improves segmentation accuracy and temporal alignment, highlighting the value of structured physical interaction signals in segmenting complex robotic tasks.

Keywords: Artificial Intelligence Systems, Human-Robot Interaction, Robotic Applications
Abstract: We present a modular robotic system designed for precision assembly tasks in flexible manufacturing environments. Our method addresses three critical challenges: (1) Generating robot motion for assembly tasks: manual planning or teaching is complex and inefficient due to variability in tasks and fixtures. Our solution leverages human demonstration and teaching, enabling the robot to learn effective strategies directly from human experts. (2) Collecting demonstration data: traditional data collection methods are cumbersome and impractical. We introduce a portable and user-friendly teaching interface incorporating a GoPro camera, HTC Vive trackers, and a gripper encoder. This interface supports real-time demonstrations through physical interactions, simulation, or teleoperation, facilitating rapid and intuitive data acquisition. (3) Lack of precision and explicit constraint handling in learned models: most learning-based motion models implicitly encode constraints, resulting in insufficient precision and potential violations of motion limits. To address this, we apply a refinement step using finite-horizon online optimization that incorporates motion constraints and goal-state accuracy into its cost optimization process. Our approach involves an initial motion trajectory generated by a learned model, followed by constraint-aware optimization refinement. Experimental evaluation on a precision pick-and-place task in power relay assembly demonstrates that a learned motion model combined with refinement ensures robust performance and high precision, achieving successful task completion across various test scenarios

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Robotic Applications
Abstract: An adaptive coverage path planning (CPP) method is proposed for multiple fixed-wing unmanned aerial vehicles (UAVs) operating in dynamic and unstructured environments. The method enhances computational efficiency through a two-stage framework: region partitioning and path computation. A learning-based initial solution estimation module is introduced to improve convergence in the baseline algorithm. To address the dynamic constraints of fixed-wing UAVs, a path computation method based on convex polygonization and principal axis analysis is developed. Simulations and field experiments demonstrate significant improvements in planning speed and coverage reliability under time-constrained conditions.

Keywords: Autonomous Vehicle Systems, Human-Robot Interaction
Abstract: As the aging society progresses, the demand for personal vehicles (PVs) is increasing to support elderly mobility. Research has focused on improving the operability of PVs and applying autonomous driving technologies. Mixed reality (MR)-based training systems improve driving skills, but the system relies on manual operation, which raises safety concerns. Therefore a brain-computer interface (BCI) using steady-state visual evoked potentials (SSVEPs) provides an alternative for autonomous driving. Although lowering the confidence threshold improves the detection speed, however it could increases the false detections, requiring reliability improvements. This study proposes an LSTM-based deep learning system to predict a driver's steering direction using SSVEP confidence values and gaze data.

Keywords: Autonomous Vehicle Systems, Robot Vision, Robotic Applications
Abstract: This paper presents a 360-degree surround-view generation method using projective transformation and homography estimation from multiple wide-angle cameras. Each camera was individually calibrated, and homography matrices were computed by matching detected chessboard corners to predefined templates in a unified bird’s-eye view (BEV) layout. To achieve seamless integration, the BEV canvas was divided into eight logical regions, and a distance-based spatial blending technique was applied in overlapping areas. This blending strategy computed pixel-wise weights based on the Euclidean distance to the outer boundaries of each camera's valid projection zone, effectively minimizing visual seams, ghosting, and illumination artifacts. The system demonstrated consistent performance under both daytime and nighttime conditions. With an average processing time of approximately 0.28 seconds per frame, the proposed framework enables near real-time operation on embedded platforms and provides a robust foundation for downstream perception tasks such as road extraction and autonomous navigation. Code to reproduce our results is available at: https://github.com/souldeathz/360-degree-Surround-View-appl ication.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Navigation, Guidance and Control
Abstract: This paper presents a leader–follower formation control framework for differential-wheeled mobile robots (DWMR), combining Artificial Potential Fields (APF) for obstacle avoidance with fuzzy-tuned Proportional– Integral–Derivative (PID) controllers for trajectory tracking. The leader follows time-parameterized trajectories—specifically circular and lemniscate paths—while follower robots maintain a dynamic, rotating formation by computing real-time relative offsets. APF-based repulsive forces handle collision avoidance with static obstacles and neighboring agents. To improve responsiveness, fuzzy logic adaptively adjusts PID gains based on position and heading errors. Simulations with six follower robots, initialized randomly, validate the proposed method’s effectiveness in maintaining formation, avoiding four static obstacles, and handling trajectory curvature. Quantitative evaluations using root mean square errors confirm low tracking errors and consistent inter-agent spacing. The results demonstrate that the proposed control scheme is robust and adaptable for multi-robot coordination in cluttered environments. Future works will include extending the approach to dynamic obstacles, heterogeneous agent systems, and real-world implementation.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications
Abstract: This research proposes a safe trajectory planning method for AMR (Autonomous Mobile Robots) using game theory and Control Barrier Function (CBF). In game theory, trajectory planning is performed using ILQ Games, which is a nonlinear extension of dynamic games. However, ILQ Games have two difficult problems: safety and deadlock. Therefore, we define aggressiveness between AMRs and add CBF constraints using Augmented Lagrangian method based on aggressiveness. Vehicles with low aggressiveness are assigned CBF constraints, while those with high aggressiveness are assigned position constraints before changing to CBF to consider the advantages of both game theory and CBF. The effectiveness of the proposed method is verified through simulations in multiple scenarios and comparisons with existing conventional research results.

Keywords: Sensors and Signal Processing, Autonomous Vehicle Systems, Robot Vision
Abstract: The long-term deployment of autonomous vehicles often suffers from gradual degradation in sensor alignment due to vibration, thermal variations, and mechanical stress. This paper proposes an online extrinsic calibration framework that compensates for time-varying drift without relying on prior calibration information. By using the globally fused trajectory as a reference, the system detects and corrects inter-sensor inconsistencies in real-time at each frame. The framework operates asynchronously and supports heterogeneous sensors, including LiDARs, monocular cameras, IMU, and RTK-GPS. Experiments conducted in real-world urban driving scenarios demonstrate that the proposed method successfully reconstructs body-frame trajectories with significantly reduced drift for all sensors. These results confirm the effectiveness of the framework in maintaining extrinsic consistency and improving odometry quality, even in the absence of static calibration baselines. The system is computationally lightweight and deployable in embedded automotive platforms, enabling long-term robust operation without manual recalibration.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems
Abstract: Machine learning models including state-of-the-art deep learning models are intrinsically vulnerable to adversarial examples, where imperceptible perturbations could deceive AI recognition system in various intelligent applications, altering its response with high confidence. In this work, we investigate the effectiveness of physical adversarial attacks and defenses based on simulated environments with effective adversarial examples. The study presents successful attack results, demonstrating the vulnerability of object detection systems to physical adversarial attacks. The research utilizes synthetic image data and simulation to generate adversarial examples, providing a cost-effective and scalable method for testing the robustness of object detection systems. Experimental results on vehicle detection demonstrate that the physical attacks are quite feasible and adversarial training for defense techniques can reduce the impact of the underlying attacks.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Robot Vision
Abstract: The inquiry into the essence of intelligence is a fundamental and stimulating inquiry that transcends multiple academic disciplines, such as philosophy, neuroscience, cognitive science, computer science, and brain science. This paper delves into the concept of embodied intelligence from foundational principles, emphasizing the role of embodiment in cognitive systems and proposing a framework for embodied cognitive systems. By incorporating pertinent artificial intelligence technologies, we apply this framework to the development and simulation of intelligent driving systems. Our methodology integrates the autonomous vehicle's perception of its surroundings with its self-state estimation to create a unified cognitive system. This interdisciplinary strategy seeks to connect theoretical insights with practical applications, furthering our comprehension and capabilities in the advancement of intelligent autonomous vehicles. Through comprehensive simulations and practical implementations, we illustrate how embodied cognition can enhance the performance of autonomous vehicles in intricate real-world settings. This research not only improves the operational effectiveness of intelligent driving systems, but also contributes to the broader domain of artificial intelligence by underscoring the significance of embodied cognitive principles.

Keywords: Artificial Intelligence Systems, Information and Networking, Multimedia Systems
Abstract: The development of critical thinking skills has become a central priority in contemporary education, especially in preparing students for complex problem-solving, informed decision-making, and lifelong learning. This paper presents the design and initial deployment of a bilingual (Kazakh–Russian) AI-powered educational tool aimed at fostering critical thinking among secondary school students. The application leverages NLP and large-language models to analyze literary or instructional texts and generate Bloom-aligned questions at all cognitive levels, from basic recall to analysis, evaluation, and creative synthesis. The system enables teachers to upload text-based content, upon which the AI automatically produces structured level-specific questions tailored to the material. An integrated scoring and feedback mechanism supports formative assessment and encourages self-regulated learning. The preliminary evaluation includes an expert review of 30 generated items (three teachers of the literature) and early engagement metrics from a pilot with Class 6A (n=18). Baseline data collection is underway to allow a comparative analysis of the impact of the tool on the development of critical thinking. This case study illustrates the potential of AI to deliver personalized and cognitively engaging learning experiences in formal education. The paper details the system architecture, instructional logic (including retrieval-augmented, prompt-driven generation), and pilot evaluation framework, providing a foundation for future empirical studies and scalable implementation in diverse educational contexts.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Robot Vision
Abstract: This study explores the application of deep learning techniques for classifying targets such as people, drones, and cars using micro-Doppler radar signals. Micro-Doppler radar excels in capturing subtle motion features critical for target differentiation. We conduct a comparative analysis of several deep learning architectures, Convolutional Neural Networks (CNN), DenseNets, and Vision Transformers (ViT), together with classical preprocessing (FFT, PCA, SVD). Our focus is on lightweight models optimised for real-time embedded deployment. The SVD-augmented CNN (rank-6 SVD with 3-frame stacking) delivers the best mean accuracy of 98.19% (5-fold), surpassing our lightweight baseline CNN (98.12%) and DenseNet (98.05%), with ViT at 94.33%. FFT in the preprocessing pipeline further enhances frequency-component extraction for stability. These advances improve the accuracy-efficiency trade-off in radar target classification and support practical deployment in surveillance and autonomous navigation scenarios.

Keywords: Artificial Intelligence Systems
Abstract: Recent advances in foundation models have enabled the broad transfer of prior knowledge to diverse downstream tasks. However, fully fine-tuning these models is often infeasible in practice due to prohibitive memory and storage costs, especially in resource-constrained physical systems. Parameter-efficient fine-tuning methods mitigate this challenge by updating only a small subset of parameters, allowing efficient adaptation while preserving pretrained knowledge. Among these, parameter selection has demonstrated strong effectiveness, but existing approaches typically require storing gradients of all parameters, leading to substantial memory overhead. To overcome this limitation, we propose a hybrid framework that integrates gradient-based parameter selection with low-rank matrix decomposition. Our approach significantly reduces memory usage compared to prior methods while maintaining competitive performance across multiple benchmark datasets.

Keywords: Artificial Intelligence Systems, Robotic Applications, Industrial Applications of Control
Abstract: Dexterous manipulation using multi-finger hands require precise control and adaptability. Training multifinger robot hands to imitate expert-like behaviors still remains challenging, since traditional imitation learning (IL) algorithms suffer from state occupancy mismatches. In this paper, we propose a novel IL algorithm, which employs a state-based discriminator to explicitly align the policy’s state distribution with demonstrations, thereby imitating expertlike behaviors. We also introduce a novel task environment that utilizes tactile sensors, which allows for more sensitive feedback on complex deformable objects. We validate our approach in tasks involving deformable objects, including a challenging pour task, where it is shown that tactile feedback is critical for precise force control and dynamic adjustments of the multi-finger hand.

Keywords: Robot Mechanism and Control, Artificial Intelligence Systems, Robot Vision
Abstract: In recent years, vision-language-action models (VLAs) have become dominant in robotics domain. In this work, we leverage both visual and textual information to build a versatile grasp network. To test our algorithm, we set up an experimental environment where objects of identical color were grouped and placed on the same plate. Evaluated in our environment, our framework demonstrates superior grasping performance.

Keywords: Robot Mechanism and Control, Autonomous Vehicle Systems, Human-Robot Interaction
Abstract: In this paper, we introduce a compliant spoke track mechanism that enables stable driving on rough terrain. The hollow sprocket and main roller are connected to the inner and multiple suspensions, and can absorb impacts from various directions. Dynamic analysis for modeling was performed to optimize the stiffness of the suspension.

Keywords: Robotic Applications, Sensors and Signal Processing
Abstract: Accurate motion estimation of mobile robots operating in unstructured environments with uneven terrain is a critical challenge in agricultural robotics. LiDAR sensors, widely used for environment perception, suffer from motion-induced distortion. This distortion becomes especially problematic in outdoor, off-road scenarios where terrain variability introduces rapid and unpredictable robot motion. Traditional de-skewing methods often rely solely on IMU data, which may be insufficient under erratic dynamics or in regions with highly uneven terrain. This paper presents a sensor fusion-based approach that integrates GPS and IMU data for robust real-time LiDAR distortion correction. By combining high-frequency IMU readings with globally referenced GPS data, the system estimates continuous motion trajectories to compensate for motion during each LiDAR scan. The proposed method improves point cloud integrity and spatial alignment across frames, facilitating reliable odometry estimation and localization in complex field conditions. Extensive field tests conducted in agricultural environments demonstrate that our fusion-based deskewing significantly reduces point cloud distortion and improves the accuracy of downstream tasks such as mapping and pose estimation. The system runs in real-time and is designed to be deployed on agricultural robots for precision farming and autonomous field operations.

Keywords: Robot Vision, Robotic Applications, Sensors and Signal Processing
Abstract: Visual-Inertial Odometry (VIO) systems often struggle in visually challenging environments due to two common issues: ensuring reliable feature tracking and managing uncertainties that arise from integrating diverse observations. This paper presents a robust multi-camera VIO system that addresses these limitations through two key innovations. First, we develop a learning-based feature extraction module specifically designed for multi-camera configurations, which generates stable descriptors to validate optical flow correspondences and retain only high-quality features while reducing computational overhead. Second, we introduce an adaptive weighting scheme in the backend that employs weighting techniques across multiple cameras to quantify tracking uncertainties and assign appropriate weights to different viewpoints based on their reliability. The system seamlessly integrates an arbitrary number of cameras to enhance robustness in challenging scenarios. Extensive evaluation on datasets featuring visually degraded conditions demonstrates the effectiveness of our approach, achieving up to 90% reduction in Absolute Trajectory Error (ATE) compared to state-of-the-art VIO methods.

Keywords: Robotic Applications, Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: Recent advancements in deep learning and transformer architectures have accelerated research on Large Language Models (LLMs), enabling multimodal reasoning across diverse domains. However, deploying LLMs in mobile robotics remains difficult due to high computational costs and output instability. This paper proposes a framework that integrates a lightweight multimodal LLM into the path-planning layer of mobile robots. Given a terrain map and mission directive, the LLM selects the most appropriate algorithm from a predefined set of classical planners (A*, Theta*, PRM, RRT*, RRT*-Smart), based on high-level reasoning. By restricting the LLM’s role to algorithm selection, the framework minimizes hallucination and improves reliability. Experimental results using Gemma3 variants in obstacledense environments show that the LLM consistently chooses algorithms that balance path length, computation time, and trajectory smoothness. These findings demonstrate the feasibility of incorporating lightweight LLM reasoning into resource-constrained robotic platforms and highlight its potential for intelligent, cost-effective autonomous navigation.

Keywords: Robotic Applications, Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: Legged robots possess unique capabilities to navigate unstructured environments, thanks to their articulated legs and advanced autonomy systems. However, maintaining fully autonomous operation for a long duration without human intervention remains a significant challenge in large-scale settings. Although topological mapping with interactive annotations has recently been used to make quadruped robots more flexible, these methods typically face major bottlenecks in decision making. In this paper, we present ATBT, a unified framework managed by behavior tree to provide robust and adaptive autonomy for quadruped robots in a wide variety of environments. ATBT builds a sparse topological graph whose nodes and edges encode semantic information, mission behaviors, failure recovery strategies, and dynamic parameter adaptation rules that boost the system’s autonomy. By embedding interactive annotations directly into this graph, our system gains additional resilience and flexibility. ATBT runs entirely on the robot's edge computer with a ROS 2 native stack. It combines a high-performance SLAM and low-level control under a high-level behavior tree layer. We validate ATBT on the GhostRobotics Vision 60 platform in a challenging urban surveillance scenario, constructing a topological graph of 27 nodes and 44 edges augmented with four mission subtrees, two behavior strategies, and four dynamic parameter adaptation rules. We demonstrate that topological map-based BT autonomy not only performs effectively in varied environments but also supports complex missions such as industrial inspection.

Keywords: Robotic Applications, Artificial Intelligence Systems, Robot Vision
Abstract: In cluttered scenes, effective object manipulation often requires both precise grasping and proactive scene rearrangement. We propose a dual-branch reinforcement learning framework that separately predicts grasp position and orientation, trained via supervised pretraining and shaped rewards to ensure stable and sample-efficient learning. To minimize unnecessary pushing, we model the coordination between grasp and push agents as a Stackelberg game, where the push agent acts only when grasp success is unlikely, to enhance downstream grasp success.. Experimental results in simulation show that our method improves grasp success and action efficiency, outperforming existing baselines in both success rate and policy economy.

Keywords: Robotic Applications
Abstract: Teleoperated mobile robotic systems are critical in remote and hazardous environments, but they pose significant control challenges due to nonlinear dynamics, system uncertainties, and external disturbances. While traditional PID controllers offer simplicity and ease of implementation, their sensitivity to parameter variations and external perturbations limits their applicability in complex teleoperation scenarios. To address these limitations, this paper proposes an enhanced Self-Coupling PID (SC-PID) controller that incorporates an exponentiallydefined funnel boundary into the speed factor formulation. This integration improves robustness, reduces the tuning burden, and maintains structural simplicity. The stability of the proposed controller is analytically demonstrated. Simulation results confirm its superior tracking performance and resilience in both disturbance-free and disturbance-present conditions, validating its effectiveness for robust teleoperation control.

Keywords: Robotic Applications, Artificial Intelligence Systems, Process Control Systems
Abstract: Accurate Ground Reaction Force (GRF) estimation is crucial for legged robots to perform complex tasks, yet conventional sensor-based methods suffer from fragility and impracticality. Indirect estimation is challenged by intricate system dynamics and the complex, dynamic nature of friction. This paper proposes a novel MIMO Hysteretic Friction-aware Lagrangian-Based Network (M-Net) for sensorless GRF estimation. M-Net is comprised of a Lagrangian-based Network (L-Net) for modeling energy-conservative rigid-body dynamics and a Hysteresis-aware Network (H-Net) utilizing Temporal Convolutional Networks (TCNs) to capture dynamic friction characteristics, including hysteresis and coupling effects. By integrating physics-informed constraints into the loss function, enhanced data efficiency and model generality are achieved. The performance of M-Net is validated through MuJoCo simulations on a quadruped robot leg, where it is compared against conventional DeLaN and classical friction compensation methods. Results demonstrate that M-Net significantly improves joint torque and GRF estimation accuracy, enabling robust sensorless GRF prediction by precisely modeling the complex dynamics of systems with dynamic friction.

Keywords: Robotic Applications, Robot Vision, Autonomous Vehicle Systems
Abstract: This paper proposes a semantic information-based autonomous docking framework that enables mobile robots to recognize and approach docking positions autonomously without manual reconfiguration, even when charging station locations are altered by external factors.

Existing autonomous docking systems have limitations in that they require charging station positions to be set as fixed coordinates and need manual configuration when positions change. To address this problem, this study proposes a method where robots utilize RGB-D cameras to recognize stations and estimate positions based on semantic information.

This approach enables docking using only cameras without additional devices such as IR sensors, thereby simplifying hardware configuration. Furthermore, this represents a practical application of semantic information processing to real-world robots, demonstrating the potential for autonomous docking technology that can adapt to environmental changes.

This paper shows that robots can reliably navigate to charging stations autonomously, even when the physical environment has changed, by leveraging semantic-based recognition in the docking system.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: Impediment shirking could be a key necessity for independent unmanned aerial vehicle operations, particularly in complex areas. This consideration presents a reenactment system created in MATLAB/Simulink, which coordinates real-world landscape information inferred from Advanced Territory Height Information (DTED). The chosen landscape speaks to the zone around the Affect Presentation and Tradition Center in Nonthaburi, Thailand, and is imported into the UAV Situation Originator to form a practical flight situation. The UAV demonstration incorporates a six-degree-of-freedom (6 DOF) energetic representation and a 360-degree LiDAR sensor to identify impediments in genuine time. A crossbreed control approach utilizing both Model Predictive Control and PID controllers is utilized to empower an exact and responsive route. Reenactment came about to illustrate that the UAV can effectively maintain a strategic distance from both inactive and moving deterrents without abusing security edges. The integration of real geographical highlights with sensor-based deterrent location altogether upgrades the authenticity and pertinence of recreation. This system serves as a dependable stage for testing and approving UAV route frameworks some time recently real-world sending.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Robotic Applications
Abstract: In this paper, we propose a 3D path planning method that integrates the A* algorithm with the octree structure. Unmanned Ground Vehicles (UGVs) and legged robots have been extensively studied, enabling locomotion across a variety of terrains. Advances in mobility have enabled obstacles to be regarded not only as hindrances to be avoided, but also as navigational aids when beneficial. A modified 3D A* algorithm generates an optimal path by leveraging obstacles during the planning process. By incorporating a height-based penalty into the cost function, the algorithm enables the use of traversable obstacles to aid locomotion while avoiding those that are impassable, resulting in more efficient and realistic path generation. The octree-based 3D grid map achieves compression by merging high-resolution nodes into larger blocks, especially in obstacle-free or sparsely populated areas. This reduces the number of nodes explored by the A* algorithm, thereby improving computational efficiency and memory usage, and supporting real-time path planning in practical environments. Benchmark results demonstrate that the use of octree structure ensures an optimal path while significantly reducing memory usage and computation time.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: Inertial measurement units (IMUs) are widely used for estimating the state of mobile robots due to their ability to provide inertial data, including acceleration and angular velocity, at high measurement rates. However, low-cost IMUs are susceptible to errors and noise. To overcome these limitations, one effective approach is the use of a virtual IMU (VIMU), which fuses measurements from multiple IMUs to form a VIMU. Conventional VIMU methods based on the Kalman filter (KF) typically adopt the standard KF, which uses a fixed process noise covariance matrix. This fixed process noise covariance matrix can lead to performance degradation under fast dynamic maneuvers where model uncertainty increases. To address this issue, this paper proposes a multi-IMU fusion method using an adaptive Kalman filter (AKF) that dynamically adjusts the process noise covariance matrix. The proposed method is evaluated on a mobile robot platform by comparing its position estimation accuracy against that of the standard KF under the fast dynamic maneuver condition. Experimental results demonstrate that the proposed method achieves significantly improved performance compared to the standard KF in fast dynamic maneuvers.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Control Theory and Applications
Abstract: Autonomous Underwater Vehicles (AUVs) are essential platforms for various underwater missions; however, their limited onboard battery capacity restricts the duration of fully autonomous operations. To overcome this constraint, Unmanned Surface Vehicles (USVs) can be deployed to cooperate with AUVs by providing intermittent charging and communication relay. This paper proposes a simple and effective waypoint generation strategy that allows the USV to track an AUV in a more energy-efficient and stable manner. The key idea is to generate a target point for the USV based on the midpoint between the AUV’s current position and its closest point on the predefined mission path. This approach smooths out the USV trajectory, avoids unnecessary oscillations, and maintains a communication range between the two vehicles. A MATLAB-based simulation environment was developed to evaluate the performance of the proposed strategy. The results demonstrate that, compared to the conventional Line-of-Sight (LOS) tracking method, the proposed algorithm yields smoother trajectories and reduced maneuvering effort, which are crucial for long-duration missions.

Keywords: Navigation, Guidance and Control, Robotic Applications, Robot Mechanism and Control
Abstract: This paper presents ALaM (Adaptive Locomotion and Manipulation), a novel approach for quadruped robots that enables pedipulation with any leg as a manipulator while maintaining balance and locomotion with the remaining three legs. Unlike previous methods that only utilize a designated leg as a manipulator, ALaM enables any leg as manipulator based on the task goal position. We propose a gravitational moment minimization reward for training pedipulation using the centroid of feet as a pivot point. Experimental results demonstrate that ALaM achieves better stability during manipulation, reduces tracking error by approximately 50% compared to state-of-the-art methods, and completes multi-goal tasks more efficiently by dynamically switching manipulation legs.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems, Robot Mechanism and Control
Abstract: This paper presents a torque vectoring control system that integrates Model Predictive Control (MPC) with Gaussian Process (GP) regression. Traditional MPC approaches may experience performance limitations when confronted with model uncertainties or varying operational environments. To overcome these challenges, this work utilizes GP to learn and compensate for unmodeled vehicle dynamics and external disturbances. The proposed framework incorporates a dynamic online adaptation mechanism for the GP model, continuously updating its predictions through recent operational data via a buffering strategy and periodic hyperparameter optimization. This adaptive learning approach, referred to as Streaming GP, improves system control performance and robustness. Simulation studies on a comprehensive vehicle model demonstrate the superior torque vectoring capabilities of the proposed method across challenging driving conditions compared to baseline approaches.

Keywords: Control Theory and Applications, Sensors and Signal Processing, Industrial Applications of Control
Abstract: Stabilized gimbal systems are now widely used in both civil and military applications for operations, including observation, target tracking, target identification, and communication. As a result, systems with various sizes, inertial, and elastic properties will be expected during their operations. Fixed-coefficient PI/PID controllers, which are only meant for use under normal circumstances, suffer performance loss when faced with various unknowns and/or unknown variables, such as changing payload, disturbance, and shifting dynamics. This work used a Model Reference Adaptive Control - Proportional Integral (MRAC-PI) controller based on Lyapunov rules that can vary with the system and its operational environment for speed control of the gimbal. The controller was tested using sine and step speed commands in various inertial setups, and its stabilization accuracy was also evaluated under external disturbances. It has been noticed that despite system changes and different speed commands, the MRAC-LYAPUNOV-PI controller behaves quite similarly in all situations and continues to function without the need for any adjustments during operation.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems
Abstract: We propose a trajectory tracking control for differential drive mobile robots using tube-based model predictive controller. The proposed controller constructs a reference tracking error model by generating a reference model matrix based on the error. While conventional Tube MPC approaches address the regulation problem, this work extends the framework to handle the trajectory tracking problem, enabling the system to follow time varying reference trajectories more effectively. Considering the linear time-varying system characteristics of the mobile robot, the tube area is updated online at each sampling step. The nominal control input is computed by solving an optimization problem subject to tightened constraints based on the update tube, while the feedback control input is obtained derived from the discrete time riccati equation. The effectiveness of the proposed controller is validated through numerical simulations. Compared to the conventional Tube MPC, the proposed method achieves significantly lower tracking errors, demonstrating improved tracking accuracy and faster convergence.

Keywords: Control Theory and Applications, Industrial Applications of Control, Robot Mechanism and Control
Abstract: The cylinder system of a hydraulic excavator was modeled, capturing the relationship from lever input to cylinder piston speed. The model incorporated equations representing valve motion induced by lever input, oil pressure variations resulting from valve motion, and piston displacement changes driven by pressure differences across the piston. A simulator was built based on these differential equations, and a transfer function model was derived through system identification. Based on the transfer function, a two-degree-of-freedom control system combining feedforward and feedback control was designed. The feedforward controller uses adaptive control to estimate the effect of external forces based on the deviation between actual and target piston speeds and adjusts its gain accordingly. The feedback controller applies sliding mode control, ensuring the system trajectory follows a specified sliding surface toward the target. The proposed method was validated using the constructed simulator and successfully tracked the target speed under variable external force conditions.

Keywords: Control Theory and Applications, Industrial Applications of Control, Control Devices and Instruments
Abstract: Ride height control systems adjust the distance between a vehicle’s chassis and the road surface to improve ride comfort, aerodynamics, and underbody protection. Among various types, electro-hydraulic systems use hydraulic actuators to lift or lower the vehicle body. In such systems, ride height is usually measured by sensors mounted on the vehicle body. However, these sensors do not directly measure the actuator’s internal position. As a result, the controller may continue commanding motion even after the actuator reaches its stroke limit. This can lead to noise generation, mechanical wear, and reduced ride quality. To solve this problem, we propose a model-based method to estimate the internal position of a single-acting actuator using only measurable signals. First, a full-order model of the system is developed, including the motor, pump, valves, and actuator dynamics. Then, a reduced-order model is derived using singular perturbation to simplify the computation. An extended Kalman filter is designed based on this reduced model. Simulation results are validated against experimental data obtained from vehicle-level tests. The proposed method shows accurate and robust performance under various operating conditions and sensor noise, making it suitable for practical ride height control systems.

Keywords: Navigation, Guidance and Control, Robotic Applications, Artificial Intelligence Systems
Abstract: Efficient task execution in automated warehouses is often challenged by limited task endpoints relative to the number of agents. To address this, we propose a Multi-Target Safe Interval Path Planning (MTSIPP) method for the Multi- Agent Pickup and Delivery (MAPD) problem. Unlike prior approaches that restrict endpoint sharing to prevent conflicts, MTSIPP enables agents to sequentially visit a pickup location, a delivery location, and a designated parking location by leveraging Safe Interval Path Planning (SIPP) in continuous time. Pre-planned paths of other agents are used as dynamic constraints to ensure collision-free navigation, even when task endpoints are shared. Simulation results with 10 agents and 50 tasks show that our method reduces the makespan by 22.79% compared to a Token Passing-based baseline, validating its effectiveness in constrained environments

Keywords: Robotic Applications, Robot Mechanism and Control
Abstract: This paper proposes a virtual verification system for standardized robot framework technologies, focusing on enhancing the efficiency and reliability of multi-robot cooperation control. The system consists of a user interface, a centralized multi-robot cooperation control module, and distributed robot terminals. It enables end-to-end virtual validation from task input and status monitoring to path planning and traffic control within a digital simulation environment. Real-time synchronization is achieved through a task status table and a multi-robot path planning database, while each robot autonomously executes tasks based on SLAM-based localization. The proposed system facilitates interoperability verification and automated testing of robotic software, proving its practical value in pre-deployment validation stages.

Keywords: Autonomous Vehicle Systems, Sensors and Signal Processing, Robotic Applications
Abstract: Effective anomaly management is essential for ensuring the reliability and performance of multi-robot systems operating in complex indoor environments. Increasing deployment of robots across logistics, manufacturing, and facility management requires capabilities to detect, classify, and respond to abnormal operational conditions. This paper proposes a web-based monitoring and notification framework designed to detect and visualize key anomalies during mobile robot operations, built on the Robotics Middleware Framework (RMF). The proposed system identifies and classifies anomalies by analyzing both robot-specific data and inter-robot behavior dynamics. Detected anomalies are shared in real time via a centralized dashboard, enabling both operators and system components to perceive and respond to issues collaboratively. The system consists of anomaly definition rules, data-driven detection logic, and a visualization interface implemented on the web. Validation in an RMF-based simulation environment demonstrates improved visibility into abnormal conditions and enhanced responsiveness in robot operations.

Keywords: Robot Vision, Artificial Intelligence Systems, Robotic Applications
Abstract: With the advent of high-performing vision foundation models, many computer vision tasks now achieve both strong performance and impressive zero-shot generalizability. Visual Place Recognition (VPR) is no exception, demonstrating high accuracy and cross-domain robustness across diverse datasets. However, existing VPR approaches primarily focus on exploiting foundation models to obtain large-dimensional global descriptors, often overlooking a key practical requirement: a lightweight memory footprint and low computational cost for deployment on resource-constrained platforms such as mobile robots and autonomous vehicles. These challenges can be addressed through network quantization or descriptor dimensionality reduction. In this work, we focus on the latter and propose a framework, textbf{MatVPR}, which leverages Matryoshka representation learning to enable a single trained model to generate dimension-adaptive global descriptors optimized for various dimensional configurations. Unlike conventional dimension reduction methods that tailor descriptors to a fixed dimensionality, our approach maintains high retrieval performance across multiple target dimensions with a single model. We demonstrate that our technique achieves comparable performance to existing methods on various VPR benchmarks, highlighting its flexibility and practical applicability for real-world deployment on resource-constrained platforms.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Robotic Applications
Abstract: High adaptability is crucial for swarm robotics to operate effectively in dynamic and constrained environments. However, previous swarm models commonly rely on averaging-alignment interactions among robots, which hinders the emergence of adaptive collective motion when responding to environmental changes. In this work, we propose a visual context-based selective interaction mechanism to endow collective adaptability in swarm robotics. Using a semi-physics simulator, we show that the proposed swarm model exhibits high adaptability in navigating through a narrow passage, and also outperforms the Vicsek model in terms of time efficiency and straightness of the crossing path. Finally, we adopt the proposed swarm model in a group of 50 real robots, further demonstrating the advantages of visual context-based selective interactions in real-world tasks.

Keywords: Navigation, Guidance and Control, Robotic Applications, Control Theory and Applications
Abstract: This paper presents a waypoint-scheduled path planning method for multi-robot navigation through narrow passages while maintaining formation and ensuring safe inter-robot distances. The proposed approach uses map-based self-localization and coordinated planning on an expanded occupancy grid map, allowing the robot formation to follow a predefined trajectory and arrival schedule. When the formation cannot pass through constrained regions, a fallback mechanism is triggered to generate a single-robot path. Each waypoint is then evaluated to classify whether the full formation can traverse that segment. The resulting classification enables adaptive scheduling, where robots sequentially navigate narrow areas by adjusting their arrival times at critical entry and exit points. The planning process includes B-spline smoothing and uniform waypoint resampling to ensure smooth motion and synchronized timing. Experimental results in a structured environment with two mobile robots demonstrate the system’s effectiveness in maintaining formation, adjusting behavior in tight spaces, and recovering formation post-traversal. The proposed method enhances multi-robot coordination in environments where narrow passages pose spatial constraints and collision risks.

Keywords: Artificial Intelligence Systems, Biomedical Instruments and Systems, Autonomous Vehicle Systems
Abstract: This paper develops a unified zero–sum H∞ differential-game framework for robust formation control of quadrotor UAVs in a single-leader, multi-follower scenario. We consider one leader executing a take-off → figure − 8 trajectory and three followers maintaining fixed ±5 m NSE/W offsets while tracking the leader. By augmenting each follower’s translational error dynamics with a single integrator state, we enforce zero steady-state height error under constant gravity bias. A static output-feedback gain is synthesized via a generalized algebraic Riccati equation, with only a simple graph-scaling factor distinguishing global and local formulations. The outer-loop H∞ “optimal” acceleration commands are combined with a quaternion-backstepping inner loop to guarantee thrust-vector alignment and torque convergence. 3-D simulations confirm that followers maintain the prescribed formation, reject worst-case disturbances, and achieve zero height bias. The proposed scheme offers a scalable, real-time-capable solution for single-leader multi-UAV formation tasks.

Keywords: Robot Vision, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: This paper presents the Gaussian splatting database (GSDB), a database construction pipeline for Gaussian splatting map-based visual localization. GSDB consists of three stages: (1) View filtering using pixel-wise gradient variance and frustum constraints to remove visually uninformative views; (2) Covisibility graph construction by analyzing spatial overlaps between frustums of among viewpoints; and (3) Viewpoint ranking using perceptual quality (BRISQUE) and structural reliability (Edge-Aware Uncertainty Concentration, EAUC) to select representative views from each covisibility group. This process ensures that only the most informative and stable views are retained in the final database. The proposed GSDB framework is compatible with Gaussian Splatting-based maps and does not require additional geometry or reference supervision. It enables significant database compression while maintaining localization accuracy. Experimental results on the Replica dataset show that GSDB effectively reduces database redundancy, improves retrieval efficiency, and results in up to about 10% faster localization process than conventional baselines, while compressing the database size by up to 95%.

Keywords: Artificial Intelligence Systems, Process Control Systems
Abstract: Reinforcement learning (RL) agents trained through simulations often exhibit degraded performance when directly implemented in real-world processes due to simulation inaccuracies. Sim-to-Real approaches have been developed to address this problem and enable successful real-world adaptation. In this study, we used a Sim-to-Real method to enhance the generalization performance across the Sim-to-Real gap of an RL-based Energy Management System (EMS) framework for the operation of photovoltaic generation, battery storage, and anion-exchange membrane (AEM) electrolyzer systems. An RL framework may fail under real-world conditions where simulation inaccuracies exist, resulting in significant performance degradation. To address this, we introduce domain randomization, where hydrogen crossover parameters are randomized during simulation, enabling the RL agent to learn a robust, generalizable policy under model plant mismatches. Experimental results showed that domain randomization improved initial performance by up to 160% across environments with varied parameter deviations, ensuring stable initial operation despite discrepancies between simulation models and real-world processes. This approach is anticipated to ensure robust startup performance and rapid learning despite modeling inaccuracies during the practical deployment of water electrolyzer systems.

Keywords: Control Theory and Applications, Sensors and Signal Processing, Robotic Applications
Abstract: This paper address the problem of linear-quadratic-Gaussian control (LQG) in systems where the state is observed through nonlinear quadratic measurements. Existing methods rely on linearization and approximation, potentially leading to suboptimal performance. To overcome this limitation, we first establish a connection between the LQG cost structure and the second-order moment of the state, and propose a nonlinear LQG framework that leverages the Quadratic Kalman Filter (QKF). The QKF directly estimates the second-order terms, enabling more precise cost evaluation and control design in systems with nonlinear quadratic measurement models. Our numerical simulations demonstrate improved estimation accuracy and control performance over Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) baselines, validating the effectiveness of our approach for bridging estimation and control in nonlinear quadratic measurement models.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Robot Mechanism and Control
Abstract: This paper presents a heterogeneous marsupial robotic system for object retrieval in hazardous Chemical, Biological, Radiological, Nuclear, and Explosive (CBRNE) environments. The system combines a ground vehicle, robotic manipulator, winch, and tethered UAVs that perch to act as mobile pulleys, enabling flexible aerial–ground cooperation. To support robust perching maneuvers, we develop a full-state trajectory generation framework formulated as a nonlinear programming problem with a simplified quadrotor model. The framework enforces dynamic feasibility by explicitly accounting for position, orientation, velocity, and acceleration constraints. For execution, we implement a modified model predictive controller that computes thrust and body rates for accurate tracking. We validate the approach in simulation with a 9, kg quadrotor across multiple scenarios. Results show successful trajectory generation and tracking for surface inclinations from 0^circ to 90^circ, highlighting the potential of the proposed system to enable reliable aerial–ground robotic coordination in complex and hazardous environments.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Control Theory and Applications
Abstract: This paper presents a ROS2-based distributed task allocation framework for multiple unmanned vehicles performing complex and time-critical missions. The framework incorporates a multi-waypoint task modeling approach and implements two representative distributed algorithms, the Asynchronous Consensus-Based Bundle Algorithm (ACBBA) and the Distributed Auction Algorithm (DAA), as modular ROS2 nodes. A Mission Planning Computer (MPC) provides a graphical interface for configuring mission parameters and distributing designed tasks to vehicles via ROS2 messages. The framework is validated through hardware-in-the-loop experiments across multiple mission scenarios, where the performance of the distributed algorithms is benchmarked against a centralized Hungarian method. The proposed approach offers a practical foundation for advancing autonomous swarm technologies toward real-world applications.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control
Abstract: This paper proposes an evasion strategy based on matrix game theory, assuming the pursuer adopts the Proportional Navigation (PN) guidance law, while the evader employs a corresponding evasive guidance law, with both sides having a finite set of selectable guidance parameters. By pairing the selectable guidance parameters of the evader and pursuer, the expected miss distance is generated. An enumeration strategy is used to comprehensively generate all possible expected miss distances, and the results are populated into a game matrix. A min-max approach is employed to select the corresponding guidance parameter for the evader, which demonstrates strong robustness against different pursuers. Similarly, a max-min approach can select a reasonably effective guidance parameter for the pursuer, which is utilized in simulation verification. Simulation results show that the selected guidance parameters can increase the miss distance, particularly by significantly raising the lower bound of the miss distance, while exhibiting strong robustness.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Robot Mechanism and Control
Abstract: This paper presents a complete control system for the longitudinal dynamics of an F-16 aircraft. The nonlinear equations of motion are linearized around cruise conditions, and the trimmed state, input, and derivative values are calculated to form the state-space model. To improve damping and transient response, a state-feedback Stability Augmentation System is developed. The Stability Augmentation System has been reduced time. To improve tracking performance, a linear quadratic integral-action controller is developed. The controller gain ˆ𝐾 is calculated by solving the Riccati equation with carefully selected weighting matrices. The linear quadratic integrator enables accurate pitch tracking to eliminate the steady-state error and reduce the settling time to approximately 4.27 seconds. The system’s eigenvalues have been confirmed to improve stability and damping characteristics. Simulation results in both the time and frequency domains confirmed the effectiveness of the proposed SAS and LQI designs. Bode analysis shows an increase in the system bandwidth, particularly in autopilot mode, with a closed-loop bandwidth of approximately 1 rad/s. Overall, the proposed control architecture provides a robust solution for stabilising and precisely controlling high-performance aircraft.

Keywords: Autonomous Vehicle Systems, Robot Vision, Sensors and Signal Processing
Abstract: For advanced autonomous driving systems, minimizing sensor blind spots and accurately perceiving the driving environment are critical. This paper proposes a sensor fusion method that combines camera and 3D light detection and ranging (LiDAR) data to estimate the state of objects located in blind spots using a road convex mirror. The camera and LiDAR sensors are calibrated to enable the integration of image and range data. A deep learning–based detector identifies objects reflected in the convex mirror using real-time camera images, and data association is performed to track the object across frames. To improve the accuracy of state estimation for blind spot objects, the error characteristics of the measurement are analyzed and incorporated into a Kalman filter. The proposed method is validated through scenario-based experiments by comparing the results with ground truth data. Experimental results demonstrate that the proposed approach can effectively estimate object states and assess collision risk in blind spot regions using road convex mirror reflections.

Keywords: Autonomous Vehicle Systems, Sensors and Signal Processing, Navigation, Guidance and Control
Abstract: This study introduces a semantic mapping framework for indoor environments, integrating visual and geometric observations to detect and represent structural elements such as doors and elevators. These objects are critical for semantic understanding and navigation in structured spaces such as offices or residential buildings. The system combines object detection and instance segmentation from RGB-D data with planar clustering from LiDAR to generate semantic point cloud and geometric structures. These complementary modalities are aligned using a Point-to-Plane Iterative Closest Point (ICP) algorithm to produce object-level maps that are both semantically meaningful and spatially consistent. The proposed method addresses common challenges in indoor perception, including depth distortion, occlusion, and reflectivity, by leveraging the strengths of each sensor. The resulting semantic–geometric representations are relatively compact and accurate. Experimental validation using a mobile robot platform demonstrates the effectiveness of the framework in real-world indoor scenarios and confirms its practical feasibility.

Keywords: Artificial Intelligence Systems, Robot Mechanism and Control, Robotic Applications
Abstract: This paper presents a fully sinusoidal control framework for snake robot locomotion, where all joints, including both horizontal and vertical ones, are controlled using parameterized sine wave functions. The proposed policy generates amplitude, frequency, and phase values for each joint, enabling smooth wave propagation and adaptive body shaping along the entire robot. To encourage forward motion, stable posture, and smooth transitions, a tailored reward function was designed and used to train the policy through the Proximal Policy Optimization algorithm in a large-scale simulation environment. During training, domain randomization and low-pass filtering techniques were applied to improve robustness in real-world conditions. The learned policy was deployed on a 16-degree-of-freedom snake robot without requiring additional parameter tuning. In real-world experiments, the robot reproduced stable and periodic locomotion that closely followed the commanded sinusoidal trajectories. Although minor differences such as amplitude reduction and phase lag were observed, the overall movement remained coherent and stable. These results demonstrate that the proposed policy can effectively transfer from simulation to real-world deployment, even with minimal sensors and a constrained control structure.

Keywords: Artificial Intelligence Systems, Robot Vision
Abstract: Traditional knapsack spraying poses economic and environmental challenges due to the imprecise application of agrochemicals. Aerial sprayers offer advantages in precision agriculture by enabling access to challenging terrains and providing high-resolution imagery. However, existing uniform spraying practices may result in the overuse or underuse of agrochemicals. To address this issue, variable flow rate spraying can be guided by vegetation indices, whose accurate measurement is critical. This study benchmarked three state-of-the-art semantic segmentation models: U-Net, Feature Pyramid Network (FPN), and DeepLabv3+ for crop segmentation in aerial imagery. Among them, FPN consistently achieved superior performance across accuracy, Dice Score, and IoU. In particular, the FPN (Resized) model at a 0.5 threshold demonstrated the best trade-off between accuracy and computational efficiency, achieving an accuracy of 98.71%, IoU of 88.77%, Dice Score of 93.39%, and a latency of 32.2 ms on CPU-based platforms. Furthermore, the proposed segmentation-aided averaging method improved Green Leaf Index (GLI) estimation by effectively excluding background pixels such as soil and weeds, thereby reflecting true vegetative regions. The findings of this study demonstrate the potential of segmentation-aided averaging method for crop health monitoring operations.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Industrial Applications of Control
Abstract: Analog gauges remain common in industrial and nuclear facilities, yet their manual inspection is labor intensive, time-consuming, and sometimes hazardous. To overcome the limitations, we propose an automated analog gauge reading system for mobile robots that integrates traditional computer vision with deep learning to ensure robust performance, even with limited training data. The system combines ellipse detection and perspective correction for accurate gauge region extraction, and applies a hybrid approach of fixed-box analysis and a DenseNet169-based CNN for precise pointer angle and scale range detection. Experimental results show that the proposed method achieves high accuracy in classifying scale regions and estimating pointer angles, despite a small and diverse dataset. The system demonstrates mean absolute errors below 5 degrees for scale angles and below 1 degree for pointer angles, confirming its reliability for real-world applications. This study highlights the system’s potential as a practical solution for real-time, automated gauge monitoring, supporting safer and more efficient industrial inspections and paving the way for predictive maintenance.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: Fault prognosis and accurate prediction of gearbox remaining useful life (RUL) are challenging due to the lack of mathematical models and external factors like operating conditions and ambient temperature. Data-driven approaches are widely used for fault prognosis and RUL prediction, but extracting informative features from data is time-consuming and challenging. In this study, we propose an effective deep learning model, Inception Convolutional long short-term memory (LSTM) Autoencoder (ICLSTMAE), for automatic extraction of degradation features from run-to-fail data in both the time and frequency domains. The ICLSTMAE model using convolutional LSTM (CLSTM), that combines the advantages of convolutional neural networks (CNNs) and LSTM networks to extract both spatial and temporal features from the data. In the proposed method, the extracted degradation features are from current signals collected from the control motor attached to the gearbox, and they are used as input to a deep LSTM (DLSTM) model for RUL prediction. Experiments on a gearbox failure testing system demonstrated the effectiveness of the proposed method compared to the method developed in previous studies.

Keywords: Artificial Intelligence Systems, Biomedical Instruments and Systems
Abstract: In this study, we propose a medical ultrasound image data augmentation method based on a fine-tuned Stable Diffusion model. Due to limited access to medical data, especially for conditions like Chronic Venous Insufficiency (CVI), data scarcity remains a critical challenge for developing deep learning-based diagnostic systems. To address this, we fine-tuned a pretrained Stable Diffusion model using a small number of labeled patient and non-patient ultrasound images. Two class-specific text prompts were used to condition the model for generating anatomically distinct images. The results show that the model can generate qualitatively realistic ultrasound images, especially for patient cases, where some diagnostic features such as artery-vein pairs were partially reproduced. However, the model struggled to consistently replicate key features, particularly in the non-patient class with limited training samples. This limitation highlights the need for future work to incorporate explicit diagnostic annotations and structural constraints to improve the clinical applicability of generated data. Our approach demonstrates the potential of generative models in medical image synthesis.

Keywords: Human-Robot Interaction, Artificial Intelligence Systems, Robotic Applications
Abstract: We present a modular prompting framework for task-aware natural language understanding in spatially grounded environments. The system transforms free-form user queries into executable instructions by leveraging struc- tured semantic representations of the environment. The first stage classifies the user’s intent and extracts contextual cues using an Intent Understanding module. This information guides an Automated Chain-of-Thought Prompting module, which generates intermediate reasoning steps without relying on manually authored exemplars. Based on the inferred reasoning, a Few-Shot Prompt Refinement module reformulates the query into a structured instruction suitable for down- stream execution. The full pipeline enables a language model to produce goal-aligned, context-aware responses grounded in environmental semantics. Designed for flexibility and reuse, the framework supports a wide range of instruction types through composable reasoning and adaptive prompting strategies. Preliminary results demonstrate promising alignment with task expectations. Ongoing development focuses on deeper integration with spatial planning and highlights the system’s potential for scalable prompt-based interaction in embodied agents.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems
Abstract: Hierarchical reinforcement learning enables learning in long-horizon environments with goal-based tasks. Previous hierarchical structures require Higher Layer (HL) to generate good sub-goals for learning efficiency. However, sparse reward causes subordinate agents to propose only sub-goals that are easy to reach. Therefore, reducing the diversity of paths and limiting the exploration by the HL’s sub-goal. To solve these problems, we propose a structure in which the HL identifies sub-goals that are already sufficiently learned and probabilistically hops between them. The HL generates N sub-goal candidates and estimates the Q-value of each sub-goal through the critic of the Lower Layer (LL). Then, it probabilistically selects the less-learned sub-goal using the softmax function. It simultaneously improves the diversity of paths and sample efficiency and enhances learning opportunities for unreached sub-goals. We demonstrate improved effectiveness of our method in the OpenAI Gym goal-reaching environment.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Robotic Applications
Abstract: Autonomous racing presents challenging continuous control problems requiring precise decision-making and adaptation to complex vehicle dynamics at performance limits. Traditional Reinforcement Learning (RL) methods using Multi Layer Perceptron (MLP) architectures fail to capture the inherent temporal patterns and periodic structures in racing dynamics, leading to suboptimal performance and poor sample efficiency. We propose RAPTOR (Racing Adaptive Periodic Trajectory Optimization with Residuals), integrating a Multi-Branch Implicit Fourier Learning Network (MIL-FAN) with Soft Actor-Critic (SAC). Rather than explicit frequency decomposition, our approach employs multiple independent weight matrices that process the same input through different transformations, implicitly specializing in different temporal patterns through training dynamics. Our method introduces beneficial inductive biases through diverse initialization schemes, enabling effective capture of multi-scale periodic patterns without hand-crafted features. Experimental validation in Assetto Corsa simulator across three Formula 1 circuits demonstrates substantial improvements: RAPTOR achieves lap time gains of 2.229s on Monza, 0.905s on Imola, and 2.060s on Silverstone compared to best baselines, representing 1.3-2.3% performance improvements.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Robot Vision
Abstract: Autonomous driving in off-road environments presents unique challenges that differ significantly from structured urban settings, due to irregular terrain, variable appearance, and the ambiguity of semantic boundaries. While most prior work has focused on urban driving scenarios, autonomous driving in off-road environments remains unexplored, in part due to the lack of comprehensive, multi-sensor datasets. Existing benchmarks primarily rely on monocular cameras or LiDAR, and often fail under challenging conditions such as occlusion, dust, or degraded visibility. We introduce textbf{RoboTerra}, a novel multi-modal dataset for off-road scene understanding and traversability modeling. RoboTerra includes per-point Doppler velocity from 4D radar, dense 3D point clouds from LiDAR, and 360-degree RGB camera from a synchronized multi-camera system, offering comprehensive and complementary sensor observations for off-road environments. These modalities collectively provide diverse cues for terrain understanding, helping reduce blind regions and offering visual and geometric signals for robust perception through vegetation and uneven terrain. By integrating these sensors into a unified dataset, RoboTerra becomes the first publicly available resource of its kind, facilitating research in multi-modal perception, sensor fusion, and autonomous planning in unstructured terrain.

Keywords: Control Devices and Instruments, Process Control Systems, Robot Mechanism and Control
Abstract: Reinforcement learning (RL) has recently been applied to quadruped robot locomotion control; however, its reliance on large datasets and high computational requirements makes real-time implementation challenging. To address these limitations, this paper presents a Gaussian Process-based Model Predictive Control (GP-MPC) framework that performs disturbance compensation using comparatively fewer data samples. The quadruped robot is modeled as a Single Rigid Body (SRB) dynamic system. Leg joints are controlled using Proportional-Integral-Derivative (PID) controllers, while body motion is governed by Model Predictive Control (MPC). Gaussian Process (GP) regression is used to model the residual dynamics between the nominal SRB model and the actual system under external disturbances such as uneven terrain. The GP model is updated online with relatively small amounts of data, enabling the MPC to incorporate these corrections at each control cycle. The proposed method includes GP-based modeling of SRB dynamics, Euler discretization of the system, and formulation of a constrained MPC problem. Simulation experiments are conducted using a quadruped robot traversing stochastic, uneven terrain, and the performance of the GP-MPC controller is compared with that of a baseline MPC controller. The results show that the GP-MPC controller reduces trajectory tracking error and maintains stable locomotion under terrain-induced disturbances, without increasing computational cost.

Keywords: Robotic Applications, Rehabilitation Robot, Exoskeleton Robot
Abstract: Hemiparetic gait following stroke presents substantial challenges for rehabilitation and mobility, often characterized by asymmetric kinematics, spasticity, and reduced functional independence. To address the limitations of conventional exoskeletons—typically designed for spinal cord injury patients with symmetric gait assumptions—we developed RoboWear20, a wearable robotic system tailored to the unique biomechanics of hemiplegic individuals. RoboWear20 features side-specific torque actuation using asymmetric motor configurations (AK80-64 for the paretic limb and AK70-10 for the sound limb), a three-layer adaptive control framework (gravity compensation, PD control, and disturbance observation), and a dual-interface system that allows both patients and therapists to interact with the robot in real time via smartwatch and smartpad, respectively. A preliminary usability study with five chronic stroke patients (FAC level 3) confirmed safe and comfortable operation, with highest ratings in joint alignment and end-of-walk posture recovery. However, user feedback also revealed limitations in static balance, suggesting the need for future improvements in postural control. These findings demonstrate the feasibility and clinical potential of RoboWear20 as a personalized, scalable solution for gait rehabilitation and daily assistance in post-stroke populations.

Keywords: Robot Mechanism and Control, Robotic Applications, Human-Robot Interaction
Abstract: Traditional inspection methods for infrastructure such as bridges, towers, and high-rise buildings often pose safety risks to personnel and require significant time and resources. To address these challenges, this paper presents the development of a 6-DOF caterpillar robot designed for vertical surface navigation primarily intended for infrastructure inspection. The robot design includes two key features to enhance robot’s robustness. First, a three-legged configuration ensures that at least two legs remain in contact with the surface at all times, supporting the robot’s weight while the third leg moves. Second, each leg is equipped with two suction cups to provide redundancy in case one cup fails to maintain a vacuum seal. The robot is equipped with a pneumatically tethered suction system and servo motor-driven joints. The suction system can adhere to a rough concrete surface at 0.4 MPa with holding force of 68 N. In addition, the suction cup has successfully adhere to rough surface using the 2-layer bellow-type suction cup. Based on the numerical data, the robot can support 6 kg payload with a safety factor of 2.0 while keeping lower body weight of 2.4 kg. During climbing tests, the robot achieved stable vertical movement with an average climbing speed of 0.9340 cm/s. The experimental tests and numerical results validated the climbing capability and high payload capacity of the proposed robot.

Keywords: Robotic Applications, Robot Vision, Control Theory and Applications
Abstract: In human–robot interaction (HRI), handheld fiducial markers can simplify perception and lighten computational load. We present a cube-shaped marker that embeds a low-cost inertial measurement unit (IMU) beneath five ArUco patterns and a sensor-fusion pipeline for its high-rate localization. A Kalman filter (KF) blends 50 Hz inertial data with 30 Hz RGB-D measurements, producing drift-free, 6-DoF estimates of the cube’s centroid. The multi-face geometry affords robust detection from wide viewing angles, while the embedded IMU sustains tracking through occlusions and motion blur. Experiments spanning slow (0.4 m/s) and fast (2.0 m/s) trajectories show that the proposed fusion raises update frequency three-fold and lowers RMS position error by up to 10% relative to vision-only baselines, delivering a portable and inexpensive solution for real-time HRI.

Keywords: Robotic Applications, Multimedia Systems
Abstract: This paper presents the design and implementation of an autonomous voice-controlled robot system using ChatGPT and a Raspberry Pi-based robotic car. The aim is to integrate modern AI capabilities to process natural language commands and control the robot's movements. In particular, the robot receives voice inputs and processes them through Google’s Speech-to-Text (STT) API, then utilizes OpenAI’s ChatGPT to interpret the commands. With the help of a proposed algorithm for analyzing multi-step command sequences, the robot decomposes complex instructions into actionable steps and performs them accordingly. The robot demonstrates high accuracy and responsiveness in an indoor environment by employing wheel encoders for precise movement and an ultrasonic sensor for obstacle detection and avoidance. Additionally, the robot executes the commands and provides voice feedback to the user using a Text-to-Speech (TTS) API. Several experiments are conducted in an indoor setting, demonstrating the ability to follow complex voice commands with high accuracy and low latency. This research advances the state-of-the-art in voice-controlled robotics by enabling more flexible and intelligent human-robot interactions on low-cost hardware.

Keywords: Robotic Applications, Robot Vision, Robot Mechanism and Control
Abstract: This paper proposes a vision-based robotic system for high-torque bolt fastening in unstructured industrial environments. The system combines a mobile dual-arm manipulator, a commercial impact wrench, and dual RGB-D cameras to perform bolt detection, 3D pose estimation, and compliant insertion without the use of force/torque sensors. Bolt positions are detected using a YOLOv12n Turbo model, and surface normals are estimated via SVD applied to clustered 3D points. A SAM2-based segmentation approach enables fine alignment through image-based distance estimation. Fastening is achieved via a self-aligning spiral insertion strategy that compensates for positional errors. Experimental results on M22 bolts demonstrate an 83% success rate and an average cycle time of 12 seconds. The system offers a low-cost, robust solution suitable for real-world field applications.

Keywords: Navigation, Guidance and Control, Robotic Applications, Control Theory and Applications
Abstract: Safe navigation in cluttered and unknown environments remains a significant challenge for differential-drive unmanned ground vehicles (UGVs). Classical reactive planners exhibit low computational cost but often fail in narrow passages, while end‐to‐end learning approaches lack formal safety guarantees. This paper presents a hierarchical navigation framework for autonomous ground robots with limited onboard sensing, operating in densely cluttered environments. The proposed system integrates a global path planner with a local risk-aware reference trajectory generator and a nonlinear model predictive controller (NMPC) to ensure kinodynamic feasibility, real-time reactivity, and collision avoidance. A spatio-temporal confidence metric is used to modulate local planning behavior based on environmental complexity. Convex free-space decomposition around the reference path enables safe corridor construction, ensuring that the robot’s entire footprint remains within obstacle-free regions. The framework was extensively evaluated in simulation on the BARN Challenge benchmark, where it achieved higher success rates and superior navigation scores compared to both classical and learning-based baselines. The results demonstrate the proposed method's effectiveness in constrained and complex environments with limited perception.

Keywords: Navigation, Guidance and Control, Civil and Urban Control Systems, Autonomous Vehicle Systems
Abstract: This research conducts an empirical verification of a path planning algorithm for managing multiple unmanned aerial vehicles (UAVs) in a structured urban airspace, responding to the rising demand for efficient and safe drone delivery systems. The algorithm, incorporating inner loop optimization for route selection and outer loop sequencing for coordination, is implemented to control Crazyflie 2.1 quadcopters in real-flight experiments. The evaluation centers on a multiple sources to multiple sinks (M-to-M) delivery scenario, assessing trajectory adherence, tracking precision, conflict avoidance, and operational efficiency. Experimental results highlight precise path following, with both horizontal and vertical tracking errors remaining within reasonable levels, alongside conflict-free operations supported by temporal gaps exceeding a 2-second separation threshold. The algorithm exhibits optimized path planning and robust performance in high-density very low-level (VLL) environments, confirming its practical applicability. These findings provide substantial insights for urban airspace management, contributing to the development of unmanned aircraft system traffic management (UTM) by offering a reliable framework to enhance safety in complex aerial operations.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: Robust localization of autonomous robots remains a critical challenge in environments where LiDAR performance is degraded, such as nuclear facilities or underground tunnels. Conventional Strapdown Inertial Navigation Systems (SINS) accumulate errors over time due to sensor bias and noise, a problem that is further exacerbated in environments with sparse or ambiguous LiDAR observations. In this study, we propose a GRU-based learning motion prediction model to replace the SINS-based inertial state prediction within a UKF-based LiDAR-Inertial SLAM framework. By directly learning the robot’s motion dynamics and time-varying sensor biases from raw IMU data, the proposed model mitigates the cumulative errors that arise when integrating discrete IMU measurements and provides more accurate motion priors. The effectiveness of the proposed approach was validated in both simulation and real-world environments. In simulation experiments, UKF+GRU achieved equal or lower Absolute Trajectory Error (ATE) compared to conventional SLAM methods, while also significantly reducing computational cost. In real-world environments, the proposed model provided stable pose estimates, although further validation in more diverse environments and against high-precision ground-truth measurements is necessary.

Keywords: Navigation, Guidance and Control, Control Theory and Applications, Industrial Applications of Control
Abstract: In the attitude control of an asymmetric spacecraft, oscillation inevitably occurs, imposing significant loads on the two actuators. This leads to a decrease in actuator lifespan as well as fuel or battery reserves. In this paper, an adaptive control gain is proposed to reduce oscillation and energy consumption. To suppress the angular velocity increase of the uncontrollable axis, a small gain is initially used. As the spacecraft nears the desired attitude, the control gain is increased to accelerate the convergence rate. However, the adaptive gain is affected by the initial attitude. Therefore, the proposed method is divided into the direct method and the indirect method, based on the initial attitude. The direct method guides the spacecraft directly to the target attitude. In contrast, the indirect method guides the spacecraft to an intermediate attitude that is easier to reach from the initial attitude, and subsequently converges to the target attitude. The stability of the proposed method is proven using homogeneous vector field-based stability analysis. The proposed method shows superior energy efficiency compared to the conventional method.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Robot Mechanism and Control
Abstract: Mobile robots equipped with differential drive modules on both the front and rear—referred to in this study as Dual Differential Drive Mobile Robots (D3MRs)—pose unique control challenges due to a structural mismatch between modeling and actuation. While trajectory tracking is often modeled using the kinematic bicycle model, low-level control on the D3MR must be executed via differential drives on both the front and rear modules. This mismatch makes it difficult to apply conventional controllers such as Stanley and Pure Pursuit. To resolve this issue, this study introduces a velocity mapping method that converts acceleration and steering commands into individual wheel velocities through kinematic decomposition. This mapping enables conventional controllers to operate effectively on the D3MR platform. Furthermore, a nonlinear model predictive control (NMPC) method is designed to improve path tracking performance. The proposed framework, which combines velocity mapping method and NMPC, is implemented and tested in the Webots simulator across geometric benchmark paths. Comparative experiments show that the NMPC with velocity mapping method outperforms conventional controllers, maintaining tighter adherence to reference paths. These results demonstrate the importance of bridging the modeling-control gap for reliable and precise autonomous driving in D3MR platform.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Robotic Applications
Abstract: This study introduces an autonomous navigation control strategy for mecanum wheel omnidirectional mobile robots that integrates real-time dynamic obstacle avoidance with precise path following, addressing the challenges of autonomous navigation in complex environments. The system, built on the Jetson TX1 and utilizing specific motors, controllers, a LiDAR, and an RGBD camera, features a reactive obstacle avoidance algorithm based on segmented laser scan data of avoidance and path-following vectors. Enhanced by a Kalman filter for robust localization and optimized PID steering control, experimental results demonstrate the system's ability to balance path deviation and obstacle safety, outperforming traditional methods and offering a valuable solution for service robots, intelligent warehousing, and industrial automation.

Keywords: Control Theory and Applications, Industrial Applications of Control, Sensors and Signal Processing
Abstract: This paper presents a systematic co-design framework for an output feedback event-triggered tracking controller for nonlinear systems. The nonlinear dynamics are modeled using a polytopic representation with state-dependent weighting, enabling the application of linear matrix inequality (LMI) techniques. An observer estimates the system states, while an event-triggering mechanism updates control signals only when necessary to reduce communication and computation. A unified LMI condition is proposed to jointly design the controller gain, observer gain, and triggering matrix, ensuring asymptotic tracking and closed-loop stability. A minimum inter-event time is also derived to avoid Zeno behavior. Simulation results confirm that the proposed method achieves stable tracking with fewer transmissions, making it well-suited for networked or embedded systems with limited resources.

Keywords: Control Theory and Applications, Industrial Applications of Control
Abstract: Achieving improved durability and performance in vehicle systems requires effective suppression of driveline vibrations. This study presents an automated tuning approach for an active driveline vibration control system, with particular attention to the complex nonlinear behavior induced by mechanical backlash clearance. To enhance performance under nonlinear conditions, the control framework incorporates an adaptive mode-switching mechanism, driven by a jerk threshold-based estimation, which enables seamless transitions between operational modes. Importantly, the parameters governing this switching logic are optimized in an automated fashion through the use of the whale optimization algorithm (WOA), eliminating the need for manual calibration. Extensive simulation-based evaluations confirm that the proposed controller achieves superior vibration mitigation compared to conventional approaches.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Process Control Systems
Abstract: Descriptor systems are expressed as differential-algebraic equations, which allow them to represent algebraic constraints inherent in the systems such as robots or linkage mechanisms with physical constraints. In addition to exponential modes associated with finite poles, descriptor systems may also have static modes and impulse modes corresponding to infinite poles, that requires the design of control systems tailored to the specific modes present in the system. This paper proposes a numerical computation method for identifying the modes of descriptor systems. Conventional algorithms depend on QZ decomposition, which is susceptible to numerical errors. In contrast, the proposed algorithm employs orthogonal similarity transformations, making it more robust against such numerical inaccuracies. The effectiveness of the proposed algorithm is validated through numerical examples.

Keywords: Control Theory and Applications
Abstract: This work considers the stability problem of nonlinear time-varying cascade systems. Under general assumptions, we propose a new stability criterion for nonlinear time-varying cascade systems by relaxing the negative definiteness requirement on Lyapunov function derivatives.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Robotic Applications
Abstract: This paper proposes an enhanced least-squares (LS) approach integrated with a composite adaptive update law for environmental parameter identification. The proposed method accelerates the convergence of parameter estimation while mitigating the windup effect that typically arises under weak persistence of excitation (PE). By preventing the adaptive gain from oscillating or growing without bound, the approach improves numerical stability and robustness. A Lyapunov-based stability analysis is provided to guarantee the convergence of parameter estimation errors. The effectiveness of the proposed approach is demonstrated through comprehensive simulations of environmental parameter estimation and robot dynamic uncertainties, which confirm its superior convergence speed and stability.

Keywords: Sensors and Signal Processing, Biomedical Instruments and Systems
Abstract: Respiratory rate (RR) is a vital clinical parameter that can indicate a patient's health condition. Existing rPPG methods often suffer from motion artifacts, uneven illumination, and reliance on a single region of interest (ROI), limiting their robustness and accuracy in real-world scenarios. This study presents a novel multi-ROI ensemble method that addresses these limitations using advanced skin segmentation followed by dual-channel signal extraction and adaptive signal fusion. In this framework, we incorporated wavelet-based filtering and a signal quality-aware weighted ensembling to enhance robustness against environmental interference. For model performance analysis, we used a custom dataset focusing on diverse demographic variation. Our proposed method achieved a mean absolute error (MAE) of 1.5 breaths per minute (bpm) and root mean square error (RMSE) of 2.37 bpm, demonstrating significant improvement over existing conventional approaches for noncontact RR estimation.

Keywords: Artificial Intelligence Systems, Robotic Applications
Abstract: Existing Coverage Path Planning (CPP) methods often fail to align with the core objective of maximizing defect detection. These conventional methods typically pursue uniform geometric coverage, overlooking the fact that defect probability and criticality vary across a component's surface. This results in inefficient inspections where critical areas may be inadequately inspected. To address this gap, this paper introduces a novel hierarchical, multi-agent framework that translates high-level user intent into an optimized, executable robotic inspection path. The primary contribution is an end-to-end framework that intelligently allocates inspection resources, ensuring that critical regions receive greater attention, thereby enhancing overall reliability and efficiency.

Keywords: Industrial Applications of Control, Robotic Applications
Abstract: As deep learning-based artificial intelligence technology spreads across industries, autonomous robots and drones are spreading across industries. In this study, as the construction of apartment complexes is expanding nationwide rather than just in the metropolitan area, and as a solution to the problem of a shortage of underground parking lot control personnel in the case of new apartment complexes, we conducted research on the development of a parking management robot based on autonomous and image processing technologies to address the problem of a shortage of underground parking lot control personnel. In particular, in this study, first, an autonomous mobile robot body using two wheels and two casters capable of autonomous driving, second, real-time spatial information and 3D spatial information of underground parking lots were acquired using lidar, and third, an SLAM map generation system based on the acquired spatial information was built and an indoor space control device was developed using a Raspberry Pi V2 camera.

Keywords: Artificial Intelligence Systems, Robot Vision, Sensors and Signal Processing
Abstract: This paper proposes a real-time automatic attendance system that combines pose estimation and face recognition to operate without any manual input. The system uses the YOLOv11-Pose model to detect hand-raising gestures, which serve as an explicit signal of attendance intent. Only individuals detected with raised hands are passed to the face recognition module, thereby reducing unnecessary computation. The pose estimation is performed based on the relative position of the wrist and shoulder keypoints, allowing robust detection even in multi-person classroom environments. For identity verification, a FaceNet-based embedding model is fine-tuned using Supervised Contrastive Learning (SupCon) to better reflect East Asian facial characteristics. This approach improves intra-class compactness and inter-class separability in the embedding space. Experimental evaluations confirm that the proposed system achieves high precision in gesture detection and improved accuracy in face recognition, showing practical applicability for automated attendance tracking in real-world educational settings.