Keywords: Navigation, Guidance and Control, Robot Mechanism and Control, Autonomous Vehicle Systems
Abstract: Fully actuated omnicopters offer high maneuverability, but rotor failures can severely distort their flight envelope through coupled force and torque generation. This paper investigates post-fault wrench feasibility and command-preserving actuator reallocation for a tilted-lattice eight-rotor omnicopter, where directional actuator phasing makes the remaining control authority highly dependent on the failed rotor location. Under a single rotor failure, asymmetric forward/reverse thrust limits, hover support, and torque requirements reshape the feasible wrench set. We formulate actuator allocation as a constrained feasibility problem and evaluate the degraded translational envelope under a zero-torque condition, representing a task-oriented mode that preserves attitude or payload direction. The analysis reveals two geometric failure classes: a severe-degradation class whose weakest horizontal translational authority drops to 4.48 N, and a moderate-degradation class retaining 15.66 N. For transit-oriented motion, optimization-based reallocation relaxes the zero-torque constraint within a bounded auxiliary torque margin. The selected auxiliary torque maximizes feasible translation while maintaining the required vertical force for hover. In the most constrained directions, this relaxation expands the limit from 4.48 N to 22.93 N in the severe class and from 15.66 N to 28.70 N in the moderate class, providing a feasibility-level basis for attitude-aware fault-tolerant trajectory planning.

Keywords: Control Theory and Applications
Abstract: This paper presents a data-driven predictive control method for discrete-time linear systems based on input-output measurements. The offline data collection phase considers only input-output trajectories corrupted by bounded noise, while the corresponding noise bound is assumed to be known. A data-consistent set is constructed to represent all system dynamics compatible with the collected noisy data and the prescribed noise bound. An auxiliary input-output representation is introduced to describe the system behavior, and the controller is designed by imposing robust conditions over all admissible system dynamics in the data-consistent set. Sufficient linear matrix inequality conditions are derived to ensure closed-loop stability and to incorporate ellipsoidal input-output constraints within a unified framework. Simulation results demonstrate that the proposed method achieves asymptotic output regulation to a prescribed operating point while satisfying both input and output constraints.

Keywords: Robotic Applications, Artificial Intelligence Systems, Robot Mechanism and Control
Abstract: Long-horizon tasks can be decomposed into phase-specific action generators, which can simplify task execution. However, selecting an appropriate action generator under the current state remains important. A fixed temporal schedule, as a general approach, can achieve competitive performance, but it may execute an inappropriate action generator for a long time. To address this issue, we propose Chain-of-Phases, which stabilizes action generator selection using a latent phase-CoT reasoner and a monotonic phase-state update. We evaluate the proposed method in three FrankaKitchen long-horizon manipulation tasks with intermediate-phase initial states. Experimental results on long-horizon manipulation tasks show that our approach reduces wrong generator execution and suppresses unstable switching under the same action generator bank.

Keywords: Information and Networking, Sensors and Signal Processing
Abstract: This study introduces a novel active noise control (ANC) approach, termed the combined step-size filtered-x affine projection exponential hyperbolic sine algorithm (CSS-FxAPEHSA). To practically handle secondary path effects, the filtered-x architecture is integrated into the robust affine projection algorithm. Moreover, the combined step-size of the algorithm is derived, which can dynamically adjust the mixing parameter via l1-norm error minimization. This strategy effectively resolves the trade-off between convergence rate and steady-state misalignment, while preserving strong resilience against impulsive interferences. Comprehensive simulations verify that the CSS-FxAPEHSA achieves superior convergence rates and enhanced steady-state accuracy compared to conventional techniques.

Keywords: Navigation, Guidance and Control, Robot Vision
Abstract: LiDAR-inertial odometry (LIO) provides accurate state estimation by combining the high-rate motion measurements of an inertial measurement unit (IMU) with the geometric observations of LiDAR. However, conventional LIO methods that use IMU measurements as state-prediction inputs can be severely degraded when the IMU output is saturated, because clipped angular velocity or acceleration directly corrupts the motion prior. Although Point-LIO improves robustness by treating IMU measurements as correction measurements, it does not fully exploit reliable inertial prediction during non-saturated intervals. This paper proposes IMS-LIO, an IMU mode-switching point-wise LIO framework for robust state estimation under IMU saturation. The proposed method switches the role of IMU measurements according to real-time saturation detection. In the nominal mode, IMU measurements are used for state prediction to provide an inertial motion prior. In the saturation mode, angular velocity and acceleration are included in the filter state, and IMU measurements are used for correction while saturated axes are excluded from the update. Experimental evaluations are conducted on the TIGS dataset, which contains aggressive tumbling motions and repeated IMU saturation, to analyze the effectiveness of the proposed mode-switching strategy for robust LiDAR-inertial odometry.

Keywords: Control Devices and Instruments, Control Theory and Applications
Abstract: When validating multiple scenarios or generating high-quality data for reinforcement learning, efficiently handling a large number of structurally identical problems is essential. CPU-based solvers can accurately solve quadratic programming (QP) formulations of model predictive control (MPC) problems, but their computational cost grows linearly with the number of instances. This paper presents B-IPM, a throughput-oriented GPU batch primal-dual interior-point QP solver that simultaneously solves up to 10,000 MPC problems formulated as QPs sharing the same cost and constraint structure. B-IPM formulates multiple cart-pole regulation problems with varying initial conditions and constraint bounds as condensed QPs and solves them using a log-barrier primal-dual interior-point method. Experimental results demonstrate that all 100 instances are stably controlled to the upright equilibrium, even when position constraints are active, and zero falls are observed across all tested batch sizes up to 10,000. Furthermore, a batch-size sweep from 1 to 10,000 reveals that the per-step latency of a six-core CPU cyipopt baseline increases almost linearly, whereas B-IPM scales sublinearly, achieving an approximate 3.3-fold speedup at a batch size of 10,000. These results indicate that B-IPM is well-suited for large-scale, multi-scenario MPC verification and for generating large volumes of trajectory data for reinforcement learning.

Keywords: Robotic Applications, Industrial Applications of Control
Abstract: Lithium-ion battery degradation proceeds through different electrochemical processes depending on temperature and operating conditions, and the components extracted from electrochemical impedance spectroscopy (EIS) by equivalent-circuit-model (ECM) fitting are widely used as diagnostic indicators. However, such component-based degradation signatures are mostly established on a few cells with laboratory-grade EIS, and how far they remain observable in large-scale, realistic (low-cost cycler) EIS has rarely been verified. In this work, we fit an ECM to the room-temperature EIS of a large public dataset of cells aged under a wide range of conditions, and extract the ohmic resistance (R0), the charge-transfer resistance (R2), and the constant-phase-element parameters Q and p. We track these together with capacity-based state of health (SOH) across aging temperature, charge protocol, and state of charge (SOC). A single-ZARC model reproduces the spectra with low error, and R0 and R2 increase while p decreases consistently across almost all cells as the cells age. The growth rate of R2 per unit capacity loss varies with condition: under high-rate charging at low temperature the capacity loss is decoupled from the R2 increase, which we link to lithium plating with loss of lithium inventory, whereas at high temperature Q and p change in a way corresponding to the surface-area and heterogeneity increase of particle cracking. These trends are broadly consistent with previous reports.

Keywords: Rehabilitation Robot, Biomedical Instruments and Systems, Control Devices and Instruments
Abstract: Ankle inversion is important during walking and turning movements; however, excessive inversion can lead to lateral ankle sprain. Quantitative evaluation of structural changes around the calcaneus during inversion is therefore important, but stable ultrasound imaging during motion remains challenging because probe position and contact conditions can change. In this study, we propose an ultrasound-based method for quantitatively evaluating changes in the calcaneal bone surface and surrounding soft tissues under dynamic inversion conditions. A dedicated fixation holder was developed to stabilize the ultrasound probe, and an image processing workflow was constructed to extract the calcaneal region. Ultrasound images were acquired from five healthy adult participants under neutral and inversion conditions. The x3 coefficient of the cubic curve fitted to the calcaneal surface and the low-intensity pixel ratio in the soft tissue region were used as evaluation indices. For both indices, all participants showed changes in the same direction between conditions, and significant differences were observed using one-tailed paired t-tests. However, the 95% confidence intervals for the effect sizes were wide and included zero. These results suggest that ultrasound derived indices around the calcaneus may exhibit condition dependent changes under the dynamic conditions used in this study.

Keywords: Rehabilitation Robot, Human-Robot Interaction, Robot Vision
Abstract: In close-proximity frontal following scenarios for smart walkers, recognizing the user's walking intention and relative position remains challenging due to the cost and complexity of multi-sensor systems and the limited availability of real walking data. To address these challenges, this study proposes a data-driven approach that recognizes the user's walking intention and position primarily using depth information acquired from a low-cost depth sensor and improves data diversity through simulation-based data augmentation. Specifically, the proposed method employs a Transformer-based model for walking intention prediction, an ROI-based depth processing algorithm for user position estimation, and Isaac Sim for simulation data generation. Experimental results show that walking intention recognition using depth information alone achieved an average F1-score of 0.88. The estimated user-walker distance was 0.70 pm 0.049 m (SD), corresponding to a comfortable distance during walking assistance. The results demonstrate that the proposed data-driven method can effectively recognize the user's walking intention and position using depth information, while a 2D PID-based following control simulation verifies the feasibility of applying the proposed method to smart walker control.

Keywords: Rehabilitation Robot, Robot Mechanism and Control, Human-Robot Interaction
Abstract: Upper-limb rehabilitation robots have been widely studied to provide repetitive and task-oriented training for patients with motor impairments. In our previous study, a three-degree-of-freedom (3-DOF) end-effector-type upper-limb rehabilitation robot was developed, and its mechanical design, workspace optimization, and structural validation were presented. This paper extends the previous work by focusing on the overall system integration, control hardware architecture, and control strategies for the developed robot. To clarify the implementation framework, the control hardware configuration and signal flows based on a real-time operating system (RTOS) network are established. Furthermore, a non-switched cascade control architecture is formulated, in which a task-space admittance controller generates a modified reference trajectory for task-space impedance tracking, while a conditional joint-space impedance loop enforces the spatial tilt-angle boundary. The formulated framework serves as an initial structure for future real-time control and interactive rehabilitation experiments.

Keywords: Rehabilitation Robot, Robot Mechanism and Control, Robotic Applications
Abstract: Bed-ridden patients require repeated lower-limb exercise to prevent mobility decline, but access to therapist guided rehabilitation is often limited in bedside environments. This paper presents Pilates-Bot, a cable-driven bedside rehabilitation robot designed to provide guided Pilates-based lower-limb exercise in a supine posture. The system uses a modular mobile frame positioned around a bed, while actuators are separated from the patient and assistive forces are transmitted through lightweight cables to reduce mechanical burden near the user. For real-time control, encoder-based cable length measurements were used to estimate the end effector position with pulley geometry compensation. A virtual spring controller generated position-error-based assistive forces, and quadratic programming allocated the desired force to individual cable tensions under tension constraints. Position tracking experiments with a 4.5 kg load were conducted along 300 mm reciprocating trajectories. The measured trajectories followed the references, with delay-compensated RMS errors of 5.3% and 14.0% of the motion range in the y- and z-axis directions, respectively. A preliminary usability evaluation with five elderly participants showed that Assist-As-Needed(AAN) mode maintained or improved tracking performance while providing higher robot-generated assistance than Minimal-Support mode mode, with a System Usability Scale (SUS) score of 91.67 and low perceived workload. These results support the feasibility of Pilates-Bot as a bedside lower-limb rehabilitation system.

Keywords: Information and Networking, Biomedical Instruments and Systems, Rehabilitation Robot
Abstract: Advanced perception and motion-planning libraries for robotics are usually developed for non-real-time operating systems. However, safety-critical medical robotics applications require deterministic, hard real-time execution. We propose an architecture that separates non-real-time deliberative functions, such as planning, perception, and learning, from real-time industrial safety and motion control. A supervised interface validates commands sent from the deliberative layer to the control layer to ensure safety. We present two open-source interfaces that connect ROS 2, which runs on a standard Linux kernel, to a Beckhoff TwinCAT 3 real-time PLC: (i) a UDP-based interface with application-layer security and (ii) a hardware-isolated EtherCAT master–master bridge. Both preserve safety authority within the real-time domain while enabling integration of advanced computer vision and machine learning algorithms. The architecture is demonstrated in a semi-autonomous orthopedic laser application.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Robotic Applications
Abstract: Estimating traversability in unstructured off-road environments remains challenging, as traversability often cannot be determined from visual appearance or geometric properties alone. However, existing approaches often rely on only one of these cues, limiting their ability to leverage both semantic and geometric information. In this paper, we propose a learning framework that jointly exploits visual and geometric cues for traversability estimation in unstructured off-road environments. To avoid the need for manual annotation, we introduce an automatic label generation framework that produces visual traversability and geometric risk labels using a pretrained visual traversability labeler and geometric terrain analysis. We further propose a visual-geometric fusion strategy that combines visual traversability and geometric risk predictions to achieve more reliable traversability estimation. Experimental results across multiple seasons and diverse off-road terrains demonstrate that the proposed method achieves more balanced traversability estimation than visual-only and geometric-only approaches.

Keywords: Control Devices and Instruments, Control Theory and Applications
Abstract: This paper proposes a 13-state direct–quadrature (dq) grid-following (GFL) digital twin for 220kW battery energy storage power-conversion systems (BESS-PCS) built from low-rate supervisory control and data acquisition (SCADA) logs under unknown grid impedance. Only low-rate SCADA observations are accessible while the converter dynamics are much faster. The equivalent line inductance and shunt capacitance are identified from multiple field logs by a robust iteratively reweighted least-squares (IRLS) estimator, while the inner and outer controller gains stay at commissioned nominal values, so the twin is calibrated rather than fully data-identified. The plant, controller, and disturbance are combined into one nonlinear model, and the available logged operating envelope is numerically checked for local exponential stability, with the angle-augmented closed-loop Jacobian remaining Hurwitz across the observed non-saturated strong-grid envelope. In replay, the twin reproduces active power, reactive power, and current with R2≥0.986 on a representative field test. The reconstructed point-of-common-coupling (PCC) voltage ˆ VLL is reported only as an auxiliary consistency metric and shows a 2.12V RMSE. AC-compiled twin runs in a 100Hz live-replay worker, maintaining RTF=0.9998 over a 33s field test. Grid-forming extension, online state estimation, and hardware-in-the-loop tests remain future work.

Keywords: Control Theory and Applications, Robotic Applications, Robot Mechanism and Control
Abstract: Differentiable simulation yields analytic first-order policy gradients that can improve the sample efficiency of policy learning relative to model-free reinforcement learning, as shown on smooth systems such as quadrotors. For legged robots, however, the non-smoothness of contact renders the analytic gradient of a hard-contact step ill-suited to gradient-based policy search: evaluated within a single contact mode, it carries little information about the making and breaking of contact. Recent quadruped studies therefore smooth the backward pass, for example by back-propagating through a reduced single-rigid-body surrogate or through a dedicated differentiable contact model. We instead adopt the analytic contact-impulse gradient under a relaxed complementarity constraint, originally introduced for contact-implicit model predictive control, as the backward pass of a differentiable simulator for analytic policy gradient (APG) learning of quadruped locomotion. The simulator takes a strict hard-contact step in the forward pass and applies the relaxed Signorini gradient in the backward pass. Under this hard-forward/smooth-backward scheme the policy learns to track velocity commands, whereas APG with the non-smooth hard-contact gradient fails and the robot loses balance.

Keywords: Robotic Applications, Sensors and Signal Processing, Navigation, Guidance and Control
Abstract: This paper presents a contact-aided equivariant filter for legged robot state estimation. Many nonlinear state estimators for legged robots still rely on linearization about the current estimate, which introduces linearization errors that degrade performance. To mitigate this issue, we formulate legged robot state estimation as equivariant filtering. The proposed estimator uses IMU propagation and kinematics measurements while incorporating bias states through an expanded tangent group symmetry. This yields state-independent propagation linearization for the navigated states. Simulation results show that the proposed method improves estimation performance compared with non-invariant and invariant baselines.

Keywords: Robotic Applications, Robot Mechanism and Control, Navigation, Guidance and Control
Abstract: This paper presents a residual learning framework for model predictive control (MPC) in humanoid non-prehensile loco-manipulation. The target task is heavy-object pushing, where a humanoid robot must coordinate locomotion, balance, body posture, foot placement, and hand-object interaction forces. A nominal MPC based on simplified robot-object dynamics provides a structured and physically interpretable controller, but its performance can be limited by heuristic references and modeling assumptions on object motion and friction. To improve task expressiveness without replacing the MPC backbone, a residual planner is trained to modify high-level references, including footstep positions, body angle references, and body-height references. Simulation results on a Unitree G1 humanoid show that the residual-augmented controller learns physically meaningful pushing strategies: the feet are shifted behind the body, the torso leans toward the object, and the body height is modulated according to the gait phase. These results suggest that residual learning can enhance MPC-based humanoid loco-manipulation while preserving the structure of model-based control.

Keywords: Autonomous Vehicle Systems, Robot Vision, Sensors and Signal Processing
Abstract: Visual-inertial odometry can provide accurate state estimation for mobile robots and autonomous systems, but its performance often degrades in structured indoor and construction-like environments. In such scenes, weak texture, repetitive patterns, illumination changes, motion blur, and rapid motion can make visual correspondences unreliable and introduce unstable residuals into the backend optimization. This paper presents an enhanced OKVIS2-based visual-inertial odometry pipeline that improves both frontend data association and residual weighting. In the frontend, CLAHE preprocessing, SuperPoint feature extraction, and LightGlue matching are integrated to obtain more reliable feature correspondences under challenging visual conditions. To preserve compatibility with the original OKVIS2 pipeline, BRISK descriptors are computed at the learned keypoint locations, and the original geometric verification path is retained. In the backend, visual residuals are adaptively weighted using online reprojection residual statistics, while IMU residuals are adjusted using RMS jerk to reduce the influence of aggressive-motion intervals. The proposed method is evaluated on multi-camera visual-inertial sequences from the HILTI 2022 dataset and compared with the original OKVIS2 baseline. Experimental results show that the proposed pipeline improves trajectory estimation robustness in challenging structured environments.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems, Autonomous Vehicle Systems
Abstract: Blind corners create a coupled navigation problem because the same wall that blocks direct goal motion can also provide cover from an observer. This study considers randomized L-corner scenes with a single observer and proposes an event-triggered framework that switches between ADVANCE and RETREAT. A common local-target generator rejects collision-prone segments and ranks finite-distance targets using goal progress, clearance, and risk derived from the latest observer estimate. Under onboard detection limited to a 22 m range and a 90◦ field of view, the implementation using SAC for ADVANCE and PPO for RETREAT completed 94 of 100 held-out scenes with no collisions or RETREAT failures. Under the same switching rule, target generator, RETREAT controller, and collision guard, DWA and MPPI achieved comparable completion rates, resulting in a 93–96% range across the three ADVANCE controllers. These results indicate that the proposed switching and target-generation logic can be paired with different low-level ADVANCE controllers.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: This paper proposes a Neural Signed Distance Field (SDF)-based Augmented Lagrangian Differential Dynamic Programming (AL-DDP) method for local path planning and obstacle avoidance of ground vehicles in rough terrain. Since environmental information obtained from onboard sensors is typically represented as a discrete map, conventional gradient-based methods are difficult to apply directly. The proposed method addresses this by constructing a Neural SDF from onboard sensor data and incorporating it as an inequality constraint within the AL-DDP framework. Simulation results demonstrate that the proposed method generates obstacle-avoiding trajectories in the evaluated scenarios, and comparisons with an APF-based DDP baseline confirm a distinct safety--tracking trade-off between cost-based and constraint-based avoidance formulations.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: This paper proposes an optimization-based gain-synthesis framework for a variable gain guidance law with prescribed impact-angle constraints in planar surface-to-surface engagements against stationary and nonmaneuvering moving targets. A time-varying navigation gain, formulated as a quadratic function of the line-of-sight (LOS) angle, is adopted to satisfy impact-angle requirements while respecting seeker field-of-view (FOV) constraints and the lateral-acceleration boundedness property. Closed-form expressions are derived for the impact angle and maximum look angle, enabling guidance gains to be selected by minimizing the peak look angle subject to admissibility conditions. Numerical simulations demonstrate that the proposed guidance law achieves the desired impact angles while maintaining FOV constraints. Comparative results show improved performance in terms of interception time relative to existing guidance laws.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems
Abstract: Efficient path-following is crucial in most of the applications of autonomous aerial vehicles (UAVs). Among various guidance strategies presented in literature, the look-ahead distance (L_1)-based nonlinear guidance has received significant attention due to its ease in implementation and ability to maintain a low cross-track error while following simpler reference paths and generating bounded lateral acceleration commands. However, the constant value of L_1 becomes problematic when the UAV is far away from the reference path and also produces higher cross-track error while following complex reference paths having high variation in radius of curvature. To address these challenges, the notion of look-ahead distance is leveraged in an original way to develop a two-phase guidance strategy. Initially, when the UAV is far from the reference path, an optimized L_1 selection strategy is developed to guide the UAV towards the vicinity of the start point of the reference path, while maintaining minimal lateral acceleration command. Once the vehicle reaches a close neighborhood of the reference path, a novel notion of corrector point is incorporated in the constant L_1-based guidance scheme to generate the guidance command that effectively reduces the root mean square of the cross-track error and lateral acceleration requirement thereafter. Simulation results validate satisfactory performance of this proposed corrector point and look-ahead point pair-based guidance strategy, along with the developed mid-course guidance scheme. Also, its superiority over the conventional constant L_1 guidance scheme is established by simulation studies over different initial condition scenarios.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Control Theory and Applications
Abstract: This paper presents an asynchronous guidance and control framework for the HTTP-4 vertical-takeoff-and-vertical-landing hybrid rocket. Extended Non-dominated Sorting Genetic Algorithm III (NSGA-III) guidance balances propellant consumption, maneuver time, and horizontal landing error through feasibility-first handling and warm-started replanning. A 20-Hz nonlinear model predictive controller (NMPC) incorporates six-degree-of-freedom dynamics, actuator constraints, identified thrust dynamics, and a four-step command queue for the nominal 0.20-s transport delay. The hot-fire mean delay is 0.1828~s; mean AM437x execution times are 49.37~ms for NMPC and 3.868~s for online NSGA-III. NMPC reduces the representative-mission three-dimensional tracking root-mean-square error from 9.981 to 4.428~m. Across 500 randomized closed-loop landings, the mean and maximum horizontal errors are 1.814 and 4.365~m.

Keywords: Human-Robot Interaction, Robotic Applications
Abstract: Tabletop robotic tasks often require visual understanding of changing object locations, spatial relationships, and execution states, while cloud-based large models may introduce API cost, network dependence, and privacy concerns. This paper presents LocalVLM-TPR, a locally deployed vision-language model framework for robotic task planning and re-planning. The system combines eye-to-hand and eye-in-hand vision to support scene perception, task planning, execution validation, and failure recovery. A logic audit is used to check plan feasibility, while pre-action, post-action, and final-goal validation form a closed-loop execution process. When a local failure occurs, plug-in re-planning generates a corrective subtask. Experiments on three tabletop tasks show that LocalVLM-TPR achieves a 93.33% average success rate, close to the cloud-based ReplanVLM baseline with 96.67%, while avoiding cloud-based inference.

Keywords: Artificial Intelligence Systems, Robot Mechanism and Control, Robotic Applications
Abstract: This paper introduces a disturbance-aware reinforcement learning (RL)-based adaptive PID control framework for robust tracking control of manipulators operating on dynamic naval platforms. The control architecture integrates a feedback-linearizing PID controller with a nonlinear disturbance observer, yielding safety-constrained PID gain bounds. An RL agent utilizes disturbance estimates and runs asynchronously to adaptively tune the PID gains within the gain bounds for safety, robustness, and optimal performance. To improve training efficiency, an automated expert policy generation pipeline is designed based on large language models to initialize the RL policy through behavior cloning. The agent is trained in simulation under varying sea-state conditions and domain randomization, and validated on a ROBOTIS OpenMANIPULATOR-X robot mounted on a Stewart platform. Our results demonstrate improved training efficiency over standard RL baselines and non-expert behavior cloning methods, with superior performance and robustness compared to conventional PID tuning and prior adaptive RL-PID approaches.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Robotic Applications
Abstract: Quadrotor research benefits from accurate system models and simulation environments, facilitating novel controller design and evaluation. This is especially important for quadrotor applications which diverge from the conventional near-hover regime, such as drone racing and acrobatic flight. To assist with such research, we develop QuaSI: an integrated Python-based quadrotor identification and simulation framework. QuaSI allows users to identify customizable models of their own quadrotor platforms and deploy these directly in an accompanying (or their own) simulation environment. To accommodate diverse user needs and quadrotor designs, QuaSI allows one to define their own model structure and basis variables, in addition to the standard first principle-based terms. As a sanity check, QuaSI also reports on the expected validity of the identified models. These insights can help users find regions of the flight envelope where their models produce reliable predictions alongside those where it struggles and may require additional data. QuaSI therefore assists researchers in quickly deploying and improving upon their (custom) quadrotor and subsequent controller designs.

Keywords: Human-Robot Interaction, Robotic Applications, Robot Mechanism and Control
Abstract: Vision-Language-Action (VLA) models generate robot actions in fixed-length inference chunks, amortizing neural evaluation cost but introducing two practical problems: the robot idles during inference, and a freshly predicted chunk rarely starts from where the robot actually is, producing a jarring velocity jump at every chunk boundary. A third issue appears on startup, where the first predicted action commands a pose far from the robot’s current configuration, causing an initialization lurch. This paper describes AsyncFlow, an execution-layer system addressing all three problems without any changes to the VLA itself. AsyncFlow stays permanently ahead of the next inference: while the robot executes the current chunk, the next chunk runs in the background. At chunk boundaries, AsyncFlow blends consecutive chunks smoothly over a 6-step overlap window, substantially attenuating the velocity jump at each handoff (smoothstep blending achieves C1 continuity). It publishes multiple future waypoints per control tick rather than a single action, giving the lowlevel controller a 400 ms lookahead horizon (4 waypoints × 100 ms at 10 Hz evaluation rate). On startup, it gently eases from the robot’s observed pose into the model trajectory over 16 steps, removing the initialization lurch. On a ROBOTIS AI Worker running GR00T 1.6N on a paint brush pick-and-place task, AsyncFlow achieves 99.2% prefetch hit rate, 12× lower velocity jump, 9× lower RMS jerk, and 98% task success – a +33 percentage-point gain over the 65% synchronous baseline (+50.8% relative) – without any modification to the underlying VLA. On this task and model, the remaining 2% failure rate is attributed to gripper hardware limitations; broader evaluation across tasks and models would be needed to confirm generalizability.

Keywords: Industrial Applications of Control, Control Devices and Instruments, Artificial Intelligence Systems
Abstract: This paper presents an integrated modeling, optimization, and control framework for the energy management of a hybrid marine propulsion system comprising a two-stroke dual-fuel main engine, a shaft-mounted Power-Take-Off/Power-Take-In (PTO/PTI) generator, a battery energy storage system (BESS), and three auxiliary generator sets. An Artificial Neural Network (ANN) model trained by a collaborating institution is used to represent the nonlinear fuel consumption map of the main engine; an equivalent-circuit battery model and piecewise-linear genset maps complete the plant description. The ANN is embedded into a Mixed-Integer Linear Programming (MILP) solver via a convex piecewise-linear (PWL) approximation, yielding a computationally tractable Model Predictive Control (MPC) scheme, with Dynamic Programming (DP) as the offline global benchmark. Results on a real-world 24-hour voyage show that MILP achieves 5.49% cost reduction relative to the conventional baseline and 1.87% over the existing rule-based max-PTO strategy. A key finding is that recurring MILP control patterns, encoded as five causal rules, yield a real-time controller that outperforms the extended MPC with a 60-minute horizon by 0.5% — a direct consequence of the computational budget constraining the online optimizer to a short prediction window, preventing it from anticipating the main engine shutdown and executing the battery pre-charge strategy identified by the full-horizon solution.

Keywords: Robot Vision, Sensors and Signal Processing, Robotic Applications
Abstract: Event-based cameras asynchronously detect brightness changes at each pixel and are promising for high-speed perception and low-latency robotic control. However, collecting real event data remains challenging due to sensor availability, experimental reproducibility, and ground-truth acquisition. In this paper, we propose a fast event-based camera imulatorbased on same-pixel refiring interval distributions estimated from real event data. The proposed method generates event candidates from inter-frame log-intensity differences, and then selects events and interpolates their timestamps using a probabilistic refiring model. This enables the simulator to reproduce sensor-like temporal firing structures without explicitly simulating detailed pixel circuit responses. Experiments using DAVIS346 event data show that the proposed method achieved the smallest Wasserstein distance of 0.16 among the evaluated simulators while maintaining a processing speed of 631 fps. The proposed method also produced spatial event structures closer to those of real sensors than existing methods. These results suggest that the proposed method efficiently generates event streams with temporal and spatial characteristics close to those of real sensors.

Keywords: Sensors and Signal Processing, Navigation, Guidance and Control, Robotic Applications
Abstract: A three-axis magnetometer supplies the heading that inertial sensing cannot bound, but only once calibration removes the hard- and soft-iron, scale-factor, non-orthogonality and mounting-misalignment errors that displace the reported field direction. Calibration from field magnitude alone is autonomous yet structurally blind to the mounting rotation, which is precisely the quantity heading reads; full-attitude calibration recovers that rotation at the cost of presupposing the heading it exists to establish. This paper shows that gravity-referenced calibration escapes both. The magnitude and inclination constraints identify the complete twelve-parameter model wherever the platform’s tilt varies, and lose a rotation about the local vertical only at constant tilt, so the deficiency is a property of the motion rather than of the constraint set. Gravity Referenced Attitude Constraint Estimation - Magnetometer Calibration(GRACE-MC) measures the tilt diversity of its own record through a single eigenvalue and selects its parameterization from the result. Evaluated against five baselines on real fluxgate data from a moderate-excitation motion profile repeated six times, it holds 1.6◦ mean heading error with no heading reference at any stage and under an injected mounting rotation it attributes the entire injection to the rotation channel, leaving the recovered scale factors, non-orthogonality angles and bias magnitude unmoved, as an estimator supplied with the full attitude does.

Keywords: Biomedical Instruments and Systems
Abstract: Biochemical reaction networks in living organisms can adapt their behavior according to past experiences. Reconstructing such adaptive behavior using artificial molecular systems is an important challenge in DNA computing and synthetic biology. In this study, we propose a refined DNA strand-displacement circuit for Pavlovian classical conditioning with learning and forgetting functions. The circuit changes its response through repeated simultaneous stimulation and subsequently exhibits a conditioned response to a previously neutral stimulus. Numerical simulations reproduce learning, conditioned response, and forgetting processes, demonstrating improved learning performance compared with previous conditioning circuits. Sensitivity analysis under internal concentration perturbations showed that the baseline design was representative of the overall perturbation distribution. Input-strength analysis further showed that stimulus strength modulated both learning performance and memory retention; the learning rate reached a maximum under moderate input enhancement, whereas the forgetting rate continuously decreased with increasing input strength. These results suggest that the proposed circuit provides a feasible basis for implementing molecular learning and memory functions using DNA strand displacement.

Keywords: Biomedical Instruments and Systems
Abstract: Detecting peripheral nerve tissue in histopathology supports perineural-invasion assessment and nerve-sparing surgery, yet nerves are easily confused with stroma and vessels. We present a controlled comparison of four frozen pretrained vision encoders: two domain-specific pathology models (cTransPath and Phikon), a general-purpose vision–language model (CLIP ViT-L/14), and an ImageNet baseline (ResNet-50), plus an early-fusion variant, under an identical linear probing protocol. On 62 H&E images at 10× from 31 patients, patches were sampled by a sliding window and evaluated with patient-level 5-fold cross-validation to prevent leakage. Both domain-specific encoders far outperformed the rest; Phikon was numerically strongest (ROC AUC 0.999 vs. 0.991 for cTransPath), though the two were not separated at five folds. Each beat CLIP (0.948) and ResNet-50 (0.937) in every fold (paired t-test p < 0.05); fusion gave no gain. CKA shows the two pathology encoders learn highly similar representations (CKA 0.91) that are distinct from CLIP (0.58). In full-image detection the two pathology encoders were also the most precise (Phikon: 0.80 sensitivity at 0.58 false positives per image; cTransPath: 0.71 at 0.86), dominating the low-false-positive region of the FROC curve, and Eigen-CAM shows cTransPath concentrates on the nerve while CLIP responds to scattered points. For fine-grained histopathology, the pretraining domain, rather than the specific model, is the dominant factor, and a linear probe on a domain-specific encoder is sufficient and label-efficient.

Keywords: Biomedical Instruments and Systems, Control Theory and Applications, Sensors and Signal Processing
Abstract: Real-time and objective evaluation of virtual reality (VR) sickness requires an index that remains valid when the visual stimulus is not controlled in advance. We propose a model-based evaluation framework built on an eye movement reproduction model that combines the vestibulo-ocular reflex (VOR) and smooth pursuit eye movement (SPEM) and is identified individually for each user. Because the model is driven by head motion and by the motion of an arbitrary gaze target, it does not presuppose any specific visual stimulus. Two detection methods are derived from the model: Method 1 quantifies the overall deviation between the model output (identified before exposure) and the actual gaze measured after exposure, whereas Method 2 quantifies the changes in each model parameter between these two conditions. In an experiment with 29 participants, the post/pre ratio of the plant time constant was larger in severe VR sickness cases than in mild cases. While this was the largest effect among all estimated parameters, it did not survive correction for multiple comparisons at the present sample size. The results demonstrate that our framework localizes the physiological change to a specific subsystem using an index that does not presuppose a specific visual stimulus.

Keywords: Biomedical Instruments and Systems, Information and Networking, Artificial Intelligence Systems
Abstract: Metabolic diseases such as diabetes result from the breakdown of complex multi-cellular and multi-organ networks. Traditional ordinary differential equation models struggle with the computational complexity of simulating the interactions of billions of heterogeneous cells. To address the problem, we propose an approach that integrates a population-level approximation with reinforcement learning to analyze liver energy metabolism. By modeling individual hepatocytes as autonomous agents, we employ a deep deterministic policy gradient algorithm to comprehensively optimize their individual actions---such as local reaction gains and hormone-like responses---to maintain internal Acetyl-CoA homeostasis within a shared circulatory environment. Simulations using the proposed topological model demonstrate that this local optimization successfully regulates macroscopic blood glucose levels. Furthermore, applying the learned policy to a heterogeneous population of 100 cells with varying metabolic parameters showed robust collective homeostasis. The system maintained stability under parameter noise up to 20%, beyond which it exhibited systemic destabilization analogous to pathological decompensation. This population-level framework effectively bridges microscopic cellular behaviors and macroscopic organ functions, offering a new approach for analyzing metabolic network diseases.

Keywords: Biomedical Instruments and Systems, Information and Networking, Sensors and Signal Processing
Abstract: Pancreatic cancer has an extremely poor prognosis due to difficulties in early diagnosis. To achieve a highly sensitive screening method, microRNA (miRNA) biomarkers have attracted significant attention. In this study, we propose a cascade signal amplification model that integrates two DNA computing circuits: Catalytic Hairpin Assembly (CHA) and an Entropy-Driven Circuit (EDC). Numerical simulations confirmed a synergistic amplification effect and an ideal sigmoidal response that mitigates leaky reactions. Furthermore, in vitro experiments were conducted to evaluate the individual behaviors of the constituent circuits. The results validate the feasibility of miRNA detection using the proposed framework while highlighting specific sequence design refinements required for seamless physical integration.

Keywords: Artificial Intelligence Systems
Abstract: Facial emotion recognition (FER) in the wild remains challenging due to pose variation, illumination, and occlusion. While Vision Transformer (ViT) models achieve strong results, they treat all image patches equally, discarding the anatomical prior that emotions arise from localized muscle-group activations. We propose KGAB-ViT (KeypointGuided Attention Bias ViT), which injects face keypoint information as additive Gaussian heatmap biases directly into the ViT attention logits at every encoder layer. A per-layer learned gate grows with network depth, allowing early layers to explore freely while later layers lock onto emotion-relevant face anatomy. On AffectNet-8, KGAB-ViT achieves 72.27% top-1 accuracy, surpassing our plain ViT-B/16 baseline by 1.62 pp and the best published AffectNet-8 result by 8.50 pp.

Keywords: Artificial Intelligence Systems
Abstract: Seagrass ecosystems support coastal biodiversity and blue-carbon storage, but conventional monitoring under the DA-BFAR Participatory Resource and Socio-Economic Assessment framework remains labor-intensive and observer dependent. This study developed a camera-equipped quadrat and a two-stage deep learning pipeline for automated seagrass-cover assessment in Kauswagan, Lanao del Norte, Philippines. Stage 1 localized the quadrat frame from tide stratified underwater video. The selected YOLO26m-seg model achieved a test mask mAP@50 of 99.5%, with 100% precision and 99.85% recall. Stage 2 classified the quadrat’s 100 grid cells into seven modified Saito–Atobe coverage indices. Among ten evaluated variants, YOLO26l-seg achieved validation and test mask mAP@50 values of 95.73% and 86.98%, respectively. Across 16 test quadrats, model-derived percent cover showed strong agreement with expert verification, with an MAE of 3.11 percentage points, RMSE of 6.01 percentage points, and bias of −1.72 percentage points. The system automatically generates percent-cover estimates, coverage-index heatmaps, and habitat-condition classifications, providing a consistent approach to cell-level seagrass assessment.

Keywords: Artificial Intelligence Systems
Abstract: Insects play a significant role in the stability of ecosystems, biodiversity conservation and environmental monitoring. However, traditional methods of insect identification and preservation are time-consuming, tedious, and require taxonomic expertise. This paper introduces SmarTax, a unified system that integrates artificial intelligence-based insect identification and high-resolution 3D digitization of selected Coleoptera specimens. The system uses a controlled single-camera imaging setup with a motorized turntable to obtain multi-angle insect images. The captured multi-view images were used for photogrammetric reconstruction, while representative images from the same acquisition process were selected to train a YOLOv8-based identification model to detect morphological features and classify insect specimens. This study covers the families of beetles: Scarabaeidae, Curculionidae, Cerambycidae, Tenebrionidae and Passalidae. The YOLOv8m model achieved 81.88% precision, 73.68% recall, a 77.57% F1-score, and 76.89% mAP@0.5, while comparison of model dimensions with ImageJ-based references yielded an overall mean absolute percentage error of 4.18%. These results demonstrate the capability of SmarTax to support automated insect identification, interactive examination of pre-generated 3D insect models, and digital biodiversity documentation.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Navigation, Guidance and Control
Abstract: The integration of Vision-Language Models (VLMs) in autonomous Unmanned Aerial Vehicles (UAVs) offers unprecedented semantic reasoning capabilities. However, real-time closed-loop navigation requires not only low inference latency but also obedience to structured flight commands. This paper proposes a hybrid FSM-VLM control architecture for UAVs in GPS-free environments. The system combines a deterministic Finite State Machine (FSM) for low-level physical control with an asynchronous VLM copilot for high-level semantic pathfinding. We evaluate three models with different parameter scales in a Software-In-The-Loop (SITL) simulation. The framework isolates and measures syntax errors at the format level versus semantic hallucinations at the logic level in a normal and degraded scenario. The results indicate that structured reasoning reliability, rather than inference latency alone, is more closely associated with successful mission execution in the proposed hybrid architecture.

Keywords: Artificial Intelligence Systems, Autonomous Vehicle Systems, Robotic Applications
Abstract: The rapid growth of e-commerce requires efficient and scalable last-mile drone delivery systems. This paper presents a comparative evaluation of three scheduling approaches for a fleet of four homogeneous Unmanned Aerial Vehicles (UAVs) serving 100 customers within a 480-minute horizon under a strict minimum battery level of 30%. We investigate: (1) a Classical Greedy Optimization, (2) a Hybrid Greedy-PPO Approach, and (3) an end-to-end PPO approach trained from scratch. All methods account for payload-dependent energy consumption and recharging constraints. Experimental results show that the end-to-end PPO approach reduces makespan by 9.9% and charging operations by 57.7% compared with the greedy baseline. All schedules are validated in Gazebo with a simulation gap limited to 2.7-3.1%, confirming practical feasibility of the proposed framework.

Keywords: Artificial Intelligence Systems, Control Theory and Applications
Abstract: Ensuring the safety of Large Language Models (LLMs) is a critical challenge. While existing activation engineering methods offer computationally efficient alternatives to fine-tuning, they typically rely on static and indiscriminate interventions that can degrade the model's inherent inference abilities. To address this, we propose a novel, dynamic activation control framework for Mamba utilizing a low-dimensional latent space and Control Barrier Functions (CBFs). First, we map the high-dimensional activations into a low-dimensional latent space via a supervised autoencoder, defining a safe set using a Gaussian Mixture Model (GMM). During inference, if the internal state breaches this safety boundary, an optimal intervention vector is dynamically computed via gradient-based optimization to steer the state back into the safe region. Unlike conventional activation engineering, our proposal operates only when necessary, minimizing unnecessary modifications to the model's representations. Experimental evaluations using the mamba-2.8b model on the Real Toxicity Prompts dataset demonstrate that the proposed method significantly reduces toxic text generation while providing clear interpretability and mathematical safety guarantees.

Keywords: Robotic Applications, Artificial Intelligence Systems, Navigation, Guidance and Control
Abstract: This paper proposes a D^*-certified learning-guided mission reallocation framework for multi-UAV coverage missions under dynamic threat environments. In surveillance and reconnaissance missions, newly emerging threats can invalidate pre-planned target points and route segments, requiring online replanning and target reassignment. Although D^*-style incremental planning provides reliable local path replanning in changing maps, it does not directly handle mission-level decisions such as target priority, skipped-value loss, and workload balance among UAVs. To solve this problem, the proposed framework retains deterministic D^* planning as a safety core and introduces a lightweight learning module to estimate target-level mission values. The learned mission values are incorporated into a unified reallocation objective with D^*-based travel cost, risk, makespan, workload balance, skipped-value penalty, and coverage reward. Thus, the learning module guides mission reassignment without directly generating executable paths, while D^* guarantees that the final routes satisfy hard grid-path safety constraints. Simulation results show that the proposed method preserves zero unsafe crossing and improves mission-level reassignment performance, particularly in local dynamic-threat scenarios.

Keywords: Artificial Intelligence Systems, Robot Vision, Robotic Applications
Abstract: 6-DoF visual localization in complex indoor environments confronts the fundamental challenge of perceptual aliasing, exacerbated by long-term dynamic scene variations and severe occlusions. To overcome the limitations of purely geometric methods, integrating 3D Gaussian Splatting (3DGS) with semantic awareness has emerged as a breakthrough paradigm. Vision-Language Models (VLMs) play a pivotal role in this context by providing powerful viewpoint-dependent spatial relational priors. However, this integration encounters a core theoretical barrier: the modality gap under partial observability. Specifically, VLMs are highly prone to geometric hallucinations when reasoning about occluded objects, while static 3D landmarks fail to represent the viewpoint-dependent centroid shift. We propose a generalized localization framework that embeds occlusion awareness into a spatial optimization model. By applying an epistemic-bounded Set-of-Mark (SoM) mechanism, we compel the VLM to infer relationships solely from strictly visible regions. Furthermore, by redefining 3D objects as a distribution of spatial hypotheses and establishing a continuous ray-casting pruning mechanism, the proposed framework drastically reduces the search space, setting a new benchmark for robust single-shot global localization under extreme environmental conditions.

Keywords: Autonomous Vehicle Systems, Robot Vision, Sensors and Signal Processing
Abstract: Visual-Inertial Odometry (VIO) systems often suffer performance degradation in visually challenging environments containing illumination variation, dense vegetation, dynamic objects, and occlusions. Although multi-camera systems improve visual coverage and geometric redundancy, most existing frameworks directly aggregate observations from all cameras into a unified optimization process without explicitly considering the observability quality of each individual viewpoint. As a result, unreliable visual constraints from degraded cameras can deteriorate the numerical stability of the optimization process and reduce localization robustness. To address this limitation, this paper presents a robust non-overlapping multi-camera VIO framework with camera-wise degeneracy awareness. First, a degeneracy detection module is proposed to analyze the observability condition of each camera independently using the visual Hessian matrix, enabling the identification of locally weakly observable visual constraints. Second, a degeneracy-aware mitigation strategy is introduced to suppress unreliable optimization updates caused by degraded visual measurements while preserving reliable geometric information from healthy viewpoints. Finally, the proposed framework is evaluated on both public benchmarks and self-collected datasets under diverse perceptually degraded environments. Experimental results demonstrate that the proposed system achieves superior robustness and estimation accuracy compared to state-of-the-art (SOTA) multi-camera VIO methods.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems
Abstract: This paper presents a memory-based navigation framework fusing deep reinforcement learning with gap-based method to tackle local minima and partial observability in unknown environments. We integrate an LSTM-augmented Soft Actor-Critic (SAC) policy to model sequential sensor data into latent belief states for improved environment awareness. To ensure robust training without reward hacking, a dense reward function utilizing a gated trajectory progress penalty is introduced. Additionally, a sub-goal manager is deployed to extract the robot from deadlocks detected via LiDAR. Evaluated using ROS and Gazebo, our framework demonstrates capability over reactive baselines in escaping traps while maintaining high computational efficiency during real-time online deployment.

Keywords: Robot Vision
Abstract: Sparse class-agnostic masks from the Segment Anything Model (SAM) provide rich object hypotheses for outdoor mapping, yet per-frame local IDs break temporal consistency and degrade object-aware 3D Gaussian Splatting (3DGS). We propose a pose–LiDAR association module that warps recent keyframe masks with LiDAR–visual inertial odometry (LIVO), fuses a short multi-keyframe prior, and recovers global IDs by bipartite Hungarian matching under sparse detections, with an explicit indoor/outdoor warp-gating rule. Associated IDs supervise 3DGS through pose consistent identity seeding (PCIS), association-confidence weighting (ACW), and an associated-ID 3D consistency loss Lid3d. On outdoor LIVO sequences and KITTI-360, re-associating identical detections approaches a mask-IoU Hungarian oracle and surpasses MOT/VOS peers (KITTI ID persistence 0.97 vs. oracle 0.94). Consistent IDs enable strong holdout instance quality: Replica mIoU 0.85 / fwIoU 0.98; KITTI-360 fwIoU 0.49, exceeding Gaussian Grouping and SGS-SLAM under a matched evaluation protocol.

Keywords: Robot Vision, Robot Mechanism and Control, Robotic Applications
Abstract: Vision-Language-Action models (VLAs) provide a unified interface for mapping visual observations and language instructions to robot actions, but dexterous hands remain difficult to include because multi-finger actions are high-dimensional, coupled, and sensitive to contact timing. We present an end-to-end dexterous VLA pipeline that predicts arm and hand actions together without a separate hand controller. Hand Synergy Latent compresses high-dimensional hand joint commands into a compact latent action, while Hand Intent predicts desired hand-pose progress and selects the decoded hand candidate that best matches the task pose from the VLA action chunk. Across grasp, push, and hook- pull tasks, the compact representation reaches the same convergence criterion with 31.8% fewer optimization steps than direct 16-dimensional hand-action prediction. Intent-guided execution further reduces hand-transition timing error by 78.7–88.2% and pose-progress error by 42.9–55.6% compared with VLA-only execution.

Keywords: Navigation, Guidance and Control, Sensors and Signal Processing, Control Theory and Applications
Abstract: Heterogeneous track-to-track fusion (T2TF) must combine local tracks represented in different state spaces, such as active Cartesian tracks and passive azimuth/elevation tracks. Information matrix fusion (IMF) reduces repeated information counting in T2TF with memory by fusing the information increment of each local update. A passive angular increment, however, represents line-of-sight uncertainty whose natural support is the unit sphere rather than the azimuth/elevation plane. This paper presents geodesic-normal information fusion (GN-IMF), which reparameterizes the passive predicted/posterior angular Gaussian pair as a geodesic-normal density pair on S2. The posterior-to-prior density ratio defines a directional information factor that is pulled back through the passive line-of-sight map and locally approximated as a Cartesian-compatible Gaussian information increment for the standard IMF update. In 500-run Monte Carlo simulations, GN-IMF reduces nominal position RMSE from 41.62 m to 37.51 m relative to conventional IMF and reduces mean 6-D ANEES from 10.39 to 5.27. Additional passive-noise and high-elevation evaluations show that the improvements in accuracy and consistency persist under the tested conditions.

Keywords: Navigation, Guidance and Control, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: Reliable vehicle localization is essential for autonomous driving, particularly in environments where GNSS signals are unreliable. In such environments, map-based localization improves robustness by associating sensor observations with prior road geometry. In particular, Standard-Definition (SD) maps such as OpenStreetMap (OSM) provide lightweight road information, but their limited geometric precision makes robust localization challenging. Existing image-based SD-map localization methods primarily rely on sparse structural cues or rasterized map features and often suffer from degraded performance in feature-poor environments such as highways and rural roads. In contrast, road centerlines provide road-level geometric information that can be directly associated with SD-map road networks. In this study, we propose a monocular-camera-based road centerline representation framework for SD-map-based localization. The proposed method predicts road centerline representations in the Bird’s-Eye-View (BEV) domain using a transformer-based set prediction network with feature-guided dynamic queries, representing each road centerline as a compact cubic Bezier curve. Experimental results on a dataset show that the proposed method improves the F1-score from 53.0% to 71.3% and reduces the Chamfer Distance from 1.81,m to 1.42,m compared with a DETR-style baseline. Furthermore, particle-filter-based localization experiments demonstrate that the predicted road centerlines provide effective observations for SD-map-based localization.

Keywords: Navigation, Guidance and Control, Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: Model-free deep reinforcement learning can learn sensor-based navigation policies without a map, but it is physics-agnostic: exploration causes thousands of collisions, sample efficiency is poor, and learned paths exhibit excessive turning. We propose PI-PPO, a physics-informed extension of Proximal Policy Optimization that injects prior physical knowledge through four complementary modules: (i) a PINN-style auxiliary loss that penalizes the residual of the contact-inclusive equation of motion, with dense collision labels derived for every action from lidar ray geometry; (ii) a separable Hamiltonian network H = T(p) + V(q) trained with a minimum-control-effort prior, whose learned potential V is fed back through policy-invariant potential-based reward shaping; (iii) a nonholonomic action constraint with a curvature penalty reflecting unicycle kinematics; and (iv) a discrete-time control barrier function safety filter defined directly in lidar sensor space. Across three randomized mazes of increasing difficulty and five seeds per variant, PI-PPO eliminates training collisions entirely (versus ~2,500 for PPO), raises the success rate from 53% to 100%—on the hardest maze vanilla PPO never succeeds while PI-PPO always does—and reduces final-path turns from 10.0 to 6.0 while staying within 2.6% of the A* optimal length, at 1.8× wall-clock cost. Ablations isolate the contribution of each module—attributing most of the headline gains to the nonholonomic constraint and CBF shield, while the PINN and Hamiltonian modules mainly add an interpretable, data-driven dynamics model in this discrete setting—and expose a practical lesson: the conservativeness of the safety filter directly governs whether learning remains possible.

Keywords: Navigation, Guidance and Control, Sensors and Signal Processing, Autonomous Vehicle Systems
Abstract: This extended abstract summarizes a Wasserstein distributionally robust optimization-based extended Kalman filter (DRO-EKF) for robust GNSS/INS integrated navigation under nonstationary measurement errors. Conventional EKF-type filters rely on a nominal measurement-noise covariance and can therefore suffer severe performance degradation when GNSS measurements include outliers, bias, heavy-tailed noise, or spoofing components. The proposed method constructs an empirical distribution from recent measurement residual samples and defines a Wasserstein ambiguity set centered on it. In the measurement update, the residual mean shift is compensated using a mean-centered residual, while residual covariance distortion is reflected through an adaptive covariance update based on Wasserstein shrinkage. Numerical simulations and real flight-test-based GNSS/INS validation show that the proposed filter suppresses both average position error and instantaneous peak error more effectively than EKF, AKF, MCKF, HEKF, and DRKF.

Keywords: Control Theory and Applications
Abstract: In this paper, we present a data-driven output feedback controller for nonlinear systems that achieves practical output regulation, using noise-free input/output measurement data. The proposed method has two main components: (i) an inverse model of the system identified via kernel methods, which is a mapping from the current state and a desired output to the corresponding control input; and (ii) a data-driven reference selection framework that actively chooses a suitable reference output trajectory from the dataset which has been used for the identification. We establish a verifiable sufficient condition on the dataset under which the proposed controller guarantees practical output regulation. Numerical simulations demonstrate the effectiveness of the proposed controller, while additional evaluations under output measurement noise provide empirical evidence of its robustness.

Keywords: Control Theory and Applications
Abstract: Networked multi-agent systems are vulnerable to integrity attacks because local state corruption and forged information can propagate through the communication graph. To address this issue, we propose a verifiable consensus framework that mitigates vulnerabilities in networked multi-agent systems. By using a distributed protocol that securely aggregates proofs from individual agents and employing elliptic-curve cryptography, we reduce computational and communication burden. In addition, by using Pedersen commitment, we enhance privacy protection by preventing possible data leakage. Finally, we empirically demonstrate the efficacy of our method through a distributed simulation of a consensus problem.

Keywords: Control Theory and Applications
Abstract: This paper proposes a hierarchical framework integrating soft actor-critic reinforcement learning with differentiable nonlinear model predictive control (Diff-NMPC) for path tracking in rear wheel steering autonomous racing. In the proposed framework, the stage and terminal cost weights of the NMPC problem are treated as learnable parameters and are optimized end-to-end using analytical sensitivities obtained from the optimal control solver. Simulation results show that the analytical sensitivity closely matches finite-difference estimates, with absolute errors on the order of 10e−6 to 10e−8. With a prediction horizon of 40, the validation lap time reached 5.08 s, corresponding to a 15.9% reduction relative to the case with a horizon of 25 (6.04 s) and a 3.8% improvement over conventional NMPC with fixed cost weights evaluated at a horizon of 40. On the tested desktop platform, the average total computation time remained below 3.7 ms, which is compatible with the 20 ms sampling time used in the simulations.

Keywords: Control Theory and Applications
Abstract: This paper investigates the absolute stability of feedback systems with slope-restricted, idempotent and repeated nonlinearities. It is well known that, within the framework of integral quadratic constraints (IQCs), suitable multipliers can be employed to derive (primal) linear matrix inequality (LMI) conditions for ensuring the absolute stability. However, a limitation of this approach is that no conclusion on the absolute stability can be obtained if the corresponding primal LMI is (numerically) infeasible. Recent studies have shown that new insights can be obtained by focusing on the dual formulation of the primal LMI. Motivated by these developments, we address the analysis based on dual LMIs. As the main results, conditions are derived for constructing destabilizing nonlinear operators and extracting non-trivial equilibrium points, thereby certifying that the considered systems are not absolutely stable. Numerical examples are provided to demonstrate the validity of the proposed results and to compare them with existing dual-LMI-based approaches.

Keywords: Control Theory and Applications
Abstract: Many existing software packages, including MATLAB, perform index reduction for DAEs with an index greater than one using the Pantelides method. However, the Pantelides method may fail because numerical cancellation can occur during the reduction process. Iwata et al. proposed a method for constructing DAEs to which the Pantelides method can be successfully applied. In this study, we implement a transformation algorithm based on mixed matrices using MATLAB's Symbolic Math Toolbox and develop software that converts general linear DAEs into forms amenable to the Pantelides method.

Keywords: Control Theory and Applications
Abstract: This paper investigates the stability and L_2-gain performance of periodic piecewise systems with time-varying subsystems. Under a piecewise continuous time-varying Lyapunov function framework, an augmented linear matrix inequality is constructed by introducing a positive definite matrix and an auxiliary matrix satisfying a positivity constraint to address the negative definiteness verification of a matrix polynomial arising from the Lyapunov derivative. Compared with existing methods, a less conservative matrix polynomial negativity criterion is proposed. Based on this criterion, a sufficient condition for exponential stability is established, and an L_2-gain performance criterion is derived. All conditions are expressed directly in terms of linear matrix inequalities, and a numerical example is provided to demonstrate the effectiveness of the proposed method.

Keywords: Robot Mechanism and Control, Control Theory and Applications, Robotic Applications
Abstract: This paper proposes a residual reinforcement learning-based hybrid control architecture for the Longitudinally Extended Two-Wheeled Inverted Pendulum Robot (LE-TWIPR) equipped with a sliding mechanism. Although the LE-TWIPR provides a large space for payload placement, payload-induced longitudinal center-of-mass (CoM) shifts make it challenging to simultaneously achieve balance and velocity tracking. Model-based control provides physically interpretable nominal behavior but degrades as unmodeled factors become more pronounced, while pure reinforcement learning (RL) provides high adaptability but offers limited physical interpretability, as the contributions of individual actuators are embedded implicitly within a single learned policy. The proposed architecture uses the slider input for CoM compensation by decomposing it into a model-derived slider reference obtained from the quasi-static equilibrium condition and a learned residual, while generating the wheel torques directly through the RL policy. The policy is trained using PPO with an asymmetric actor-critic structure in Isaac Lab. Simulation results show that, compared with pure RL, the proposed architecture reduces velocity tracking error (RMSE) by up to 25%, overshoot by 58%, and slider position variation by 26%, while exhibiting faster and more consistent recovery under unseen push disturbances. The slider command decomposition shows that the model-derived slider reference represents the nominal slider behavior, while the residual term provides dynamic correction during velocity transitions. This decomposition improves interpretability and contributes to improved tracking performance.

Keywords: Robot Mechanism and Control, Artificial Intelligence Systems, Robotic Applications
Abstract: This study identifies three failure modes of Gaussian mixture model (GMM)-based indirect joint optimization for robot morphology-control co-design, namely component collapse, premature convergence, and initialization sensitivity, and proposes a failure-aware framework (F-JO) that resolves them through four targeted elements. On the Hopper and Walker2d benchmarks, F-JO improves the final return by approximately 14% and 35% over the conventional formulation (C-JO), respectively, and is the only co-design method that retains a 100% success rate on both benchmarks

Keywords: Human-Robot Interaction, Autonomous Vehicle Systems, Robotic Applications
Abstract: Combat support unmanned ground vehicles (UGVs) should follow an operator while maintaining a support position that enables continuous observation and immediate assistance. However, conventional following methods based on the operator’s current position or a fixed following distance may cause unnecessary repositioning or observation loss in obstacle-rich environments. This paper proposes an intent-aware leader-following framework using an anticipatory observation-support cost. The proposed framework estimates the operator’s short-term motion intent from position, velocity, and moving direction, predicts future observation availability, and generates candidate support positions within the support feasible region. The optimal support position is selected by jointly considering observation continuity, support availability, and movement cost. Repositioning is executed only when future support effectiveness is expected to improve. Simulation results in battlefield-like scenarios show that the proposed method maintains an Observation Loss Ratio of 0.000 in all tested scenarios while achieving the lowest average Repositioning Ratio compared with fixed-distance, current-cost, and prediction-based following methods.

Keywords: Robotic Applications, Robot Mechanism and Control
Abstract: Dielectric elastomer actuators (DEAs) are promising for small-scale robots. However, spray-coated single-walled carbon nanotube (SWCNT) electrodes often exhibit sharp surface peaks, inducing localized electric field concentrations and accelerating dielectric breakdown. To address this, we propose improving the surface morphology and breakdown strength of SWCNT electrodes by adjusting the atomization pressure, eliminating post-processing. Higher atomization pressure reduces droplet size, significantly decreasing both average surface roughness and extreme peak heights, thereby enhancing the breakdown strength of multilayer DEA structures. A large-area coating process was established with an atomization pressure of 0.20 MPa, an 80 mm nozzle height, and a 0.02 MPa supply pressure. Using the above process parameters, a multilayer rolled DEA was fabricated and compared with an actuator fabricated at a low atomization pressure. Results showed that the DEA fabricated at the higher atomization pressure withstood higher electric fields (up to 35 V/µm) without early breakdown, exhibiting superior static and dynamic actuation performance. Finally, the actuator was integrated into a small-scale robot leg, generating a periodic stroke of 5.5 mm at 15 Hz. This successfully demonstrates the feasibility of reliable DEA actuation for small terrestrial robots.

Keywords: Robot Mechanism and Control, Robotic Applications, Control Devices and Instruments
Abstract: This paper presents the design and simulation-based evaluation of a compact ground-aerial hybrid robot that combines quadrotor flight with wheel-assisted ground locomotion through a shared-actuation architecture: rather than adding independent wheel-drive actuators, each propulsion motor is reused for ground locomotion by redirecting its output to the wheel axis through a bevel-gear transmission. The platform targets compact indoor mobility, using flight to overcome obstacles or discontinuous terrain and ground travel to reduce energy consumption during surface motion. The bearing-supported bevel-gear drivetrain provides 90-degree power transmission while transferring radial and reaction loads to the bearings and gearbox housing rather than directly to the motor shaft. For a wheel radius of 0.035 m and target ground speeds of 1.0-1.5 m/s, the required wheel speed is about 273-409 rpm, which a 1:3 transmission ratio maps to about 819-1228 rpm at the motor, indicating kinematic compatibility with low-speed ground operation. A motor-load model (rolling resistance, acceleration, wheel and motor torque, and estimated current) and a MATLAB/Simulink ground-mode simulation are developed; preliminary closed-loop results track the commanded velocities, settle at motor speeds consistent with the analytic operating points, and indicate that the dominant torque and current demand occurs during startup transients rather than steady cruise. An integrated Gazebo/PX4/ROS 2 framework is organized to support mode management, flight-control integration, and future higher-fidelity validation. The study thus provides a platform-level design basis for shared-actuation ground-aerial robots; future work includes low-speed ESC bench testing, drivetrain prototyping, ground-contact experiments, and integrated mode-transition validation.

Keywords: Robot Vision, Artificial Intelligence Systems, Control Theory and Applications
Abstract: In this work, we prove that the uncontrollable linearized dynamics can be made impulse controllable and feedback stabilizable by the random projective transformation of a state equation to a differential-algebraic form. The original dynamics in the configuration space and its learned latent representation are shown to be system and feedback equivalent. The resulting differential-algebraic dynamics, symbolic of dynamics on learned latent spaces without labels, is well-defined and controllable at∞—thus characterizing a freely initializable dynamics. These assertions mathematically guarantees that learned latent dynamics of a moving agent — irrespective of the instability of the original dynamics in configuration spaces — are feedback stabilizable. Building on these results in the behavioral framework, we argue that for such a random differential-algebraic equation (RDAE), a unique solution exists for any two arbitrary states and initial conditions if the matrix pencil of the characteristic matrix A and the random multiplier R is row regular. The proposed paradigm sheds new light on the long-standing conundrum regarding augmented controllability in unsupervised learning of latent dynamics compared to joint-space control.

Keywords: Robot Vision, Artificial Intelligence Systems, Robotic Applications
Abstract: Pallet pose estimation from a single viewpoint remains sensitive to occlusion and unfavorable viewpoints. This paper presents Pallet Next-Best-View (PAL-NBV), the first NBV approach designed specifically for pallet pose estimation. PAL-NBV builds an occupancy map from synchronized RGB-D images and selects viewpoints that minimize estimated pallet occlusion, while staying within an empirically determined operating range of a given pallet pose estimator. PAL-NBV consults a large language model if viewpoint generation based on the occupancy map fails. Across 968 simulated warehouse scenarios, PAL-NBV reduces the median translation error from 0.383mathrm{m} to 0.103mathrm{m} and the median rotation error from 2.54^circ to 0.96^circ at the cost of increased path lengths.

Keywords: Robot Vision, Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: Detector class confidence alone does not provide explicit estimates of localization or velocity reliability, nor does it directly capture tracker-specific temporal stability or detection–track compatibility. We augment BEVDepth and StreamPETR with auxiliary localization- and velocity-quality branches, yielding QBEVDepth and QStreamPETR, and study how these bounded quality proxies can be consumed at different tracker interfaces. QTrack exposes temporal-confidence, association, and output-score interfaces, whereas Original PolyMOT retains its internal tracking decisions and uses detector quality only through its exported tracking score. Tracker-side models and lookups are fitted on 50 of the 150 nuScenes validation scenes, while all primary tracking results are evaluated on the remaining 100 disjoint scenes. With QTrack, detector-specific combined configurations yield AMOTA point-estimate changes of +0.0032 for QBEVDepth and +0.0033 for QStreamPETR. With PolyMOT, output-score quality conditioning produces changes of +0.0043 and +0.0053, respectively. The auxiliary detector-quality branches add a small detector-side computational footprint. These results support tracker-interface-aware quality consumption rather than a single universal insertion rule.

Keywords: Robot Vision, Autonomous Vehicle Systems, Sensors and Signal Processing
Abstract: This study addresses the problem of robust gate perception and autonomous gate traversal for unmanned aerial vehicles (UAVs) operating in GPS-denied environments. The objective is to develop a practical, perception-driven flight pipeline that enables a UAV to detect a target gate, align with it, and execute traversal using visual servoing. To this end, we propose a robust hollow gate detection algorithm (HGDA) that directly exploits the structural characteristics of the gate. The method formulates gate detection as the extraction of a geometrically consistent quadrilateral from color and contour cues in the image. The resulting image-plane features are used as inputs to a moment-based visual servoing controller for closed-loop flight control. The proposed system is validated through real-world experiments, demonstrating reliable gate detection and successful autonomous traversal in GPS-denied environments.

Keywords: Robot Vision, Navigation, Guidance and Control, Artificial Intelligence Systems
Abstract: Flying insects navigate cluttered three-dimensional environments using optic flow, achieving robust obstacle avoidance without range sensors, depth maps, or explicit metric maps. This biological strategy offers a computationally lightweight alternative to the multi-sensor fusion pipelines that dominate engineered autonomous navigation, and is particularly attractive for size, weight, and power constrained aerial vehicles. In this work, we formulate bio-inspired obstacle avoidance as a continuous-control deep reinforcement learning problem in which a simulated model perceives its environment through an optic flow field and traverses an obstacle-laden 3D environment collision-free to an end goal. A convolutional feature extractor maps the spherical optic flow field to continuous action commands. We conduct a controlled, like-for-like comparison between an on-policy method, proximal policy optimization, and an off-policy method, soft actor-critic, under an identical environment, observation space, network architecture, and collision-penalty structure. Both policies are evaluated across different obstacle densities using safety and efficiency aware metrics. The results characterize the trade-offs between sample efficiency, stability, and trajectory quality for obstacle avoidance based purely on optic flow, and provide practical guidance on algorithm selection for bio-inspired aerial navigation.

Keywords: Robotic Applications, Artificial Intelligence Systems, Robot Vision
Abstract: Robotic sorting of recyclable waste is challenging due to the deformable and geometrically inconsistent nature of target objects. We present a training-free suction grasping system for sorting deformed aseptic beverage cartons, decoupling target identification from grasp-point selection. An open-vocabulary vision-language model detects cartons from a text prompt, SAM2 refines each detection into an instance mask, and a geometric scoring method selects the suction point by combining surface flatness with normal alignment. Three geometric methods are compared: k-nearest-neighbour PCA, Sobel cross-product, and RANSAC plane fitting. Evaluated on a real robot across three deformation levels and 35 cluttered scenes, single-object grasp success reaches 88.2% and end-to-end retrieval in clutter is 72.6%.

Keywords: Navigation, Guidance and Control, Control Theory and Applications, Autonomous Vehicle Systems
Abstract: Deployment of quadrotor UAVs in time-critical missions requires time-optimal trajectory generation. While existing trajectory generation approaches are promising, they are either computationally expensive or rely on simplified motion models that inadequately capture vehicle dynamics. In this paper, we propose time-optimal trajectory generation for quadrotors using a jerk-controlled triple integrator motion model that captures key aerodynamic effects and finite attitude dynamics while maintaining computational tractability. We formulate the corresponding time-optimal control problem and derive semi-analytical solutions for a class of scenarios, enabling efficient trajectory computation. For the general case, the resulting time-optimal control problem is solved numerically using a direct shooting approach. We evaluate the proposed approach against existing trajectory generation methods in terms of trajectory tracking performance and computational time, demonstrating that it generates reasonable trajectories while requiring significantly less computation.

Keywords: Navigation, Guidance and Control, Human-Robot Interaction, Control Theory and Applications
Abstract: This paper studies socially acceptable robot roaming in indoor public venues such as museums and galleries. The robot must maintain human safety and semantic social compliance while continuing to visit different venue regions. We present a risk-gated, uncertainty-aware stimulus-MPC planner that combines stimulus-field reference tracking, effective-distance social activation, context-conditioned viewer/group/flow costs, and anchor-based coverage scheduling. The method is evaluated in a 3D museum benchmark over 20 random seeds and 540 planning steps against reactive roaming, social-force roaming, DWA roaming, deterministic stimulus control, ORCA-style roaming, and an official RVO2/ORCA baseline. Results show comparable venue coverage with substantially improved minimum human clearance, minimum time-to-collision, collision rate, motion smoothness, and semantic group/flow disturbance. Component ablations further isolate the roles of uncertainty inflation, risk gating, and semantic social costs. External replay checks on NavWareSet and SiT provide additional support beyond procedurally generated pedestrians. The results indicate that local reciprocal collision avoidance alone is insufficient for socially acceptable venue roaming, while risk-gated uncertainty-aware stimuli provide a practical control structure for public indoor navigation.

Keywords: Navigation, Guidance and Control, Robot Vision, Artificial Intelligence Systems
Abstract: Dragonflies intercept airborne prey using only visual information with no range sensor, no GPS, and no explicit knowledge of the target's position or closing velocity. Their pursuit strategy closely resembles a proportional navigation (PN) guidance law widely used in engineered interception systems. However, classical PN implementations require full state information, explicit range to the target, relative position vectors, closing velocity and a fixed navigation constant which governs how sharply the pursuer turns in response to target angular drift. We present a biologically inspired deep reinforcement learning (DRL) framework that recovers the guidance-relevant quantities: line-of-sight (LOS) direction and LOS rotation rate, purely from {optic flow}, eliminating the need for range sensing. Our system integrates three components: bio-inspired target detection from background flow, an optic-flow-based pipeline that recovers the LOS direction and its angular rate without any range measurement, and a DRL agent that {exhibits} an {emergent}, adaptive state-dependent navigation constant N(mathbf{s}). We evaluate the framework against a classical 3D true proportional navigation controller operating with full state information, a pure pursuit baseline, and a dragonfly foveation model. Results demonstrate that the DRL agent achieves improved tracking and interception performance using optic flow, and that the {emergent} navigation constants align with values measured in predatory insects. This work supports the hypothesis that insect pursuit can be implemented with image-plane information combined with a self-motion estimate, providing a practical, optic-flow{-based} tracking and interception controller for resource-constrained autonomous platforms.

Keywords: Navigation, Guidance and Control, Robot Vision, Autonomous Vehicle Systems
Abstract: This paper presents a controller-level obstacle-avoidance framework for a small drone with a limited forward-camera field of view. An obstacle may disappear from the image after a lateral maneuver, causing a memoryless controller to remove the obstacle cost near the closest encounter. The proposed Model Predictive Path Integral (MPPI) layer retains a short-term obstacle belief with confidence decay and uncertainty growth, uses uncertainty- and delay-inflated clearance, and optionally controls an independently actuated camera yaw toward the most risky memorized obstacle. A two-dimensional software-in-the-loop (SIL) simulator compares the method with memoryless, fixed-inflation, and last-observation-hold baselines under simulated position noise, dropout, and delay. Each controller--condition pair is evaluated over 100 random seeds; success confidence intervals, collision and timeout rates, and clearance variability are reported. The last-observation-hold baseline is competitive, while the full gaze-memory controller obtains 96--98% success and no collisions in the tested setting. These results constitute controller-level validation with simulated world-coordinate measurements and persistent obstacle identities, not validation of an image-based perception pipeline or physical flight.

Keywords: Navigation, Guidance and Control, Robotic Applications
Abstract: Reliable place inference is important for long-horizon navigation in large-scale environments where precise metric maps are costly to build and maintain. Humans use simplified navigational cues, such as sketch maps, place signs, and directional signs, to infer their location, but robots cannot directly rely on these cues because sign observations may be noisy, missing, or ambiguous and sketch maps may distort distance, scale, and direction. These limitations make independent local matching unreliable and motivate sequential reasoning over accumulated evidence. We propose SPIN, an offline graph-constrained sequential place inference framework. SPIN constructs a semantic observation sequence, generates candidate places on a sketch-map graph, and formulates place inference as minimum-cost sequence estimation over candidate places. Candidate sequences are evaluated using place-name costs and transition costs from sketch-map topology, directional-sign consistency, and alignment-aware sign-position consistency. Since sign-position consistency requires sketch-to-observation alignment, SPIN uses two-stage inference: the first stage estimates an initial place sequence and alignment from reliable place correspondences, and the second refines the sequence with alignment-aware costs. Skip transitions bypass unreliable observations and reduce false-positive inclusion. Experiments in the Namsan hiking-trail environment show that SPIN infers observation-consistent place sequences under noisy semantic observations and simplified sketch-map priors.

Keywords: Artificial Intelligence Systems, Control Theory and Applications, Robot Mechanism and Control
Abstract: Learning-based variable impedance control (VIC) has gained increasing attention as a promising approach for achieving adaptive compliance in contact-rich manipulation. Meanwhile, diffusion policies have recently attracted attention in robotic imitation learning, but their use in impedance control remains largely unexplored. In this paper, we propose a diffusion policy-based framework for learning VIC from expert demonstrations. Unlike conventional learning-based VIC methods that often rely on deterministic policies or reinforcement learning-based optimization, the proposed method formulates VIC as a conditional sequence generation problem. Experiments on a peg-in-hole task are conducted across different difficulty levels and compared with baseline methods. The results show that the proposed method achieves more reliable performance across varying task difficulties. In addition, the learned gain profiles indicate that the diffusion policy successfully captures phase-dependent impedance modulation consistent with expert demonstrations.

Keywords: Control Theory and Applications
Abstract: This paper studies a multi-mode approach to solve the leader-following consensus problem of Multi-Agent Systems (MASs) under random Denial-of-Service (DoS) attacks. MASs rely heavily on communication networks, making them susceptible to cyber-attacks that can impair information exchange and system performance. To address this problem, this paper proposes a novel multi-mode-based approach and introduces a general-transition probability Markov model to handle random DoS attacks, which are more complex than conventional ones. Furthermore, a new concept, the Leader-Follower Graph-Informed Gain Matrix (LFGIGM), is proposed to enhance the multi-mode approach. The proposed approach is demonstrated its effectiveness through a numerical example.

Keywords: Process Control Systems, Industrial Applications of Control, Control Theory and Applications
Abstract: Air-conditioning systems account for a major share of electricity consumption in large buildings. This paper proposes a supervisory control (SC) framework for determining the optimal reference temperature of both chilled-water supply and floor-level zones in a multi-floor cooling system. In the previous study, the heat-transfer behavior of the air handling unit was approximated using Chen’s approximation. On the other hand, this study employs a dynamic model of air handling units to obtain a more accurate physical model. The SC design problem is formulated as a quadratic programming problem that minimizes a weighted combination of normalized Total Operating Cost (TOC) and Thermal Comfort Cost (TCC) under operating limits and comfort constraints. Historical ambient-temperature and cooling-load data are used as disturbances, while floor-level cooling load profile is considered to reflect different thermal conditions among floors. Numerical simulations for a six-floor library building show that the proposed SC method increases the chilled-water supply temperature from the fixed value of 8∘C to approximately 9-13.3∘C. Compared with the fixed-reference strategy, the proposed method reduces the daily-equivalent TOC by 14.02% and the TCC by 63.3%. The proposed SC with combined chiller and AHU dynamic model provides a more realistic setpoint selection with the achievable energy-saving potential and maintaining the thermal comfort of the occupants.

Keywords: Process Control Systems, Industrial Applications of Control, Control Theory and Applications
Abstract: Air conditioning systems are the largest energy consumers in commercial buildings, particularly in Thailand’s hot and humid climate. Efficient HVAC control that balances energy savings with occupant thermal comfort is therefore of significant practical importance. This project presents a two-layer HVAC controller comprising a Supervisory Control (SC) layer, which employs constrained optimization to determine optimal temperature and humidity setpoints by minimizing total operating cost and thermal comfort cost, and a Reinforcement Learning (RL) layer based on the Proximal Policy Optimization with Lagrangian (PPO-Lagrangian) algorithm to design a tracking controller that satisfies thermal comfort and output bound constraints formulated as a Constrained Markov Decision Process (CMDP). The proposed controller was evaluated on real building data under three Coefficient of Performance values – COP = 4,5,6 – representing nominal and off-nominal operating conditions. Results show that the PPO-Lagrangian controller achieves significantly better humidity ratio tracking than Model Predictive Control (MPC), with RMSE up to five times lower, while maintaining comparable total operating cost. The controller also demonstrates robust adaptability under plant parametric uncertainty.

Keywords: Control Theory and Applications
Abstract: This paper investigates the sampled-data synchronization problem for delayed neural networks subject to communication delays. To reduce the conservatism of existing synchronization criteria based on semi-looped functionals, an augmented-state semi-looped functional is developed. In contrast to conventional approaches that utilize only the derivative information of synchronization errors in integral terms, the proposed method incorporates both the synchronization error and its derivative into an augmented-state framework. By exploiting the relationship between the state and state-derivative information, additional characteristics of the synchronization error dynamics can be reflected in the stability analysis. Based on the proposed functional, generalized integral inequalities and free-weighting matrix techniques are employed to derive novel delay-dependent synchronization conditions in the form of linear matrix inequalities. The resulting criterion preserves the main structure of the semi-looped functional approach while providing a tighter estimation of integral terms associated with communication delays and sampled-data effects. Numerical examples are presented to demonstrate the effectiveness of the proposed method. The results show that the proposed criterion admits larger maximum allowable sampling intervals than existing methods, indicating a reduction in conservatism.

Keywords: Control Devices and Instruments, Industrial Applications of Control
Abstract: Variable Frequency Drives (VFDs) are widely deployed alongside AC electric motors in industrial systems to improve energy efficiency and reduce motor wear and tear, thereby lowering maintenance costs. With proper maintenance and modernization, the drives’ lifespans can extend to decades, but they will eventually need to be replaced. This case study explores the challenges of replacing VFDs in existing industrial installations when backward compatibility is not possible. It also characterizes the retrofitting process and identifies potential for automation. We interviewed 22 employees of a large, internationally operating VFD manufacturer about retrofitting and its challenges. The analysis reveals a four-phase process for retrofitting: drive selection, physical installation, parameter configuration, and system testing. Among these phases, parameter configuration appears to be the most challenging and the most promising target for automation, as it requires deep domain expertise, lacks one-to-one parameter mappings between drive generations, and must meet the strict downtime constraints of industrial operations. Our findings also include considerations of how the VFD use case's application domain can affect the retrofitting process.

Keywords: Control Devices and Instruments, Sensors and Signal Processing
Abstract: This paper introduces a home-made ion-trap micromotion detector for quantum physics experiments. Based on an FPGA board, it detects the time-differences between the RF synchronized reference clock and the photon emissions, using the accumulated statistics to draw the histogram to indicate the micromotion. The instrument runs online in real-time, its result can be used to guide the compensation and minimizing the micromotion in an ion-trap experiment.

Keywords: Control Theory and Applications, Control Devices and Instruments, Sensors and Signal Processing
Abstract: This research realizes a controller method for a rotary inverted pendulum system, aiming to achieve swing-up and stabilization control. The system architecture was modeled and simulated using MATLAB/Simulink and validated on the Quanser QUBE-Servo 3 platform. Experimental results demonstrate the effectiveness and feasibility of the strategy even when disturbance exists.

Keywords: Navigation, Guidance and Control, Sensors and Signal Processing, Control Devices and Instruments
Abstract: This paper presents an indoor three-dimensional (3D) mapping system using a compact quadrotor unmanned aerial vehicle (UAV) equipped with a red-green-blue and depth (RGB-D) camera. Intended for disaster-damaged or confined environments where human access is difficult, the system uses Robot Operating System 2 (ROS 2) and Real- Time Appearance-Based Mapping (RTAB-Map) to process color, depth, and inertial data, estimate the UAV trajectory, and generate a colored point cloud. In an indoor experiment, the point cloud was exported in Polygon File Format (PLY) and evaluated for plane flatness, reconstructed plane thickness, spatial dimensions, outlier ratio, point distribution and density, and missing regions. The overall environmental structure was reconstructed, although noticeable dispersion and duplicated surfaces appeared near floors, walls, and object boundaries. Possible causes include UAV motion, attitude changes, sensor vibration, and accumulated pose-estimation error. The results indicate the feasibility of compact-UAV- based indoor mapping and identify the procedures needed for rigorous accuracy assessment. Future work will compare mapped dimensions with reference measurements and evaluate repeatability under different flight conditions.

Keywords: Artificial Intelligence Systems, Control Theory and Applications
Abstract: Toward the realization of control design support using Multimodal Large Language Model (MLLM), this study proposes a method in which the MLLM estimates a transfer function from a Bode plot image. The proposed method consists of two stages. In the first stage, the MLLM extracts frequency-response data points from a Bode plot image. In the second stage, the extracted data are provided to another LLM to estimate a transfer function and to evaluate the errors with respect to the input data. Using the same points extracted in the first stage, we also compare the estimation by MATLAB. The experiments showed that direct estimation from an image achieved high accuracy, whereas the two-stage method further reduced the estimation error for systems with closely spaced poles and zeros, nonminimum-phase systems, and high-order systems. The orders of estimated transfer function by the LLM often matched the true orders, whereas the orders of estimated transfer function by MATLAB tended to be higher than the true orders. Although the two-stage method increases the computational cost, the accuracy could be improved for the transfer functions with complex structures, because the search uses the explicit intermediate representation and the self-evaluation of fitting errors.

Keywords: Artificial Intelligence Systems, Human-Robot Interaction, Robot Mechanism and Control
Abstract: Robotic manipulation policies trained via imitation learning, such as Action Chunking with Transformers (ACT), can achieve strong performance under ideal conditions but often remain sensitive to small execution errors and distribution shifts. Correcting these failures typically requires dataset aggregation and full-policy retraining, which is computationally expensive and unsuitable for real-time deployment. In this work, we propose On- line Residual Policy Adaptation (ORPA), a framework that enables immediate, feedback-driven correction of robot actions without modifying the underlying policy parameters. ORPA augments a pretrained control policy with a lightweight, feedback-conditioned module that predicts residual adjustments directly in joint space, allowing the system to adapt its behavior at runtime. We evaluate ORPA on a set of precision-sensitive manipulation tasks using the ALOHA platform, demonstrating improvements in success rate and recovery from small perturbations compared to baseline control policies and rule-based inverse kinematics corrections.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control, Process Control Systems
Abstract: This paper presents a proactive inventory risk intelligence system for smart manufacturing comprising four layers: (1) MRUPS-F, a knowledge-driven risk scoring engine integrating seven risk dimensions and 25 canonical parameters; (2) a nine-category risk-type taxonomy mapping dimension profiles to operationally distinct patterns; (3) a rule-based action recommendation policy; and (4) a structured explanation layer generating actionable risk alerts. The scoring engine supports staged deployment: zero-shot (AUC = 0.618-0.630, no labels), direction-calibrated (AUC = 0.851), and fully calibrated (AUC = 0.883 +/- 0.004) on 242,075 real backorder records. Because these modes were measured under three different resampling protocols, every reported figure is labelled with the protocol that produced it, and no cross-protocol ratio is claimed. Cross-dataset evaluation reveals weight non-transferability across disruption types, a boundary condition at low parameter coverage where a single-variable heuristic outperforms the composite score, and an ablation in which purchase-order (PO) instability does not contribute on real data. We report these negative results in full, since they delimit where the framework should and should not be deployed.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control, Robot Vision
Abstract: Real-world agricultural inspection cameras rarely operate under ideal conditions. Illumination fluctuates throughout the day, lenses accumulate dust and condensation, and fruits arrive at slightly different orientations on the conveyor—all without triggering any alarm. Yet these subtle imaging shifts are enough to erode the performance of state-of-the-art anomaly detection systems that were trained on clean, controlled imagery. In this paper we ask: how much does each type of imaging degradation actually hurt detection accuracy, and can we mitigate it without retraining? To answer this, we benchmark three established methods—PatchCore (multi-view), PatchCore (single-view), and PaDiM—across 15 systematically controlled evaluation conditions on a mango surface-defect dataset (13 imaging degradation conditions, one clean reference, and one semantic-shift probe). Every baseline suffers measurable AUROC drops, with the sharpest declines under surface blur and rotation (>0.15 AUROC points). We then propose DA-PatchCore-V2, which analyses each incoming image, identifies its likely degradation type, and corrects it before feature extraction. DA-PatchCore-V2 reduces the worst-case AUROC drop to just 0.022 —an 86.0% improvement over the strongest clean-image baseline (PatchCore-SV)—while keeping clean-image accuracy within 0.01 AUROC of the top competitor.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control, Robotic Applications
Abstract: Computing power that used to be available only in supercomputers decades ago especially their parallelism is currently available in standard personal computer CPUs. We show how to effectively utilize the computing power of modern multi-core CPUs to solve the complex combinatorial problem of object arrangement and scheduling for sequential 3D printing. We achieved this by parallelizing the existing CEGAR-SEQ algorithm that solves the sequential object arrangement and scheduling by expressing it as a linear arithmetic formula. The original algorithm uses an object arrangement strategy that places objects towards the center of the printing plate. We propose alternative object arrangement strategies such as placing objects towards a corner of the printing plate and scheduling objects according to their height. Our parallelization is done at the high-level - we execute instances of the CEGAR-SEQ algorithm in parallel with a portfolio of object arrangement strategies, an algorithm called Portfolio-CEGAR-SEQ. Our experimental evaluation indicates that Portfolio-CEGAR-SEQ outperforms the original CEGAR-SEQ when a batch of objects for multiple printing plates is scheduled, Portfolio-CEGAR-SEQ often uses fewer printing plates.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control, Sensors and Signal Processing
Abstract: Automated mango grading requires severity estimates that remain reliable when lesion boundaries, lighting, and pathogen appearance vary across fruit. This paper presents Disease-Aware Model Conditioning (DAMC), a DINOv2- based framework for four-level mango defect severity estimation. The method uses a disease-conditioned gate to modulate salient patch tokens before Multi-Layer Perceptron classification, while a PatchCore memory bank and physics-guided segmentation provide coverage-based label audit and expert validation. On a verified subset of SenMangoFruitDDS, the method reaches 95.68% four-class severity accuracy and 99.59% binary healthy/diseased detection under five-fold crossvalidation, together with substantial agreement against blinded expert grading (κ = 0.7653). Although the disease-aware head does not exceed a separately fine-tuned no-gate DINOv2 baseline in raw accuracy, the results suggest that it provides a more traceable decision pathway for controlled offline inspection.

Keywords: Autonomous Vehicle Systems, Robot Vision, Robotic Applications
Abstract: Robust multi-sensor SLAM requires continuous adaptation to unpredictable sensor degradation, yet existing systems detect failure only after tracking loss has already corrupted the state estimate. We present ProActive-SLAM, a proactive reactive framework that addresses this through two complementary mechanisms. First, self-supervised DINOv2 features combined with LSTM temporal modeling forecast visual tracking quality ten frames ahead, enabling anticipatory fusion weight adjustment before degradation manifests. Second, we reformulate LiDAR surface normal computation as Rayleigh quotient minimization, where normals emerge as minimum eigenvalue eigenvectors of local scatter matrices with automatic planarity validation (λ0/λ1 < 0.3), eliminating coordinate-frame dependency. Hessian eigenvalue decomposition monitors global pose observability, and both scales share a unified eigenvalue-ratio quality metric that drives continuous, smooth sensor reweighting. Experiments on challenging construction site sequences demonstrate 16.5% lower absolute pose error and 23.2% faster optimization over state-of-the-art baselines, with 100% completion rate where fixed-fusion methods fail catastrophically.

Keywords: Autonomous Vehicle Systems, Robotic Applications, Sensors and Signal Processing
Abstract: This paper presents a robust, tightly coupled LiDAR-inertial odometry framework for autonomous mobile robots in feature-degraded environments. A cascaded median-IIR-RLS filter is developed to suppress sensor noise and non-stationary vibrations, while a UKF-based attitude estimator with magnetometer compensation mitigates gyroscope drift and magnetic disturbances. The enhanced orientation is incorporated into manifold-based IMU preintegration, and the resulting inertial constraints are tightly fused with edge and planar LiDAR factors in a centralized factor graph. Extensive experiments on agricultural and public datasets demonstrate improved localization accuracy and robustness over state-of-the-art methods.

Keywords: Human-Robot Interaction, Robot Mechanism and Control, Robotic Applications
Abstract: Recent visuomotor policies have achieved strong performance in robotic manipulation. Specifically, diffusion policy further improves action generation in various tasks. However, the diffusion policy mainly relies on image-based observation features and does not explicitly learn a dynamics-aware representation that summarizes temporal scene progression or object interactions. Therefore, we propose a world-model-conditioned diffusion policy (WM-DP) for visuomotor manipulation. WM-DP extends the original diffusion policy by concatenating a latent feature generated by a world model. Through this design, we study whether a simple early-stage world model conditioning structure can improve real-world manipulation performance without adding architectural complexity. We evaluate WM-DP on real-world pick-and-place tasks using objects with different physical properties. We also evaluate the method using a deformable object whose graspable regions and appearance change depending on its orientation. The results show that our model improves both success rate and trajectory quality under observed object conditions. Overall, our work suggests that even a GRU-based world model can provide useful cues for physical interaction and improve diffusion policy in visuomotor manipulation.

Keywords: Robotic Applications, Navigation, Guidance and Control, Control Theory and Applications
Abstract: This paper addresses the shape-formation problem for a swarm of mobile robots, proposing an adaptive capacity-density control framework on distributed Laguerre diagrams to ensure equal per-robot coverage. Conventional controllers, such as Centroidal Voronoi Tessellations, optimize dispersion but fail to guarantee uniform cell areas, which is critical for balancing sensing, communication, and energy expenditure. To overcome this, we introduce a distributed architecture with two coupled mechanisms: (i) Inter-cell Capacity Balancing, which dynamically adjusts per-robot weights to drive cell areas toward equality; and (ii) Intra-cell Density-Driven Attraction, which uses a mass-biased target to actively guide robot motion toward larger neighboring cells. Furthermore, our framework seamlessly embeds Buffered Voronoi Cells (BVC) within the Laguerre diagram to guarantee collision avoidance. Supported by distributed adaptation laws, we rigorously prove the swarm's global convergence to equal-area centroidal Laguerre tessellations of the prescribed target shape. The framework's efficacy and real-time performance are successfully validated through real-world experiments on a swarm of 20 miniature mobile robots.

Keywords: Control Theory and Applications, Control Devices and Instruments
Abstract: In this paper, a generalized super-twisting algorithm-based sliding mode control (GSTA-SMC) with radial basis function neural network (RBFNN)-based disturbance compensation is proposed for permanent magnet synchronous motor (PMSM) positioning. To reduce the disturbance rejection burden of the sliding mode controller, the RBFNN estimates the lumped disturbance online and compensates for it in the control input. Simulation results show that the proposed method improves tracking performance and reduces chattering compared with conventional SMC-based controllers.

Keywords: Robot Mechanism and Control, Artificial Intelligence Systems, Robot Vision
Abstract: Diffusion Policy has emerged as a powerful framework for robot behavior cloning, enabling multimodal action distributions that capture diverse motion strategies. However, this multimodality introduces a critical failure mode we term Decision Wavering: at each control step, the policy independently samples a trajectory, risking abrupt mode switches between consecutive steps. This causes the robot to oscillate between conflicting modes, converging toward unintended intermediate behaviors rather than committing to a single coherent strategy. We propose KaDDP, an inference-time framework that resolves Decision Wavering without retraining. Given N candidate trajectories sampled from Diffusion Policy, we estimate their densities via Kernel Density Estimation with a Gaussian kernel. A Kalman-inspired prior then incorporates the previous step's mode, computing a posterior score that favors trajectories consistent with the currently committed mode. The trajectory with the highest posterior score is selected for execution. We evaluate KaDDP on the Push-T benchmark against Diffusion Policy, demonstrating reduced steps to task completion and improved success rate over the Diffusion Policy baseline.

Keywords: Robot Vision, Robotic Applications, Artificial Intelligence Systems
Abstract: This paper presents an action recognition dataset and a comparative evaluation of action recognition models for patrol robots operating in daytime and nighttime outdoor environments. Unlike fixed surveillance camera environments, patrol robots operate as mobile platforms; therefore, viewpoint changes, perspective variations, camera shake, and dynamic background changes continuously occur. Robust recognition performance is also required under diverse illumination conditions. In this study, a daytime and nighttime outdoor action recognition dataset reflecting real-world patrol robot operating scenarios was constructed, and the performance of video-based and skeleton-based action recognition models was compared. The experimental results show that the skeleton-based models generally outperformed the video-based model. In particular, the model that converts skeleton information into two-dimensional heatmap representations and learns spatio-temporal features achieved the best performance, with an accuracy of 0.95, an F1 score of 0.77, and a mean average precision (mAP) of 0.84. These results demonstrate that skeleton-based approaches can be effectively applied to real-time action recognition for patrol robots in daytime and nighttime outdoor environments.

Keywords: Robotic Applications
Abstract: This paper presents a rotation-aware extension of Continuous-Time Conflict-Based Search (CCBS) for multi-agent path planning in graph-based workspaces with heterogeneous robots. Conventional CCBS and its improved variants model each transition as a single linear edge traversal and detect conflicts solely between linear motion segments. This abstraction underestimates arrival times and misses collisions during in-place turns, which is critical when robots have non-negligible angular inertia or low angular speed. The proposed method decomposes each graph-edge transition into (i) an in-place rotation phase and (ii) a linear translation phase, and integrates both phases into the SIPP low-level planner and the CBS high-level conflict resolution. The rotation time depends on the heading difference between consecutive edges and on the agent-specific angular velocity w_(k ); the linear time depends on the edge weight and the agent-specific linear velocity v_k. Conflict detection is extended to cover move–move, rotate–move, move–rotate, and rotate–rotate pairs, ensuring safety during turning. Because v_k and w_(k )are per-agent parameters, the framework naturally accommodates fleets of robots with heterogeneous kinematic capabilities. Experiments on graph-based roadmaps with ten agents show that the rotation-aware formulation reduces the number of high-level CT expansions by 47.2% and low-level searches by 54.7% compared with the original CCBS, at the cost of a 3.7% increase in flowtime that reflects the physical realities of robot turning.

Keywords: Control Theory and Applications, Robotic Applications, Artificial Intelligence Systems
Abstract: Repetitive industrial manipulation tasks such as pick-and-place, polishing, and assembly demand high precision under stochastic disturbances. Classical iterative learning control (ILC) exploits trial-to-trial repetition but is brittle under non-repetitive disturbances, while reinforcement-learning (RL) residual policies often ignore the trial-domain error structure that is the dominant signal in repetitive operation. This paper proposes Iteration-Domain Residual ILC (IDR-ILC), a Soft Actor-Critic (SAC) policy conditioned on the previous trial's error trajectory at the same time index, generating a per-step adaptive update to a stored ILC feedforward. A saturated accumulator and an error-multiplicative action structure make the residual vanish at convergence. A emph{contraction filter} (IDR-ILC-CF) projects each trial-domain accumulator update onto the set {Deltadelta : |Deltadelta|_F le rho |e_k|_F}, giving a sufficient condition for monotonic iteration-domain contraction. On a 7-DOF Franka Panda simulation under random payload, IDR-ILC reaches final-trial joint-space RMSE of 0.0202~rad (sinusoidal) and 0.0206~rad (Lissajous), strictly dominating PD and a same-budget Residual-PPO (paired t-test p<0.05) and matching PD-ILC and RLILC in mean. IDR-ILC-CF trades a 10% mean increase for a 3times reduction in cross-seed standard deviation. At a matched 8000-step budget, memoryless Residual-PPO is worse than PD, identifying previous-trial error conditioning as the decisive signal at this data scale.

Keywords: Artificial Intelligence Systems, Robotic Applications, Control Devices and Instruments
Abstract: Snake robots achieve locomotion through continuous, anisotropic frictional contact distributed over the whole body, which makes a single omnidirectional reinforcement learning (RL) policy difficult to train. In this paper, we take a first step toward omnidirectional snake robot control by decomposing the control problem along the direction of travel. We first calibrate the contact parameters of a MuJoCo simulation against drop-test experiments on five real surfaces, so that both the coefficient of restitution and the peak impact force are reproduced. On this calibrated model, we randomly sample compound-serpenoid gait parameters and analyze the resulting motion in terms of forward, lateral, and yaw velocity, identifying representative parameter sets that produce effective motion in each of eight principal directions. A torque-based Motion Matrix derived from each parameter set is then used as a morphological prior that constrains the action space of a direction-specific PPO policy. Simulation results show that the learned policies reliably generate motion in their assigned directions and outperform a conventionally tuned open-loop serpenoid controller in both path straightness and average forward speed.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: Many factories require high precision in assembly operations. Errors in assembly lead to defective products and result in consumer complaints. Therefore, it is essential to detect such errors as early as possible. In this paper, we use image processing to identify these errors. The correct sequence of assembly parts is known in advance. Each step of the assembly process is identified and checked against the correct sequence in real time. If an incorrect sequence is detected, an alarm sounds immediately. The correct sequence and components are also displayed. For image processing, a CNN model is trained to detect different components, their positions, and the correct sequence. This system requires only a monocular camera, and processing can be performed on an inexpensive computing board.

Keywords: Artificial Intelligence Systems, Biomedical Instruments and Systems
Abstract: Veno-arterial extracorporeal membrane oxygenation (VA-ECMO) drives oxygenated blood retrograde into the descending aorta, where it meets the antegrade output of the native heart and forms a watershed (mixing) region whose location decides which organs receive oxygenated flow---the basis of differential hypoxia (Harlequin syndrome). Patient-specific computational fluid dynamics (CFD) resolves the watershed but costs hours to days per case. We cast watershed prediction as point-cloud field regression: a PointNet++ surrogate (set-abstraction encoder, feature-propagation decoder) predicts the per-node retrograde-flow fraction throughout the aortic volume from patient geometry and boundary conditions, with no CFD field as input. As each case labels ~10^6 points, the field formulation is far more data efficient than scalar surrogates. On ten patient-specific cases under case-level leave-one-out cross-validation, it attains a watershed IoU of 0.81, an ECMO-perfused-region error of 7.2%, and a watershed-level error of 20mm.

Keywords: Artificial Intelligence Systems, Industrial Applications of Control, Human-Robot Interaction
Abstract: Skilled human operators remain indispensable in safety-critical industrial environments despite advances in automation and AI. Existing industrial AI frameworks mainly rely on perception-to-control pipelines or state-to-action learning, without explicitly modeling the cognitive processes underlying expert operation. To address this gap, this paper introduces Human Tacit Intelligence (HTI), a conceptual framework that represents expert intelligence as a hierarchical process integrating perception, situation awareness, tacit memory, operational reasoning, decision, and action. We propose an HTI cognitive pipeline together with a five-layer representation hierarchy spanning raw sensor signals, environment states, operation primitives, operational skills, and tacit intelligence. Rather than presenting a task-specific learning algorithm, this work establishes a foundational representation framework for future HTI datasets, learning architectures, explainable reasoning models, foundation models, and industrial Physical AI. HTI is expected to provide a unifying paradigm for capturing and learning expert intelligence in safety-critical industrial domains.

Keywords: Artificial Intelligence Systems, Information and Networking, Robotic Applications
Abstract: Accurate inertial and actuator parameters are important for transferring dynamic legged-robot controllers from simulation, but collecting informative identification data on hardware remains difficult. Classical optimal excitation first selects a finite-dimensional trajectory family and then optimizes its coefficients, thereby restricting the reachable motions. This paper presents LEXI, a reinforcement-learning framework that generates excitation through the robot's joint-target interface while maximizing a natural-metric-normalized Fisher information matrix (FIM). A recurrent policy conditions on causal proprioception and the accumulated FIM spectrum, and is trained with physical randomization and safety constraints. On a Raibo2 leg mounted on a rigid jig, a 30-s LEXI trajectory increases the realized FIM log determinant from 183.3 for a D-optimal ellipsoidal baseline to 193.4, while increasing the minimum eigenvalue from 0.060 to 0.101. Across 100 randomized simulation trials with known parameters, LEXI reduces base-parameter identification error from 0.305pm0.184 to 0.185pm0.137. The identified model also yields the smallest sim-to-real stride and gait-cycle errors after retraining a 4-m/s locomotion policy.

Keywords: Artificial Intelligence Systems, Civil and Urban Control Systems, Multimedia Systems
Abstract: Understanding the progression of potential hazards is critical for proactive facility management and the robust training of physical AI systems. However, real-world disaster footage is inherently scarce, making it exceedingly difficult to simulate such extreme edge cases. Furthermore, simple supervised training approaches for risk simulation are fundamentally limited by this severe lack of long-tail data. To overcome these data bottlenecks without relying on manual labor, this work outlines an automated, multi-stage generative pipeline for synthesizing highly realistic disaster simulation videos from a single still image. Anchored by an automated privacy-preserving module to securely sanitize real-world inputs, our approach orchestrates a deeply integrated pipeline where VLM-driven zero-shot risk analysis semantically guides a cohesive cascade of open-vocabulary segmentation, high-fidelity image inpainting, and state-of-the-art Image-to-Video (I2V) generation. The resulting pipeline generates context-aware, temporally consistent simulations of hazards such as fire, flood, and structural damage. Empirical evaluation demonstrates that this plug-and-play design requires no domain-specific fine-tuning or dataset accumulation, offering a scalable solution for automated risk visualization and data augmentation in extreme scenarios.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Autonomous Vehicle Systems
Abstract: This paper addresses vision-language model (VLM)-based road-hazard reasoning in autonomous driving. A perception module summarizes the weather condition into a textual weather token, which is injected into the VLM together with the image. However, weather classification often fails in adverse weather, VLMs tend to over-trust text that contradicts the image, and fine-tuning on always-correct tokens further reinforces this tendency. Given a wrong token, the model thus explains the hazard based on the wrong weather instead of the actual one. To alleviate this problem, this paper applies counterfactual token augmentation (CTA), which duplicates a small fraction of the training pairs, changes only the token of each copy to a different weather, and keeps the ground-truth explanation image-grounded. Experiments show that a small fraction of mismatched pairs largely removes the token-copying bias and grounds the explanations in the actual weather, while preserving the benefit of fine-tuning. This yields VLM-based road-hazard reasoning that is robust to incorrect weather information.

Keywords: Artificial Intelligence Systems, Robot Mechanism and Control
Abstract: Recently, unmanned inspection methods utilizing mobile manipulators have been employed for onboard inspections of ships during operation. Conventional manipulator control methods have relied on expertise in kinematics. More recently, control methods based on Large Language Models (LLMs), which enable communication via simple natural language have gained attention. However, despite the limitations of LLMs in hallucination and computational costs, few studies have directly compared control performance and stability according to the level of LLM intervention. This study investigates the appropriate level of LLM intervention required to achieve optimal control performance and quantitatively compares two control strategies: (1) a method where the LLM performs command interpretation, coordinate estimation, and joint angle calculation, and (2) a method where the LLM is used solely as a natural language command interpreter, while control of manipulator is performed using inverse kinematics-based reinforcement learning. As a result of verifying the Euclidean distance error between the actual target point and the end-effector through simulation, the LLM-only control method showed an error of 0.1654 m, while the hybrid control method showed an error of 0.0023 m. In the future, the two control methods will be implemented on physical manipulator, and additional analyses of error factors will be conducted to evaluate their practical applications.

Keywords: Artificial Intelligence Systems, Human-Robot Interaction, Robot Mechanism and Control
Abstract: This study presents the development of an AI-based humanoid surgical assistant platform designed to support both open and laparoscopic surgeries. The platform integrates Physical AI technologies to enable context-aware autonomous decision-making, surgical assistance, situational awareness, and risk response. It provides three operational modes: Passive Control, Autonomous AI, and Tele-control/Tele-mentoring, while also functioning as an integrated operating room support system that monitors patient physiological signals and intraoperative conditions. To develop and validate AI models, multimodal clinical data—including surgical field videos, operating room videos, audio recordings, and physiological signals—were prospectively collected and integrated into a comprehensive surgical dataset. Procedure-specific surgical videos were acquired from colorectal, otorhinolaryngology, and thoracic surgeries. The collected data were synchronized and preprocessed to support the development of a surgery-specific Vision AI framework and Surgical Large Language Model (LLM), incorporating real-time video enhancement, 2D–3D mapping, surgical instrument recognition, and contextual scene understanding. The proposed platform establishes the foundation for an intelligent humanoid surgical assistant by integrating hybrid compliance control, Vision AI, Surgical LLM, and multimodal data analysis. Future work will focus on enhancing physiological monitoring and developing predictive AI models capable of identifying critical intraoperative events, thereby advancing the system toward practical clinical deployment.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Robot Vision
Abstract: The existing self-reconstruction-based approaches for enhancing tiny object representations suffer from an objective conflict problem, where the optimizing the reconstruction included tiny object regions conflicts with the goal of generating dis-criminative spatial priors through the difference map. We propose the inpainting-based constrained reconstruction (ICR) framework for improving discriminative difference maps. To address this, we propose the ICR framework, which generates pseudo-background training set by replacing tiny object regions to background pixel via an inpainting model. Experiment on the VisDrone2019 dataset demonstrates that the proposed ICR framework produces more discriminative difference maps in tiny object regions throughout training, confirming that the spatial prior quality is effectively maintained.

Keywords: Artificial Intelligence Systems, Sensors and Signal Processing, Robotic Applications
Abstract: Deploying an autonomous agent on a new hardware platform typically requires manually specifying what each sensor channel measures. To examine whether pretrained models can perform this step without manual labeling, we introduce a vision-grounded benchmark in which video-capable multimodal large language models (MLLMs) assign unlabeled sensor streams to a provided list of sensors using synchronized third-person video. The benchmark is designed so that the stream-to-sensor mapping cannot be determined from signal structure alone, making visual grounding necessary. We conduct a preliminary evaluation of three open-weight video-capable MLLMs under a ten-episode budget. None achieves confirmed complete identification, and repeated passive observations do not produce stable correction of incorrect mappings.

Keywords: Artificial Intelligence Systems, Human-Robot Interaction, Sensors and Signal Processing
Abstract: Teleoperation is a widely used method for collecting demonstration data to train Vision-Language-Action (VLA) models, particularly on humanoid hardware. Such data is commonly generated by capturing human motion with hand-held VR controller setups, which typically provide only wrist poses and therefore cannot represent the operator's full-body and hand configuration. In this paper, we present a teleoperation pipeline that captures the operator's whole-body motion with a wearable Inertia Motion Capture sensor and finger motion with motion-capture gloves, and retargets the measured motion to a humanoid model in the MuJoCo simulator through a learned whole-body control policy.

Keywords: Artificial Intelligence Systems
Abstract: Agricultural domain knowledge is scattered across unstructured documents, making it challenging to construct a highly reliable QA dataset. To address this issue, we propose a pipeline for constructing a context-based agricultural QA dataset. Furthermore, we systematically optimized a RAG pipeline using the constructed QA dataset, improving the performance of an agricultural consulting system. RAG optimization was performed for both the retriever and generator, with each module evaluated using the F1 score. The final RAG pipeline, combining the optimal retriever (ColBERT) and generator (Qwen2.5-32B), achieved a 0.118 increase in BERTScore over the LLM-only baseline, corresponding to an 18.1% relative performance improvement.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Industrial Applications of Control
Abstract: Aiming at the input saturation issue during the emergency braking of urban rail trains, this paper constructs a nonlinear braking dynamic model considering actuator saturation to analyze its adverse effects on braking performance. A funnel-based arctangent anti-windup control algorithm is proposed in this work. By virtue of predefined time-varying error boundaries and adaptive compensation modules, the developed algorithm achieves high-precision trajectory tracking and rapid braking force distribution subject to saturation constraints. This scheme effectively suppresses the response delay and error divergence induced by input saturation, while guaranteeing the global uniform boundedness of the closed-loop system via rigorous stability analysis. Numerical simulation results demonstrate that the tracking error is always confined within the preset funnel boundary and converges to the predefined accuracy level within 10 seconds. Meanwhile, the control input complies with the output limitation of actuators without saturation overflow. The proposed control strategy substantially enhances stopping accuracy, dynamic response performance and operational stability. It further provides solid theoretical basis and technical reference for the high-safety and high-precision emergency braking control of urban rail trains in the presence of input saturation.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: Leader-follower formations involve role assignment and relative state tracking, which makes them fragile near clutter. A leader can brake, turn around an obstacle, or enter a narrow passage while followers still track fixed offsets. The nominal field may then point into a closing gap or toward a road edge. To address this issue, this paper combines a sampled oriented-bounding-box CBF-QP check with a rollout verification before action execution, and proposes a risk-adaptive spatiotemporal artificial potential field (APF) scheme, which provides benchmark evidence for the implemented safety stack by achieving zero collisions across 1,650 rigorous simulation runs.

Keywords: Autonomous Vehicle Systems, Robot Mechanism and Control, Robotic Applications
Abstract: Forestry robots must operate on steep, soft, and irregular terrain while supporting a heavy working arm and payload; therefore, the structural reliability of the lower mobile platform is essential. This study presents a numerical evaluation of the structural safety of a lower mobile platform for a forestry robot. A three-dimensional finite element model based on the actual platform geometry was constructed, and static, modal, impact, and frequency response analyses were conducted using Midas NFX 2016 R1. The applied load included the engine room, working arm, and payload, corresponding to a total concentrated load of 14,000 kgf (137.3 kN) at the arm joint. The criteria were a safety factor above 3 based on yield strength and a first natural frequency above 10 Hz for the main frame excluding tires. In the static analysis, the maximum Von-Mises stress was 262.2211 MPa at the hip joint, giving a safety factor of 3.23. The first natural frequency of the frame without tires was 12.317 Hz. Under impact and frequency excitation, the minimum safety factors were 3.005 and 3.426, respectively. The results demonstrate that the proposed platform satisfies the structural safety requirements under the considered operating conditions.

Keywords: Autonomous Vehicle Systems, Robot Vision, Navigation, Guidance and Control
Abstract: Fusing radar, LiDAR, and camera enables robust perception in diverse and adverse conditions, but the fusion performance critically depends on accurate extrinsic calibration among the three sensors. In this paper, we address the problem of online target-less extrinsic calibration for the radar-LiDAR-camera system. Existing target-less methods are mostly designed for a single sensor pair, and composing the pairwise results does not guarantee consistency across the three sensors. Moreover, the sparse and noisy radar measurements make the radar-involving pairs unreliable. To tackle these challenges, we propose a joint calibration framework that constructs residuals for each sensor pair and optimizes the extrinsics of all pairs together to minimize the overall residual. Furthermore, we introduce an adaptive radar noise filter that rejects spurious radar returns using a range-dependent margin, and a correspondence accumulation strategy that aggregates sparse radar correspondences over frames. We validate our method on an in-house radar-LiDAR-camera dataset covering diverse urban environments, where it reduces calibration errors across all sensor pairs over a state-of-the-art camera-LiDAR baseline.

Keywords: Autonomous Vehicle Systems, Robot Vision, Artificial Intelligence Systems
Abstract: Visual SLAM systems inevitably accumulate drift during long-term operation, and loop closure is essential for correcting this drift through pose graph optimization. However, in sparse-loop indoor environments such as long corridors and large-scale workspaces, loop events occur infrequently, making each detected loop highly important for trajectory correction. This paper proposes LACE, a loop-constraint augmentation framework with reprojection-guided correspondence and edge selection for visual-inertial pose graph optimization. Instead of relying on a single constraint from a loop event, LACE expands retrieved loop candidates using temporally neighboring keyframes to generate multiple candidate loop edges. To suppress false-positive constraints, feature correspondences are verified through Top-K descriptor matching followed by reprojection-error-based selection. Each candidate loop hypothesis is then estimated using PnP-RANSAC, and high-quality loop edges are selected based on inlier support and median reprojection error before being inserted into pose graph optimization. Experiments on KAIST indoor sequences and the EuRoC Vicon Room dataset show that LACE achieves the lowest average ATE RMSE across all evaluated settings and consistently outperforms the Expansion-only variant.

Keywords: Autonomous Vehicle Systems, Robotic Applications
Abstract: This paper presents a multi-robot coverage method that minimizes mission completion time by balancing workloads among robots. The proposed method clusters adjacent straight-line coverage paths in a polygonal environment and assigns each cluster to a robot. We adopt a two-phase approach: (i) polygon decomposition to generate turn-minimizing straight-line paths, followed by (ii) graph-based path clustering to distribute the workload evenly. Evaluated over 100 randomly generated polygonal environments with 2--8 robots, the method achieves lower average mission time than path-assignment baselines while significantly reducing planning computation time. Additional simulations with up to 100 robots and a benchmark environment confirm scalability and effectiveness for heterogeneous-speed teams.

Keywords: Autonomous Vehicle Systems, Robot Vision, Sensors and Signal Processing
Abstract: Vegetation-rich off-road driving requires perception outputs that indicate cell-level traversability rather than only fine-grained semantic categories. This paper presents a LiDAR-only BEV traversability estimation framework that predicts four classes: passable, caution, unknown, and blocked. The method estimates ground-relative height, constructs a 44-channel multi-height BEV tensor over an 80 m × 80 m area, and augments the single-frame representation with local temporal summary channels. The input features encode height-bin density and occupancy, remission and range statistics, ground-relative height statistics, vertical structure, ground support, and repeated temporal evidence. Preliminary experiments on vegetation-rich GOOSE validation scenes show strong class imbalance, with blocked cells occupying only 1.56% of labeled BEV cells after ambiguous unknown cells are excluded. The full model achieves 0.7751 mIoU, 0.8277 blocked recall, and a false-passable rate of 0.0006. A 2×2 ablation indicates that multi-height geometry is the main factor in reducing unsafe openings, while temporal summary provides complementary but not fully independent gains.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Robotic Applications
Abstract: This paper presents a simulation study of a Decentralized Graph Neural Network (DGNN) for task assignment in ROS2-based multi-robot systems. The DGNN is trained to imitate the solution of the centralized Hungarian algorithm for a task assignment problem, where each task consists of multiple waypoints. The DGNN has an encoder–GNN module–decoder architecture, and a total of 9 MLPs are trained independently of the number of agents and tasks. To handle cases where the number of tasks exceeds the number of agents, we train the model using a weighted binary cross-entropy loss. Simulation results demonstrate that the trained DGNN effectively imitates the centralized Hungarian solution. Its implementation in a ROS2 environment further confirms its applicability to asynchronous distributed task allocation.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Control Theory and Applications
Abstract: Landing a quadrotor on a moving target requires online replanning under uncertain target motion while maintaining tight final-approach geometry. Enforcing approach constraints, such as a glideslope cone, over the entire horizon is often infeasible from arbitrary initial conditions, whereas removing them entirely weakens the safety envelope near touchdown. This paper presents a receding-horizon landing planner that resolves this tension by embedding state-triggered constraints (STCs) in differential dynamic programming (DDP). The tightened approach bounds are activated only after the vehicle enters the final-approach region, and the optimizer determines when each constraint switches on. The STC residuals are accumulated through an RK4-integrated scalar state and enforced by a single terminal equality handled by the augmented-Lagrangian branch of constrained DDP. Across 100 randomized initial conditions, cSTC-DDP achieves 93 compliant landings (98 total) with a 0.95 s median cold full-horizon solve time, versus 86 landings and 4.05 s for OpenSCvx. Hardware experiments demonstrate off-board receding-horizon replanning at approximately 5 Hz (178–188 ms median solve time) under circular and random target motion.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications
Abstract: For distributed electric vehicles, multi-wheel tire-force inputs are accompanied by strong lateral–longitudinal coupling, complex multi-actuator constraints, and parameter uncertainties that affect tire-force allocation. To address these issues, a stochastic cooperative game-based collaborative decision-making method for multi-wheel tire forces considering cornering stiffness estimation errors is proposed. In this method, the longitudinal and lateral tire forces of the four wheels are treated as unified decision variables, while trajectory-tracking errors, vehicle stability requirements, and tire-force feasibility constraints are mapped into the tire-force decision space. Through probabilistic constraint transformation, the stochastic uncertainty induced by cornering stiffness estimation errors is converted into deterministic constraint bounds, which are further used to modify the tire-force feasible domain. On this basis, a stochastic cooperative game model is established to describe the cooperative relationship among front-wheel tracking correction, rear-wheel stability regulation, and all-wheel yaw coordination. The resulting distributed optimization problem is solved using an adaptive alternating direction method of multipliers. Simulation results demonstrate that the proposed method achieves coordinated multi-wheel tire-force allocation under complex constraints and balances trajectory-tracking accuracy, yaw stability, and tire-force feasibility while maintaining bounded primal and dual residuals within the prescribed co-simulation iteration budget.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control
Abstract: This paper presents an enhanced hybrid control architecture for quadrotor Unmanned Aerial Vehicles (UAVs) designed for robust path following and obstacle avoidance in cluttered environments. While Reinforcement Learning (RL) offers high-level adaptability, it often lacks the deterministic stability required for safety-critical flight. To address this, we propose an architecture that integrates a high-level RL agent with a traditional low-level cascaded PID controller. The RL agent is trained using a multi-objective reward function incorporating a potential field, which accounts for target attraction, path maintenance, and obstacle repulsion. This agent generates dynamic position displacement commands that are processed by the PID loops to maintain attitude stabilization and motor control. By offloading complex reactive guidance to the RL agent while retaining the high-frequency reliability of PID stabilization, the system achieves a balance between navigational intelligence and flight stability. The proposed model is validated through high-fidelity 3D trajectory simulations involving both static and dynamic obstacles. Preliminary analysis indicates that this hybrid approach significantly improves collision-avoidance rates and tracking precision compared to conventional end-to-end RL and pure PID methods, providing a scalable solution for autonomous UAV navigation.

Keywords: Autonomous Vehicle Systems, Information and Networking, Control Theory and Applications
Abstract: Vehicle platooning has been regarded as a cooperative driving strategy for connected and automated vehicles, since it can improve traffic efficiency by enabling automated vehicles to travel with small inter-vehicle distances. However, close formation driving makes the platoon vulnerable to string instability, where tracking errors propagate and amplify along the vehicle string. This paper investigates the error-L_2 string stability of passive vehicle platooning systems under a multiple-predecessor-following (mPF) topology and demonstrates that the error attenuation gain with respect to the first vehicle's error is bounded by 1/m. To validate the proposed analysis, a proportional-derivative controller is designed for a linearized longitudinal vehicle model to ensure passive velocity-output dynamics. Numerical simulations under varying numbers of predecessors demonstrate that the maximum error attenuation ratio decreases as m increases, consistently remaining below the theoretical bound.

Keywords: Autonomous Vehicle Systems, Sensors and Signal Processing, Control Devices and Instruments
Abstract: Frequency-modulated continuous-wave (FMCW) LiDAR precisely measures range using optical interference, offering robustness against external noise and sunlight in autonomous driving applications. Its ranging accuracy depends heavily on the optical frequency sweep linearity of the laser source. However, because the laser diode (LD) driver exhibits unpredictable variations due to temperature changes and aging, maintaining consistent linearity is challenging. To address this issue, this paper proposes a closed-loop control method that uses optical power as a feedback signal instead of directly measuring the optical frequency, which is difficult to access in the THz range. In the initial stage, a frequency-to-voltage converter (FVC)-based loop extracts a pre-distortion signal that ensures a linear sweep, while the synchronized ideal optical-power profile is stored as the reference profile. In the operation stage, this pre-distortion signal is applied to the LD driver and the closed-loop controller tracks the reference profile using the real-time optical-power signal, thereby compensating for time-varying drift in the sweep. Experimental results verify that the proposed method maintains sweep linearity under thermal variation during continuous operation, significantly improving the ranging performance of the LiDAR system.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Robotic Applications
Abstract: This paper proposes a time-delay-estimation-integrated model predictive control (TDE-MPC) framework for robust trajectory tracking of wheeled mobile robots under lumped uncertainties. A simplified nominal model is used for MPC prediction, while model mismatch, friction, and external disturbances are treated as lumped uncertainties. The uncertainty term is estimated using time-delay estimation and directly incorporated into the error-state MPC prediction model as an additional compensation term. To improve recursive feasibility and practical closed-loop stability, terminal ingredients are introduced, including a terminal cost and a tightened terminal set. The proposed method is evaluated through trajectory tracking simulations of a differential-drive mobile robot and compared with a nominal MPC that does not include the TDE term. Simulation results show that the proposed TDE-MPC improves position and heading tracking performance under disturbed conditions. These results demonstrate that the proposed approach enhances robustness while preserving the constraint-handling capability of MPC.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control, Sensors and Signal Processing
Abstract: This paper presents LiDAR-based real-time obstacle avoidance and pre-docking dynamic positioning (DP) for autonomous berthing of an overactuated unmanned surface vehicle (USV). During waypoint following, 3D LiDAR point clouds are projected onto a two-dimensional plane, where candidate trajectories are evaluated for collision risk to determine an avoidance heading correction. During pre-docking DP, a position controller maintains the vehicle’s position and heading. The docking approach begins only if the docking target is detected at the current sample, the instantaneous heading error is within its threshold, and the windowed position-stability condition is met. The methods were integrated into a five-phase autonomous berthing sequence and validated through three outdoor basin trials at distinct target berths, demonstrating obstacle avoidance during transit and stable pre-docking positioning.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Artificial Intelligence Systems
Abstract: This paper presents a simulator-in-the-loop study of model predictive control (MPC) weight calibration for autonomous vehicle path-following using a high-fidelity closed-loop simulation environment. With the controller structure fixed, the optimization target is a four-dimensional MPC weight vector for lateral tracking, heading response, steering effort, and steering smoothness. Each candidate is evaluated by a simulation objective that penalizes road departure, counted cone contacts, and tracking error. Under a common 50-trial budget, the study compares Latin-hypercube sampling, random search, Bayesian optimization (BO), and LLM-based search over five independent repetitions. The results show complementary behavior: LLM-based search reaches hit-free candidates early in four of five repetitions, while BO finds the lowest objective value when its exploration reaches a useful region. A 4096-sample Sobol-based landscape analysis finds only seven hit-free samples, indicating that feasible weight regions are sparse. These results suggest that LLM-based search is effective for rapid feasibility seeking, whereas numerical exploration remains important for discovering less intuitive weight regions.

Keywords: Autonomous Vehicle Systems, Navigation, Guidance and Control
Abstract: This paper presents an MPC-based lateral path-tracking controller combined with an unscented Kalman filter (UKF)-based equivalent front and rear lateral force estimator. The proposed method aims to improve path-tracking performance under varying road-friction conditions. A 2-DOF bicycle model is used to describe the vehicle lateral dynamics, and lateral and heading errors are added to the model for path tracking. A UKF-based estimator is designed to reduce the model mismatch caused by lateral-force variations. It estimates the equivalent front and rear axle lateral forces using the measured yaw rate, lateral acceleration, and yaw acceleration. The estimated forces are incorporated into the MPC prediction model to reflect changes in the vehicle lateral dynamics. The proposed method is evaluated in a CarSim–MATLAB/Simulink co-simulation environment. The estimator achieves an average coefficient of determination of approximately 0.99. To validate the proposed MPC-based path tracking controller, an ISO 3888-2 double lane change maneuver is conducted under laterally varying road friction. Compared with conventional MPC, the proposed controller reduces the RMS and peak lateral position errors by 45.1% and 74.2%, respectively.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Control Theory and Applications
Abstract: This paper proposes a deterministic expert interpolation layer for dead-reckoning-based vehicle localization under Global Navigation Satellite System (GNSS) outage conditions. Dead reckoning can maintain continuous localization using onboard vehicle motion signals; however, small yaw-rate errors accumulate through recursive heading integration and progressively increase trajectory deviation. Moreover, the influence of the accumulated yaw error on the trajectory varies with the vehicle motion state, which may limit the ability of a single-output prediction structure to sufficiently represent different driving regimes. To address this problem, four parallel experts generate candidate yaw-error predictions from a shared temporal representation, and their outputs are combined using deterministic interpolation weights computed from the current state in the velocity and yaw-rate space. The resulting yaw-error estimate is used to compensate the dead-reckoning heading and reconstruct the vehicle trajectory. Real-vehicle experiments were conducted under low-speed and high-speed GNSS outage scenarios to analyze variations in expert contributions according to the vehicle motion state and to compare the reconstructed vehicle trajectories with those obtained using other methods. The experimental results show that the proposed expert interpolation layer adjusts the expert combination according to different driving regimes and reduces yaw error and trajectory deviation compared with conventional dead reckoning and a single-output prediction model.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: This paper proposes a Non-Line-of-Sight (NLoS) object estimation method that leverages a road convex mirror with camera and 3D Light Detection And Ranging (LiDAR) sensor fusion. The approach begins with identifying NLoS objects through a convex mirror and utilizes the virtual LiDAR point cloud formed by the reflector for NLoS object estimation. A reflection error compensation model is designed based on the correlation between the reflection region on the reflector and the virtual point cloud recovery error. We perform object data association and fusion to mitigate estimation ambiguities caused by simultaneously acquired LoS and NLoS sensing data for the same object and discontinuities from changes in sensing path. The resulting data is then processed by an Interacting Multiple Model-Kalman Filter (IMM-KF), which incorporates the different error characteristics of LoS and NLoS sensing data. Scenario-based experiments demonstrate that the proposed method outperforms conventional methods in NLoS object estimation and shows stable estimation performance in environments where LoS and NLoS sensing data coexist and the sensing path changes over time.

Keywords: Autonomous Vehicle Systems, Robot Vision, Artificial Intelligence Systems
Abstract: Three-dimensional (3D) map generation is a core component of autonomous driving, yet recovering an accurate static background in urban environments crowded with dynamic objects is difficult in dynamic urban scenes. Existing approaches fall broadly into two categories. The 3R-family methods—such as DUSt3R, MASt3R, and MonST3R—operate on RGB images alone, but their ability to handle dynamic objects is only auxiliary or partial. Sensor-fusion-based methods, such as AD-GS and IDSplat, achieve high-quality static–dynamic separation, but they presuppose demanding sensor requirements, relying on LiDAR depth and GPS position information. In this paper, we propose a Dual-Track Gaussian Splatting pipeline that narrows the gap toward sensor-fusion based urban 3D reconstruction using only a monocular RGB camera. The proposed method predicts monocular depth and estimates ego-motion with a deep-learning-based SLAM that is robust to dynamic objects. Moving vehicles are identified by combining YOLO-based vehicle segmentation with optical flow, and they are stably tracked using a consistent-SE(3) tracking scheme. Experiments on VKITTI show that the proposed RGB-only pipeline improves PSNR from 14.43 dB and 14.06 dB for MASt3R and MonST3R to 18.40 dB. On nuScenes, replacing the GPS/IMU ego-pose with the proposed camera-only estimation costs only 0.13 dB in held-out PSNR (19.53 dB → 19.40 dB), and qualitative results further show that the method suppresses vehicle-induced ghosting artifacts without using LiDAR or GPS.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: We assess the feasibility of imitation learning (IL) for autonomous trajectory tracking. We construct a hybrid expert combining LQR (lateral) and constrained MPC (longitudinal) — each expert’s strength in its own domain — and distill it into a single LSTM (Long Short-Term Memory) Hybrid neural network (NN) (1.12M params) that outputs a vehicle-agnostic curvature target and longitudinal acceleration; a per-vehicle executor then produces the actuator steering angle. To remove the hidden bias of tuning the reference trajectory to one expert’s envelope, we restore the LQR-optimal envelope to the trajectory generator and align the MPC’s constraints to it. The distilled NN inherits LQR’s lateral precision and MPC’s longitudinal collision-avoidance behavior in a single policy, deploys across multiple electric-vehicle platforms without re-training, and validates planner abstraction through a 12–23% drop in Hard collisions with no architecture change. Residual gaps on unseen distributions (cross-map, narrow lateral coverage, simulator dynamics) are quantified and addressable via label smoothing, augmentation, and multi-town data.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Artificial Intelligence Systems
Abstract: This paper proposes a Transformer-based future parameter-scheduled linear parameter-varying (LPV)-MPC framework for autonomous vehicle lateral control. The Transformer predicts a future parameter for LPV-MPC, such as longitudinal velocity, using past parameter data, including velocity and road-geometry information. The predicted parameter is then used as a future scheduling input to construct the LPV-MPC prediction model over the control horizon. Since the optimization problem is still solved in the MPC layer, the proposed method preserves the constraint-handling capability of LPV-MPC while reducing the dependence on ideal future velocity information. The proposed method is evaluated in a time-varying autonomous-driving scenario including clothoid road-curvature variation and a single impulse disturbance. It is compared with LTI-MPC, ideal time-varying (TV)-MPC, and autoregressive integrated moving average model (ARIMA)-based future parameter-scheduled LPV-MPC. The test results show that the proposed method achieves lateral tracking performance close to the ideal TV-MPC.

Keywords: Autonomous Vehicle Systems, Industrial Applications of Control
Abstract: The effectiveness of active rear steering (ARS) control for four-wheel steering (4WS) vehicles is widely recognized in the automotive industry. At low speeds, ARS control can enhance maneuverability by steering the rear wheels in the opposite direction to the front wheels, reducing the turning radius. In contrast, at high speeds, ARS control can improve stability by steering the rear wheels in the same direction as the front wheels, preventing oversteer behavior. However, the performance of ARS control is often limited by the reference model used to generate the desired yaw rate, which is typically derived from a steady state of front-wheel steering (FWS) vehicle model. In this paper, we conduct a numerical analysis to construct an optimal yaw rate reference map for ARS control by formulating a constrained optimization problem. In the optimization problem, constraints are imposed to ensure that the vehicle operates within safe limits at steady state. Numerical simulations demonstrate the effectiveness of the proposed method in providing an optimal yaw rate reference map for ARS control.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems, Information and Networking
Abstract: Autonomous vehicles on complex urban roads must reason not only about geometry but also about the semantic meaning of the surrounding elements to decide what to do. We present a semantic behavior-reasoning approach in which a single Triplet Ontological Semantic Model (TOSM) representation describes environmental elements, their relations, and the ego-vehicle state over a semantic high-definition (HD) map. On this representation, Semantic Web Rule Language (SWRL) rules spanning behavior, regulation, event, and path reasoning infer context-appropriate maneuvers such as yielding, braking, and lane changes. To ensure correctness, the inference is cross-validated between the Pellet description-logic reasoner, which runs the SWRL rules over the Web Ontology Language (OWL) ontology, and an independent forward-chaining engine that re-implements the same rules. The two are required to produce the same conclusions. On a TOSM semantic HD map built from real Busan driving data, the two implementations agree on every behavioral conclusion across the evaluated scenarios, and the cross-validation surfaces an under-specified premise that a single reasoner would have hidden.

Keywords: Autonomous Vehicle Systems, Control Theory and Applications, Robotic Applications
Abstract: This paper presents a guide path-free multimodal trajectory planning framework that integrates model predictive control (MPC) with a turning circle-based control barrier function (TC-CBF). Unlike Euclidean distance-based CBFs that rely on proximity alone, the TC-CBF evaluates the clearance of the left- and right-turning circles set by the vehicle's maximum yaw rate, explicitly encoding feasible left- and right-side avoidance. This lets the optimizer discover topologically distinct trajectory modes within a single optimization layer, without a global reference path, mitigating the local-minimum and deadlock problems of standard MPC while remaining efficient. Deterministic and Monte Carlo simulations show higher success rates, better trajectory quality, and lower computation than ED-CBF-based MPC and guide path-based MPC.

Keywords: Autonomous Vehicle Systems, Sensors and Signal Processing, Robot Vision
Abstract: LiDAR sensors have been widely used in autonomous driving due to their ability to accurately perceive the surrounding environment. However, in adverse weather such as heavy rain, the point cloud is severely degraded by road spray, the water mist thrown up by the leading vehicle, which reduces the reliability of autonomous driving. While snow removal has been widely studied, road spray has received little attention. We observe that road spray, like snow, appears as low-intensity LiDAR returns, and therefore ask whether existing snow-removal filters that exploit this low- intensity cue can be transferred to road spray without any pipeline modification or parameter tuning. We benchmark four conventional intensity-based snow-removal filters, together with an adaptive, unsupervised pipeline (ASDR + C-GLP) that we previously developed, on the SemanticSpray dataset. The filters transferred to widely varying degrees, and even the best result was moderate: our previously developed pipeline achieved the highest F1-score while preserving the leading vehicle and running in real time. Low-intensity snow-removal filters thus transfer to road spray, but only partially.

Keywords: Autonomous Vehicle Systems
Abstract: Simultaneous Localization and Mapping (SLAM) based on a 3D map is essential for establishing exploration paths and monitoring environmental changes during indoor drone operations. This paper presents a LiDAR-based map reuse method for autonomous drone navigation and change detection in indoor environments. During the initial LiDAR-Inertial Odometry (LIO)-based SLAM process, DBoW descriptors are extracted from LiDAR frames and stored for subsequent use. During re-flight, the previously constructed map is reused to estimate the drone’s initial pose, enabling SLAM to be initialized without constructing a new map from scratch. Map reuse SLAM is achieved by performing loop closure using the stored DBoW descriptors together with the previously generated map data. The proposed framework is validated through experiments conducted in various indoor environments, demonstrating the feasibility and effectiveness of the proposed map reuse approach.

Keywords: Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: This paper proposes a human-inspired utility function for decision-making in lane-merging scenarios using explainable AI. Conventional utility functions are often designed based on heuristic assumptions and fixed weighting parameters. As a result, they may not reflect how human drivers adjust their preferences according to surrounding traffic condi-tions. The proposed method extracts factors related to human driver decision-making from real-world driving data us-ing XGBoost-SHAP and incorporates them into the safety utility. The XGBoost model learns the relationship between driving-related features and the future heading angle, which is used as an observable proxy for the initial lateral ma-neuver response associated with merging. The proposed utility function is applied to a Stackelberg game-based driver model for validation. Decision-level validation results show that the model using the proposed utility function generates lane-keeping and lane-merging decisions that are more similar to those of human drivers than those generated by a conventional utility-based model. These results suggest that factors related to human driver decision-making can be quantitatively incorporated into autonomous driving decision algorithms, thereby supporting more human-like deci-sion-making.

Keywords: Biomedical Instruments and Systems, Robot Mechanism and Control, Robotic Applications
Abstract: Gastrointestinal (GI) diseases remain challenging to treat because of the complex structures and limited accessibility of the GI tract. In this study, we propose a magnetically actuated capsule-based therapeutic platform for targeted multi-site treatment within the GI tract. The proposed system consists of a magnetically actuated capsule robot equipped with deployable therapeutic sheets (TheraSs) and a real-time imaging module for active navigation and targeted delivery. The TheraS has a multilayer structure enabling self-unrolling and conformal attachment to curved tissue surfaces. Experimental results demonstrated effective hemostatic performance and successful multi-site delivery in ex vivo environments, highlighting the potential of the proposed system as a therapeutic platform for GI diseases.

Keywords: Biomedical Instruments and Systems, Artificial Intelligence Systems, Sensors and Signal Processing
Abstract: Remote photoplethysmography (rPPG)-based blood pressure (BP) estimation enables continuous physiological monitoring in a non-contact manner; however, unstable window-level predictions may occur due to illumination changes, motion, and signal quality degradation. In this study, we propose a post-processing method that applies temporal variance-based filtering to the outputs of a long short-term memory-based BP estimation framework using rPPG features. The proposed method defines four consecutive predictions in each subject-specific systolic blood pressure (SBP) and diastolic blood pressure (DBP) prediction sequence as a chunk, and uses the prediction variance within each chunk to identify temporally unstable predictions. Predictions judged to be stable in both SBP and DBP sequences are then used as filtered predictions. In experiments using a public rPPG-based BP dataset with 10 repeated trials, the filtered predictions retained an average of 76.6% of the unfiltered predictions while showing lower mean absolute error and error standard deviation than the unfiltered predictions. These results suggest that the proposed post-processing method may help reduce error and error variability in rPPG-based BP estimation results.

Keywords: Biomedical Instruments and Systems, Sensors and Signal Processing
Abstract: Remote photoplethysmography (rPPG)-based oxygen saturation (SpO2) estimation has been investigated as a non-contact alternative to conventional pulse oximetry. Most camera-based SpO2 methods use the ratio-of-ratios (RoR) principle, where the normalized pulsatile variation of each RGB channel is computed from its AC and DC components. However, model-based rPPG algorithms such as plane-orthogonal-to-skin (POS) and CHROM are mainly used for pulse signal extraction, and their connection to RoR-based SpO2 estimation remains less direct. This paper proposes a simple gain-based RoR approach that links model-based rPPG signal extraction with SpO2 estimation. Each RGB channel signal is modeled as a combination of a component synchronized with a pulse basis and a residual component, and the AC related feature is redefined as a pulse-synchronous gain. The gain is obtained as the projection coefficient of each channel signal onto the pulse basis. A POS-based rPPG signal is used as the surrogate pulse basis. The proposed method was evaluated on the PURE dataset using record-specific calibration and three regression models. Experimental results show comparable or improved performance over conventional RoR, suggesting the feasibility of gain RoR as an alternative feature for rPPG-based SpO2 estimation.

Keywords: Biomedical Instruments and Systems, Exoskeleton Robot, Rehabilitation Robot
Abstract: This paper presents ToesiFleX, an adaptive soft robotic exosuit intended for foot drop assistance on level, inclined, and declined surfaces. The system possesses a lightweight cable-driven dorsiflexion architecture consisting of a sandal-based foot interface, a distal cable anchor, Bowden cable transmission, routing support, a shank pad, and a shank-mounted actuation unit. The waist-mounted terrain-slope detection module includes two Time-of-Flight sensors and an inertial measurement unit to estimate the upcoming planar terrain angle. Bench-top evaluation was performed using a dummy-limb prototype, an actuated waist-motion simulator, and an adjustable ramp. Tests of the terrain-slope recognition method were conducted on ramp angles from -15° to 15°, resulting in tolerance-based accuracy above 90%. The cable-driven dorsiflexion subsystem was subsequently evaluated on the dummy limb using a foot-mounted IMU. Across the tested terrain conditions, the actuation system achieved an average tolerance-based accuracy of 76.6% within ±3° of the required dorsiflexion profile. These experiments validate the terrain-sensing and dorsiflexion-actuation subsystems at the proof-of-concept level, while human-subject performance remains to be evaluated.

Keywords: Biomedical Instruments and Systems, Robotic Applications
Abstract: Glioblastoma treatment remains limited by the blood-brain barrier (BBB), which suppresses the accumulation of systemically administered anticancer drugs at brain tumor lesions. Focused ultrasound (FUS) with microbubbles can reversibly open the BBB, but conventional magnetic drug targeting based on static magnetic fields can induce magnetic nanoparticle (MNP) chain formation, reducing penetration through the opened barrier. This paper presents a compact conference summary of an integrated focused ultrasound and electromagnetic actuation (FUEM) system for enhanced drug-loaded MNP delivery. The FUS unit generated a focal acoustic field and opened a bEnd.3-MBVP in vitro BBB model using 1 MHz ultrasound and microbubbles. The 8-coil EMA unit produced 0.1 T magnetic fields and dynamic rotating fields to trap MNPs while breaking chains. Under BBB-opened dynamic magnetic field operation, MNP penetration increased by approximately seven-fold compared with the closed-BBB/no-field control, and doxorubicin-loaded MNP delivery reduced U-373MG glioblastoma cell viability to 28.51%.

Keywords: Biomedical Instruments and Systems
Abstract: Visual and textual cues in digital environments significantly influence individual food-choice responses. This study examined whether the relationship between subsequent food-choice behavior and brain activation in the prefrontal cortex (PFC) differed between high- and low-calorie virtual food-choice conditions. Seventeen healthy adults performed a food-choice task comprising high- and low-calorie main-menu conditions. During the task, brain activation in the PFC was measured using a 15-channel functional near-infrared spectroscopy (fNIRS) device, and eye movements were simultaneously recorded. Raw optical intensity data were converted to concentration changes in oxygenated hemoglobin (ΔHbO), and the mean ΔHbO within the 20–30 s analysis window was calculated for each block and channel. For each block, the opposite choice rate (OCR) was calculated as the proportion of side-menu choices made in the calorie direction opposite to that of the preceding main menu. For analysis, linear mixed-effects models were used to examine the interaction between the main-menu condition and OCR, with a random intercept for participant. The main-menu condition × OCR interaction remained significant after false discovery rate correction at Ch13, 14, and 15 (q < 0.05). The positive interaction coefficients indicated that the association between OCR and ΔHbO was more positive in the high-calorie main-menu condition than in the low-calorie condition. These findings indicate that the relationship between subsequent food-choice behavior and prefrontal hemodynamic responses differs according to the calorie condition of the preceding main-menu choice.

Keywords: Civil and Urban Control Systems, Industrial Applications of Control, Sensors and Signal Processing
Abstract: Accurate vehicle detection from a roadside radar is essential for secondary accident prevention in intelligent transportation systems. Traditional clustering methods like DBSCAN rely on isotropic distance metrics that fail to capture elongated vehicle geometries when vehicles are oriented at varying angles relative to the sensor. We propose a Heading-Adaptive Anisotropic DBSCAN (HAA-DBSCAN) algorithm that incorporates vehicle orientation directly into the clustering process. Our method combines temporal motion vectors with spatial PCA for heading estimation, and then replaces the standard Euclidean distance with an orientation-aligned elliptical distance metric to prevent merging of closely-spaced vehicles with dissimilar orientations. DBSCAN was selected as the clustering baseline after comparative evaluation against K-Means and agglomerative clustering. Experimental validation on real-world roadside radar data demonstrates successful segmentation of vehicles with different orientations. In addition, CPU-only runtime evaluation on a Raspberry Pi 4 achieves 14.396 ms/output, corresponding to 69.5 FPS, demonstrating the feasibility of the proposed method for real-time embedded roadside deployment.

Keywords: Control Devices and Instruments, Process Control Systems, Sensors and Signal Processing
Abstract: This paper proposes an economical and efficient End-of-Line (EOL) testing system for final quality assurance in Interior Permanent Magnet Synchronous Motor (IPMSM) production lines for eco-friendly vehicles. Conventional commercial EOL equipment utilizes high-voltage insulation testers, three-phase power analyzers, expensive National Instruments (NI) data acquisition (DAQ) hardware, and LabVIEW-based systems, which entail extremely high deployment costs (CAPEX) and severe reliance on foreign vendors. In this study, these expensive instruments are completely replaced with a general-purpose high-performance microcontroller unit (MCU), specifically TI's TMS320F28377, LEM voltage sensors (LV25-P), and an Analog Devices RDC (AD2S1210) chip. A low-cost architecture is realized by optimizing the MCU’s internal ADC channels and dedicated hardware circuits. Furthermore, a process parallelization algorithm is developed to simultaneously execute the back-electromotive force (back-EMF) measurement and the resolver offset measurement under a high-speed driving condition of 1000 rpm, which were previously conducted sequentially. The proposed system was deployed and validated on an active IPMSM mass-production line of a leading global automotive powertrain manufacturer. The results demonstrate that the system maintains measurement accuracy equivalent to or higher than that of high-end commercial equipment while reducing the total EOL inspection tact time from 65 seconds to 55 seconds, achieving a breakthrough 15.4% increase in manufacturing productivity.

Keywords: Control Devices and Instruments, Industrial Applications of Control
Abstract: Active magnetic bearings (AMBs) are widely used for high-speed rotating machines due to superior reliability and efficiency. Unlike symmetric AMBs with bias currents, asymmetric bearings use unbiased control which may lead to a force slew-rate limitation. In this paper, we derive the force slew rate of asymmetric unbiased AMBs in terms of coil currents and phase voltages. Since the voltages are limited by the DC-link voltage of the inverter, we provide a lower bound of the coil currents ensuring the required force slew rate. To validate the force slew rate analysis of the asymmetric unbiased bearing, a test rig is setup consisting of a prototype compressor and 9-pole asymmetric bearings. Force slew rate is obtained from the measured voltages and currents. It matches well with the numerically computed force slew rate using the electromagnetic forces, thus validating the analysis presented in the paper.

Keywords: Robot Mechanism and Control, Robotic Applications, Navigation, Guidance and Control
Abstract: This paper proposes a velocity and formation planning method with slip and tip-over avoidance for load-bearing cooperative transport, in which two non-holonomic mobile robots carry a rigid object held only by friction on their top platforms. The proposed method combines a Coulomb friction-circle (no-slip) limit and a zero-moment-point (ZMP) tip-over limit into a single lateral-acceleration bound. On top of this bound, a time-efficient velocity profile is generated via path--velocity decomposition, allowing the payload to be carried quickly while respecting the contact constraints. Unlike existing kinematic formation controllers that determine the admissible speed only empirically, the proposed method explicitly reflects the contact-level dynamics and adapts the transport speed to the local path curvature. Additionally, it reveals a formation--tip-over coupling: the in-line formation loads the weak roll axis during cornering, while the parallel formation loads the stable load-shift axis. The planned trajectories are checked by an independent rigid-body re-simulation, which also exposes an object-rotation failure mode invisible to point-mass analysis. Through experiments on warehouse-like and tight-corner scenarios in simulation, the proposed method keeps payload slip below the threshold and prevents tip-over across the tested friction range, while reducing transport time compared to the constant-velocity baseline.

Keywords: Control Devices and Instruments, Industrial Applications of Control, Artificial Intelligence Systems
Abstract: This paper presents the design, implementation, and industrial deployment of the Smart Vision-based AI Defect Inspection System (S-VADIS), developed through a traceability-driven systems engineering approach for side-edge defect inspection in hot-rolling processes. To ensure architectural consistency and robustness, the system was developed using a traceability-driven systems engineering methodology, where stakeholder requirements were systematically translated into verifiable system requirements and realized through a well-defined functional and implementation architecture. The S-VADIS enables reliable, full-coverage image acquisition under harsh industrial conditions, including high-speed operation and ambient temperatures exceeding 750 °C. To maintain production throughput while ensuring high inspection accuracy, the system employs a three-stage deep-learning pipeline—integrating YOLOv5-based localization, ensemble-based re-classification, and binary post-processing. Furthermore, the system’s data-driven architecture allows it to be leveraged as an intelligent quality management (IQM) platform; by aggregating bilateral defect data and providing intuitive quality indicators via a Human-Machine Interface (HMI), it facilitates proactive quality control and seamless integration with external analytics systems. The successful deployment of the S-VADIS in an actual hot-rolling production line validates the effectiveness of the traceability-driven design approach, demonstrating its practical feasibility and scalability for advancing smart manufacturing in the steel industry.

Keywords: Control Devices and Instruments, Robotic Applications, Information and Networking
Abstract: Vapor–liquid equilibrium (VLE) data are essential for thermodynamic model fitting and separation-process design, yet reliable data generation remains labor-intensive for emerging solvent systems, hazardous mixtures, and new operating windows. Specialized VLE apparatuses and recent laboratory-automation frameworks address complementary parts of this challenge, but end-to-end coordination across preparation, reactor operation, sample transfer, and analysis remains limited in distributed laboratory settings. This paper presents a robotic workflow that connects sample preparation, reactor-side operation, transfer, and UV–Vis analysis in one experimental sequence. The platform integrates a custom liquid-handling station, a customized reactor module for automated sequencing and logging, robot-assisted vial and cuvette handling, and centralized supervisory control. Subsystem results showed gravimetrically inferred mean delivered volumes within 0.3% of target at 1, 3, and 5 mL, successful ten-cycle repetition of robot-assisted transfer and UV–Vis interaction tasks, automated acetone–water calibration with an R-squared value greater than 0.9999, and ethanol–water reactor checks that followed the Aspen-predicted VLE trend at a 50 mol% feed. These results show how one robotic platform can connect prepared composition, reactor-side history, and analytical evidence into a repeatable workflow toward traceable and scalable VLE data generation.

Keywords: Control Devices and Instruments, Control Theory and Applications
Abstract: Active suspension systems require concurrent optimization of plant and controller subsystems (Control Co-Design, CCD) due to their strong bidirectional coupling. While direct transcription (DT) is widely used to solve CCD problems, conventional formulations implicitly nest nonlinear independent to dependent design variable mappings. To resolve these bottlenecks, this study proposes a novel Lifted CCD formulation. This method enables the solver to utilize topological shortcuts through temporarily infeasible spaces. Quantitative validation on an active suspension system demonstrates a 33.3% reduction in cost function value, while simultaneously improving vehicle ride comfort and energy efficiency. Ultimately, this framework offers a highly efficient, scalable paradigm universally applicable to various complex multi-physics co-design problems.

Keywords: Control Devices and Instruments, Robotic Applications
Abstract: To improve airport operational efficiency, Incheon Airport is introducing heterogeneous autonomous mobile robots. However, each manufacturer uses different data formats and verification methods, complicating their operation and management. This study develops a verification system for a standardized robot framework that is vendor-agnostic, based on three core principles: standardization, reliability, and scalability. The framework unifies robot data formats and verification methods, and includes multi-robot cooperative control. It organizes the software into three layers-interface, abstraction, and service-to convert raw data from different robots into one standard model. Data is collected, stored, and displayed through a dedicated network that is separate from the airport's internal network. The verification system confirms reliability through virtual-physical verification, combining simulation and real-world testing.

Keywords: Control Theory and Applications, Information and Networking
Abstract: This paper addresses the challenges in wireless ad-hoc networks, particularly focusing on the construction of a Connected Dominating Set (CDS) as a virtual backbone network. Unlike previous works that assume uniform qualifications for all nodes, we consider cases with predetermined nodes that must be either included in or excluded from the CDS due to their unique functionalities. In this context, we extend the Enforced-CDS problem to a CDS construction problem with predetermined nodes. Contribution includes the development of both centralized and distributed algorithms that utilize one-hop information, diverging from the traditional two-hop approach. The size of the computed CDS by the proposed algorithms does not exceed 6.798 times the optimum. Furthermore, simulation results show that the difference between centralized and distributed algorithms is minimal. This work provides a theoretical analysis and showcases the efficacy of algorithms in handling the dynamic requirements of wireless ad-hoc networks.

Keywords: Control Theory and Applications
Abstract: This paper investigates the reliable control problem for Takagi–Sugeno (T–S) fuzzy systems in the presence of input saturation, actuator faults, and external disturbances. The proposed model accommodates matched actuator faults together with mismatched external disturbances, thereby reflecting practical operating conditions. To cope with these challenges, a sequence of nested invariant ellipsoidal sets and the corresponding set-dependent control gains is constructed, which drives state trajectories from a sufficiently large outer set into a minimal target set. The associated design conditions are then formulated as relaxed LMIs, so that both the invariant sets and the control gains can be obtained simultaneously through convex optimization.

Keywords: Control Theory and Applications, Information and Networking, Sensors and Signal Processing
Abstract: This work studies the distributed online control problem over a network of linear time-invariant (LTI) systems in the presence of adversarial disturbances and time-varying convex costs. The network cost is characterized by the summation of local cost functions, where each local function is sequentially revealed only to the corresponding agent. The goal of each agent is to generate a control sequence, using only local observations and neighbor communication, that competes with the best {it centralized} linear policy in hindsight. We extend the recently proposed Online Spectral Control framework from the centralized setting to the distributed setting. In particular, each agent applies a spectral controller obtained by convolving past disturbances with the leading eigenvectors of a Hankel matrix, while the controller parameters are updated through a distributed online gradient descent step over the local surrogate costs. We formulate this problem as a {it regret} minimization problem based on the spectral parameterization, and under standard assumptions, we establish a sublinear regret bound of O(frac{sqrt{T}text{poly}(log T)}{gamma^3}), where T is the time horizon and gamma denotes the stability margin. The resulting bound also captures the dependence on the network size and connectivity.

Keywords: Control Theory and Applications, Industrial Applications of Control, Process Control Systems
Abstract: The startup of a packed-column batch distillation is a transient multi-phase process in which liquid phases progressively form on initially dry packing. The resulting dynamic-optimization problem combines complementarity-constrained phase transitions with a large-scale NLP that is sensitive to the initial guess and scaling, so existing studies tune these settings system-by-system. We propose a binary-system-generic DAE–MPEC framework that removes this per-system tuning. A beta-based VLE relaxation places the liquid-only, two-phase, and vapor-only regimes in one continuous variable space, and the resulting DAE coupling NRTL thermodynamics with Billet–Schultes hydraulics is discretized by Radau IIA direct collocation. An auto-initialization derives variable bounds, initial guesses, and NLP scaling from thermodynamic data, and a primal–dual warm-start continuation on delta delivers the converged solution. The framework is demonstrated on four binary systems—acetone/ethanol, acetone/water, benzene/toluene, methanol/ethanol—with optimal startup times in 34.6–50.7 s and mass/energy balance errors below 0.10 %.

Keywords: Control Theory and Applications
Abstract: This paper presents an extended robust approximate feedback linearization scheme for an electromagnetic levitation system. Unlike conventional approaches, the proposed method explicitly adopts a non-Brunovsky canonical form, which leads to more flexible controller design over the conventional Brunovsky canocnial form based methods. For stability analysis, an ϵ-scaled matrix framework is employed, which leads to a modified Lyapunov equation and an associated Lyapunov function. Based on this analysis, two sufficient ranges of the gain-scaling parameter ϵ are derived to guarantee asymptotic stability of the closed-loop system. Finally, the effectiveness of the proposed approach is demonstrated through a comparative study of the stability range and control performance against an existing method within the same analytical framework.

Keywords: Control Theory and Applications
Abstract: In this paper, we study the three-dimensional generalized Navier-Stokes-Voigt equations, which are closely related to the modeling, stability analysis, and control of incompressible fluid systems. The global existence and uniqueness of solutions are proved for alpha > 5/6. If the initial data belongs to the Sobolev space H^{-s} and H^{alpha+1}, with 5/6 < alpha < 1, 0 < s < 1/2, and 3/2 < 2 alpha + s < 5/2, we establish the corresponding negative Sobolev space estimate in Theorem 2. The obtained estimates provide theoretical support for the stability and control analysis of Voigt-regularized fluid models. Moreover, numerical simulations based on a pseudo-spectral method are performed on the periodic domain to verify the theoretical results. The numerical results demonstrate the decay behavior of several energy-related quantities and agree well with the analytical estimates.

Keywords: Control Theory and Applications
Abstract: This study presents an adaptive time-delay sliding-mode controller (ATDSMC) for inner-loop attitude tracking of a quadcopter subject to unmodeled dynamics and external disturbances. Time-delay estimation (TDE) reconstructs the lumped unknown dynamics and equivalent external torque isturbance using one-step-delayed signals, thereby reducing reliance on the exact configuration-dependent inertia and oriolis matrices. Based on the TDE, the attitude-control input is constructed using a sliding surface and a componentwise adaptive switching-gain law. Practical boundedness of the sliding variable and tracking error is established under a bounded TDE-error assumption. Simulations of the threedegree- of-freedom rotational subsystem compare the proposed ATDSMC with linear time-delay control (LTDC) and conventional time-delay sliding-mode control (CTDSMC). Based on the mean of the roll, pitch, and yaw root-meansquare error (RMSE) values, ATDSMC reduces the tracking error relative to CTDSMC by 33.80% and 38.21% in the nominal and wind-disturbance cases, respectively.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems
Abstract: This paper proposes an output feedback control scheme accompanied by a Lyapunov-based stability analysis by explicitly introducing a gain-scaling factor ϵ into both the control and observer gains. The proposed method enables a direct assessment of system stability through the scaling parameter ϵ. Based on the resulting Lyapunov analysis, the admissible range of the design parameter ϵ is derived to guarantee the asymptotic stability of the closed-loop system. Finally, the effectiveness of the proposed approach is validated through quadcopter simulations, demonstrating improved hovering performance and robustness against external disturbances.

Keywords: Control Theory and Applications, Industrial Applications of Control, Process Control Systems
Abstract: This paper proposes a model-based fault detection approach for wind turbine pitch systems using an adaptive Kalman filter (AKF). The Kalman filter (KF) relies on fixed process and measurement noise covariances, which are not suitable for different operating conditions and fault magnitudes, and may degrade estimation accuracy. To overcome this limitation, the AKF is employed, which updates the noise covariance matrices in real time according to changes in system behavior due to different fault conditions, thereby improving the robustness of state estimation and enhancing fault detection performance. Residual signals are generated using the AKF, and fault detection is performed by comparing the residuals to thresholds, which are calculated using the H∞ norm and linear matrix inequalities. Simulation studies conducted under stochastic wind conditions demonstrate the effectiveness of the proposed method in detecting realistic pitch system faults.

Keywords: Control Theory and Applications, Human-Robot Interaction, Robot Mechanism and Control
Abstract: Low-impedance rendering is essential for achieving safe and compliant physical human–robot interaction. However, practical robotic systems inherently contain actuator nonlinearities, residual damping, model uncertainty, and unmodeled dynamics, which distort the realized impedance characteristics and generate steady-state errors under low-impedance conditions. This paper analyzes the influence of such residual dynamics on impedance rendering from an actuator-space perspective and shows that low-impedance rendering inherently amplifies disturbance sensitivity and equilibrium offsets. Also, a disturbance-observer-based robust impedance control framework is proposed. The proposed controller compensates for residual disturbances while preserving the interaction torque required for impedance rendering through a force-feedback structure. Analytical derivations show that the proposed framework effectively nominalizes the actuator dynamics and recovers the desired impedance characteristics. Experimental results demonstrate that the proposed method improves impedance rendering accuracy and eliminates steady-state errors caused by actuator nonlinearities and residual disturbances.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems, Robot Mechanism and Control
Abstract: This paper presents a framework for real-time optimal control of a heavy-duty skid-steered mobile platform for trajectory tracking. Accurate real-time performance is important in situations where the system is affected by uncertainties and disturbances, and the controller must compensate for these effects to provide stable performance. A multiple-shooting nonlinear model-predictive control framework is proposed, using sensor readings and a suitable optimization routine for genuine real-time performance with high accuracy. The controller is tested on different tracking trajectories where it demonstrates desirable performance in terms of both speed and accuracy. The obtained results demonstrate millisecond-scale execution and centimeter-level trajectory-tracking accuracy on the considered heavy-duty platform.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems, Navigation, Guidance and Control
Abstract: This paper investigates a Takagi-Sugeno (T-S) fuzzy controller for the lateral dynamics of Flapping-Wing Micro Aerial Vehicles (FWMAVs) while considering actuator saturation and external disturbances. A T-S fuzzy model is derived by directly incorporating the force and moment dynamics of the left and right wing pairs, where altitude and lateral motion are regulated through total and differential thrust. In addition, realistic scenarios dealing with saturation constraints and external disturbances are rigorously addressed. Accordingly, the T-S fuzzy control gains are designed based on sufficient stabilization conditions formulated in terms of Linear Matrix Inequalities (LMIs). Finally, experimental validation on a Flapper Nimble+ demonstrates the effectiveness of the proposed approach.

Keywords: Control Theory and Applications, Artificial Intelligence Systems, Process Control Systems
Abstract: Large Language Models (LLMs) are increasingly deployed in real-world applications, yet they remain prone to generating unsafe or undesirable content. Recent work on Control Barrier Functions (CBFs) introduces a control-theoretic framework for safe decoding by enforcing token-level safety constraints during generation. However, existing approaches assume access to a perfect safety classifier, an assumption that fails in practice due to noise, model mismatch, and distributional shift. In this paper, we propose Adaptive Control Barrier Function (AdaptiveCBF) filtering, a robust extension of CBF-based decoding that explicitly accounts for classifier uncertainty. Our method introduces (i) an adaptive strictness parameter that is tighter near the safety boundary than in deep-safe regions, and (ii) a safety margin that compensates for stochastic classifier noise. We further propose a boundary-only variant (AdaptiveCBF-BO) and a smooth relaxation (SmoothCBF) to improve efficiency and generation quality. We provide a probabilistic safety guarantee showing that, under Gaussian classifier noise, the probability of maintaining safety over a sequence scales with the safety margin-to-noise ratio as Φ(ε/σ)^N. Empirical evaluations across multiple tasks (toxicity, sentiment, topic control), benchmarks (RealToxicityPrompts, BeaverTails), and models demonstrate that AdaptiveCBF consistently maintains or improves safety over standard CBF methods under classifier noise. Our approach is training-free, modular, and readily applicable to existing LLMs.

Keywords: Control Theory and Applications, Industrial Applications of Control, Autonomous Vehicle Systems
Abstract: This paper proposes an integrated fault detection and isolation (FDI) and active fault-tolerant control (AFTC) framework for in-wheel-motor electric vehicles. In conventional torque-vectoring control, actuator, tire, or steering faults may be treated as external disturbances, which can lead to inappropriate torque allocation under degraded vehicle conditions. To address this problem, a nonlinear unknown input observer is first designed to perform chassis-level preliminary fault detection under unknown inputs. Then, per-wheel observers combining Luenberger-type observation, adaptive estimation, and sliding-mode correction are introduced to isolate wheel-related faults and distinguish them from steering faults according to a fault-decision rule. Based on the isolated fault condition, the torque-vectoring controller is reconfigured by modifying the prediction model, control weights, and torque allocation rule. Simulation results under motor efficiency degradation, tire puncture, and steer-by-wire fault scenarios show that the proposed framework improves fault isolation capability and maintains post-fault vehicle stability compared with conventional disturbance-compensation-based torque-vectoring control.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Process Control Systems
Abstract: Robust quadruped locomotion on uneven terrain requires terrain-aware command adaptation under uncertain contact conditions. This paper proposes a terrain-label-conditioned GP-MPC supervisory framework for RL-based quadruped locomotion. A visual terrain recognition module assigns terrain labels, a Gaussian Process residual model predicts terrain-dependent model errors and uncertainty, and a GP-MPC supervisor refines the velocity command before it is passed to a pre-trained RL policy. Isaac Sim/Isaac Lab experiments on flat, gravel, hill, and stairs terrains show that the proposed framework improves GP prediction accuracy and command-level adaptation compared with RL-only and RL-MPC baselines.

Keywords: Control Theory and Applications, Process Control Systems, Industrial Applications of Control
Abstract: Local device controllers in a robotic material laboratory are designed in isolation and cannot guarantee global interlocks. This paper formulates the coordination layer of a robotic material synthesis–transfer–analysis workflow as a discrete-event supervisory control problem. The plant composes four modular automata: the mobile manipulator, the synthesizer, the analyzer, and a dedicated sample-state automaton that records the location and processing stage of the sample, including the configuration entered after an uncontrollable grip failure. Five specification automata encode process order, equipment and robot command–state matching, and sample-state consistency; a nonblocking supervisor is synthesized from the supremal controllable sublanguage of the legal language and verified exhaustively over the 17-state closed loop. Its necessity is established by an ablation against the supremal supervisor of the equipment-only model: after the same trace, in which the gripper releases the sample at the synthesizer, only the proposed supervisor disables the load command that follows, the other admitting it and starting a reaction on an absent sample. Sixteen of the 155 disablements are induced by the sample-state specification alone, and removing the post-failure sample states leaves the synthesis problem unsolvable.

Keywords: Control Theory and Applications, Artificial Intelligence Systems, Robotic Applications
Abstract: This paper proposes a learning-based torque-level disturbance observer (DOB) for quadruped locomotion policy. The observer predicts the equivalent applied joint torque from proprioceptive history and estimates the disturbance by comparing it with the applied control torque. The estimated disturbance is compensated online through feedforward torque correction. Simulation results under joint friction, external push, and periodic torque disturbances show improved velocity tracking, attitude stability, and termination rate compared with the baseline policy without DOB compensation.

Keywords: Control Theory and Applications, Navigation, Guidance and Control, Information and Networking
Abstract: This paper is focused on the synchronization problem of coupled Kuramoto oscillators. We treat the coupling term as a control input and analyze the synchronization phenomenon in both infinite and finite time settings from the perspective of control theory. While existing studies have mainly investigated infinite time synchronization and finite time synchronization separately, this paper analyzes and compares both synchronization mechanisms within a common Kuramoto oscillator framework. In the infinite time case, we deal with the general Kuramoto model and analyze whether the phase difference converge to zero asymptotically, by examining the Jacobian matrix at the equilibrium (synchronization) point. A numerical simulation is provided to verify the proposed approach. In the finite time setting, we derive an upper bound of the settling time and show that synchronization can be achieved within a finite time characterized by the settling time. A numerical simulation is performed to confirm the finite time synchronization behavior of the coupled oscillators.

Keywords: Control Theory and Applications
Abstract: This paper investigates the existence of solutions for fractional differential inclusions with nonlocal antiperiodic boundary conditions. The existence results for the convex case are established by employing the Leray-Schauder nonlinear alternative theorem. Furthermore, sufficient conditions for the existence of solutions in the nonconvex case are derived via the contraction mapping principle. The results obtained in this paper generalize the relevant conclusions for fractional differential equations. Finally, two examples are provided to illustrate the validity of the main results.

Keywords: Control Theory and Applications, Human-Robot Interaction, Robot Mechanism and Control
Abstract: This paper presents a unified control framework that integrates adaptive sliding mode control (ASMC), time delay control (TDC), and admittance filtering to achieve robust force and position tracking in robot manipulators. TDC is employed to estimate unmodeled dynamics using delayed measurements, while ASMC enhances robustness by adaptively compensating for time-delay estimation (TDE) errors and mitigating chattering. The admittance mechanism transforms force tracking errors into position corrections, enabling force tracking without altering the underlying position control structure. A novel adaptive law is introduced to regulate control gains more stably by incorporating a decline-rate adjust mentfactor, improving tracking under uncertainty. Stability of the proposed system is guaranteed through Lyapunov-based analysis, and simulation results demonstrate a significant improvement in position tracking accuracy—reducing RMSE from 0.0522 mm to 0.019 mm—while maintaining reliable force tracking performance.

Keywords: Control Theory and Applications, Information and Networking
Abstract: This paper proposes a Ring Learning With Errors (Ring-LWE) based encrypted iterative learning control (ILC) framework for repetitive tracking tasks over networked control systems. The architecture integrates an encrypted dynamic feedback controller with an encrypted ILC computation. During each trial, the feedback controller is evaluated in the ciphertext domain, and the encrypted output trajectory is stored directly in the cloud. After each trial, the cloud evaluates the tracking error and performs the ILC computation from the stored ciphertexts, so that the plant side does not need to store the accumulated trial data. The proposed framework uses distinct packing parameters for ciphertext multiplication, allowing the cloud to handle both lower-dimensional output feedback control and higher-dimensional ILC computation without decryption. While error growth in Ring-LWE based encrypted control is generally suppressed by closed-loop stability, the marginally stable ILC iterations cause the injected errors to accumulate continuously. To address this challenge, a range-space decomposition is introduced in the encrypted ILC formulation to allow evaluation under unlimited updates. Numerical simulations show that the range-space decomposition suppresses encryption-induced perturbation, while ciphertext packing improves the computational efficiency of the encrypted ILC update.

Keywords: Control Theory and Applications, Control Devices and Instruments, Industrial Applications of Control
Abstract: This paper presents a control-oriented modeling approach for a hybrid magnetic bearing intended for a solar array drive assembly. Conventional bearing-supported drive mechanisms are vulnerable to friction, wear, and contamination, which can significantly degrade performance in harsh operating environments. To address these limitations, a non-contact support concept based on a hybrid magnetic bearing is investigated. The proposed system combines permanent magnets and electromagnets to provide bias flux and actively controllable suspension force. For control design, a lumped-parameter force model is developed by relating the magnetic bearing force to the control current and air-gap displacement around the nominal operating point. Based on this model, the dynamic behavior of the suspended rotor is analyzed, showing that active feedback control is required to maintain stable operation. A displacement-feedback control method is introduced to regulate the rotor position and maintain the desired air gap under external disturbances. Simulation results indicate that the proposed approach can achieve stable suspension and improved disturbance rejection, demonstrating its feasibility for future non-contact solar array drive assembly systems.

Keywords: Control Theory and Applications, Autonomous Vehicle Systems, Artificial Intelligence Systems
Abstract: Control barrier functions provide a lightweight way to enforce safety in real time, but the associated quadratic program can become infeasible under input saturation unless the set defined by the barrier is control invariant. The maximal control invariant set can be obtained through Hamilton--Jacobi reachability, yet the standard grid-based solver scales exponentially with the state dimension. This paper presents a sample-based method that approximates this set with far fewer points by treating the samples as a design object rather than a passive grid. Initial samples are placed along the safe-set boundary from the safety specification, and a portion migrates inward as the value iteration proceeds, so the limited budget is spent where the value function departs from the specification. A Gaussian process with a signed-distance prior mean represents the value function and supplies a posterior variance, from which an uncertainty-aware barrier is synthesized as a lower confidence bound of the posterior. On a Dubins car avoiding a circular obstacle, the method recovers the control invariant set more accurately than fixed-sample baselines at the same budget and yields safe goal-reaching behavior.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Robotic Applications
Abstract: This paper proposes a state estimator-based robust control framework for a Motion platform to enable safe Unmanned Aerial Vehicle(UAV) takeoff and landing. Instead of IMUs or direct task-space sensing, time-of-flight(ToF) sensors measure actuator leg lengths. Based on these measurements, a super-twisting algorithm(STA)-based observer reconstructs the platform’s pose and velocity. These estimates define tracking errors, and a super-twisting sliding mode controller (STSMC) generates actuator inputs for robust tracking under disturbances and uncertainties. The method enables output-feedback control without direct task-space measurements, reducing sensor dependency. Experiments demonstrate reliable state estimation and tracking performance, contributing to safer UAV operations.

Keywords: Control Theory and Applications, Navigation, Guidance and Control, Industrial Applications of Control
Abstract: This paper proposes an observer-based event-triggered control scheme for leader-follower multi-agent linear systems. The proposed approach aims to reduce unnecessary communication among agents while guaranteeing stable leader-following synchronization. In the proposed framework, each follower agent updates its control input only when a predefined event-triggering condition is satisfied, thereby reducing both communication and control update frequencies. An observer-based controller is employed, and the stability of the closed-loop system is analyzed based on Lyapunov stability theory. Numerical simulation results demonstrate that all follower agents successfully track the leader state under reduced communication events. The results confirm that the proposed event-triggered control strategy achieves stable leader-following synchronization while improving communication efficiency and reducing unnecessary control updates in multi-agent systems.

Keywords: Control Theory and Applications, Artificial Intelligence Systems, Control Devices and Instruments
Abstract: Reinforcement learning is a method for achieving control objectives through iterative learning processes, much like humans do. A swing is an interesting physical system that can generate continuous oscillatory motion through the shifting of the rider's center of gravity. Humans also require considerable trial and error to sustain swinging. In this study, a swing device using the contraction of pneumatic artificial muscles was developed, and a reinforcement learning technique was implemented to demonstrate its capability to learn how to amplify swing motions. Experimental results showed that swing motion was possible with a stationary rider by adjusting the length of the artificial muscles instead of using fixed-length ropes.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Robotic Applications
Abstract: Robot manipulators are widely used in industrial applications that require high tracking accuracy and robustness. However, in practical robotic systems, time delays arising from sensing, computation, and communication processes may degrade tracking performance and even destabilize the robotic system. Since such delays can be regarded as disturbance-like uncertainties that degrade control performance, robustness against time delays is an important issue in robot manipulator control. This paper presents a reference modulation technique for robot manipulators to mitigate the adverse effects of time delays. To improve tracking performance and robustness, the proposed method modulates the reference signal to shape the loop-gain characteristics over a desired frequency range. In addition, the proposed method can be incorporated into conventional control frameworks without significant structural modification. The effectiveness of the technique is validated through simulations on a 2-DOF robot manipulator, where improved tracking performance and delay robustness are demonstrated in comparison with the conventional control approach.

Keywords: Control Theory and Applications, Robot Vision, Robot Mechanism and Control
Abstract: Boundary-layer Sliding Mode Control (SMC) has been applied to vision-based stabilization, but identifying its activation condition against simpler causal baselines (LPF, PID) has been elusive. The operational condition is plant binding: the disturbance velocity approaches or exceeds the actuator slew limit, forcing a bounded per-step actuation budget against a near-binding disturbance. We validate this in two independent vision domains: handheld translation stabilization (12 1080p clips; LPF, PID, SMC, L1 ) and angular re-framing against a virtual slew-rate-limited pan–tilt–zoom (PTZ) plant. Outside binding, leave-one-out cross-validation makes SMC and LPF on translation statistically indistinguishable, and PID beats SMC on the angular 99th -percentile pointing offset by 5.85◦ (95% CI clear of zero); inside binding, the translation gap stays a tie and SMC beats PID on the angular worst case (Cohen’s d up to +0.83). The PTZ plant’s 100◦ /s default sits on the cross-over, which explains a small-margin PID result previously reported in the angular setting. A complementary ablation shows raising the SMC reaching gain k costs mean tracking outside a narrow window.

Keywords: Control Theory and Applications, Human-Robot Interaction, Robot Mechanism and Control
Abstract: In position-force control tasks, unexpected external forces applied to the robot body can degrade both end-effector position tracking accuracy and contact force regulation performance. To address this problem, we propose a null-space compliance control method for maintaining both position and force tracking performance of a redundant robotic manipulator under external interactions. The proposed method preserves the primary position and force tracking tasks of the end-effector while damping interaction-induced motion through redundant null-space behavior. In addition, a virtual contact environment is introduced to model contact interactions, enabling circular trajectory tracking under a constant contact force condition. To evaluate the effectiveness of the proposed approach, comparative simulations are conducted with and without the proposed null-space compliance control. The results demonstrate that the proposed method outperforms the baseline controller in terms of position tracking and force tracking accuracy under external interactions.

Keywords: Control Theory and Applications, Industrial Applications of Control, Robot Mechanism and Control
Abstract: This paper presents a clearance-adaptive composite High-Order Control Barrier Function (HOCBF) framework for real-time safe control of a 6-DOF manipulator in cluttered environments. Imposing one HOCBF inequality per robot–obstacle pair scales poorly with the number of pairs, while aggregating them with a fixed smoothness parameter can be overly conservative and stall the robot in a narrow passage. The proposed method aggregates all normalized pairwise barriers into a single log-sum-exp composite and adapts the smoothness parameter each cycle to the current minimum clearance, sharpening the approximation only near obstacles. With analytical Product-of-Exponentials gradients, the filter reduces to one linear inequality in the joint acceleration solved in closed form. On a UR3 with N=84 pairs, it reaches the goal at a hard 1 kHz with 0% deadline miss.

Keywords: Control Theory and Applications
Abstract: This paper investigates the problem of global stabilization via adaptive output feedback for a class of uncertain nonlinear systems with function control coefficients and unknown disturbances. To this end, a novel dynamic high gain is introduced to overcome additional system nonlinearities and severe uncertainties. Then, a high-gain observer with appropriate design parameters is constructed to reconstruct the unmeasured states. Subsequently, an adaptive output-feedback controller is designed to achieve global stabilization of the closed-loop system. Finally, a numerical example is provided to validate the feasibility and effectiveness of the proposed control scheme.

Keywords: Control Theory and Applications
Abstract: This paper focuses on designing an adaptive output-feedback control scheme for exponential stabilization of a class of nonlinear systems with function control coefficients and dynamic uncertainties. To tackle the above problem, we construct a complete and effective adaptive compensation mechanism that can compensate for system uncertainties and guarantee the desired convergence rate. Significantly, we develop an elaborate dynamic gain whose update rule integrates exponential time-varying terms. With this gain, the unknown growth rate can be well managed. In addition, real-time adjustment of convergence rate based on dynamic uncertainties ensures the system achieves exponential convergence. Accordingly, we construct an adaptive output-feedback controller based on the dynamic gain observer, which enables exponential convergence of both system states and observer states.

Keywords: Control Theory and Applications, Sensors and Signal Processing, Industrial Applications of Control
Abstract: The control performance of active suspension systems strongly depends on the accuracy of the state estimates used for feedback control. However, in practical vehicle applications, direct measurement of all suspension states is difficult because of sensor cost, installation constraints, and durability issues. In particular, the relative velocity of the damper is closely related to the suspension damping force and active control input, but it is not readily available from standard sensor configurations. This study proposes a KalmanNet-based observer-control framework for active suspension systems under limited sensor measurements. The proposed observer preserves the physics-based quarter-car model and augments the nominal model-based observer with a residual-driven GRU correction module. In this structure, learning is used to compensate for the estimation errors of the nominal observer rather than to replace the suspension model with a black-box neural network. The estimated states are supplied to an acceleration-weighted LQR controller for closed-loop active suspension control. Simulation results under limited sensor configuration, parameter uncertainty, and road disturbance show that the proposed method improves the relative velocity estimation of the damper compared with the nominal observer. The results also indicate that improved observer performance can contribute to better closed-loop body response and ride-comfort indices under limited sensing conditions.

Keywords: Control Theory and Applications, Robot Mechanism and Control, Rehabilitation Robot
Abstract: This paper proposes a polar-coordinate Internal Model-Based Impedance Control (PC-IMBIC) for rehabilitation training in planar circular motion (CM), requiring coordinated shoulder and elbow movements, for patients with neurological disorders. In robot-assisted CM training, the robot endpoint being held by a patient is ideally moved tangentially along the desired path while maintaining a constant radius. Previous studies on CM control represented the desired path using discretized trajectory points and defined local motion directions from adjacent path segments. This approach requires path segmentation to determine the local motion direction at each trajectory point. The proposed PC-IMBIC, formulated in polar coordinates, was able to separately control radial and tangential motion while providing independent assistance forces in both directions, without trajectory segmentation or local direction estimation during CM. A preliminary contact experiment, comparing PC-IMBIC with conventional IMBIC, demonstrated the advantage of PC-IMBIC for providing an independent assistance force in radial and tangential directions in CM assistance.

Keywords: Control Theory and Applications
Abstract: This paper presents an integrated co-simulation and test-bench validation method for an integrated disconnect actuator system (I-DAS) using a dog-clutch mechanism. The disconnector must complete a 2WD-AWD transition rapidly while limiting impact load and avoiding baulking, stiction, and ratcheting. An Ansys Motion multibody model was coupled with a Simulink torque controller. The model includes a 10.65 gear ratio, CAD-derived inertia, actuator torque, shaft-speed synchronization, and sleeve-displacement feedback. Phase-sweeping simulations identified a maximum baulking region of 2.22 degrees, corresponding to 29.6% of the 7.5 degree tooth pitch. A four-shaft equivalent-inertia test bench was instrumented with a force/torque sensor and a laser displacement sensor. At 20 rpm output speed and 10 A actuator current, the measured engagement time was approximately 200 ms, with a peak torque of 6.3 Nm and a peak thrust of 482.4 N. The results show that phase-aware torque control and boundary-condition matching are essential for reliable engagement and model validation.

Keywords: Control Theory and Applications
Abstract: This paper presents a Lyapunov--Krasovskii functional (LKF)-based framework for evaluating the maximum allowable time delay of linear systems with input delays. A delay-dependent stability criterion is derived by constructing an appropriate Lyapunov--Krasovskii functional and formulating the stability conditions as a set of Linear Matrix Inequalities (LMIs). The maximum allowable time delay is obtained through an iterative LMI feasibility search, providing a systematic approach to delay margin analysis. To validate the proposed framework, a two-degree-of-freedom (2-DOF) helicopter system is considered as a case study. The theoretical delay margin is compared with the experimentally observed stability limit under increasing input delays. The comparison demonstrates close agreement between theoretical predictions and experimental results, confirming the effectiveness of the proposed framework. The proposed approach provides a practical and computationally efficient tool for evaluating delay margins and verifying the stability of linear time-delay systems, and it can serve as a useful basis for the analysis and design of networked control systems and other applications affected by communication and processing delays.

Keywords: Control Theory and Applications
Abstract: In this study, a powertrain based on a compound planetary gear set was designed to meet the growing demand for high-performance electric vehicles, and torque coupling of the planetary gear was implemented through a multi-motor configuration. The system consists of two motors connected via a planetary gear set, and the vehicle driving modes are defined according to motor engagement and gear ratio selection. A control logic was developed in MATLAB/Stateflow to operate each motor within its high-efficiency region, and an optimization was performed to minimize power consumption, with the total motor power consumption as the objective function and the driving mode and torque distribution ratio as input variables. A comparison with a single planetary gear model demonstrated the superiority of the Compound Planetary Gear Set architecture, which offers diverse driving modes and greater freedom in torque distribution, and an energy consumption comparison over urban and highway driving cycles validated the effectiveness of the proposed control logic.

Keywords: Control Theory and Applications, Artificial Intelligence Systems
Abstract: This study presents a model-based integrated analysis framework for consistently interpreting and controlling complex multi-mode hybrid electric powertrains. The main objective is not only to compare different vehicles, but also to establish a unified procedure that can be applied to various powertrain architectures. Four multi-mode hybrid electric vehicles are analyzed to investigate how structural differences affect optimal energy management characteristics. A lever-diagram-based backward-facing model is developed to generalize each powertrain structure and derive engine and motor operating points from the required wheel torque and speed according to the operating mode, clutch state, and gear ratio. For each feasible operating candidate, fuel consumption and battery power usage are calculated, and the Pareto Frontier Line is derived to extract efficient operating candidates. These candidates are then used as inputs for Dynamic Programming and Pontryagin’s Minimum Principle to obtain optimal energy management solutions. By comparing DP and PMP results, the consistency and reliability of the proposed analysis framework are examined. The results can provide benchmark data for future ECMS, rule-based, and reinforcement-learning-based energy management strategies.

Keywords: Exoskeleton Robot, Rehabilitation Robot
Abstract: This paper presents sparse calibration-based personalization for kernel-based gait phase and walking speed co-estimation from wearable inertial measurement units (IMUs). A 28-speed reference library was constructed from bilateral thigh and shank trajectories in a 20-subject locomotion dataset. Rather than directly applying population-average kernels, the method combines a user-specific baseline estimated from three calibration speeds with a principal-component (PC) model of baseline-centered kinematic deviations. Offline validation showed that personalized kernels reduced phase error relative to the population kernel. In a single-participant online pilot evaluation, the personalized kernels produced numerically lower mean phase and speed errors and ran in real time on embedded hardware. However, the speed effect was not significant, and pairwise phase differences did not remain significant after multiple-comparison correction. The pilot evaluation establishes wearable implementation feasibility while indicating reference-to-online dataset mismatch and the need for larger-cohort validation.

Keywords: Exoskeleton Robot, Sensors and Signal Processing, Rehabilitation Robot
Abstract: Hip assistive robots can exhibit a discrepancy between the robot-derived thigh angle and the human thigh angle because the robot and wearer do not form a perfectly aligned kinematic chain. We propose a compact causal temporal convolutional network (TCN) that estimates the human thigh angle using only the bilateral robot-derived thigh angles and trunk inertial measurement unit (IMU) yaw. No actuator-current or motion-capture input is required at inference. In a 16- fold leave-one-subject-out evaluation of no-assist walking, Overall mean absolute error (MAE) decreased from 6.720◦for the uncorrected robot-derived angle and 3.845◦for a thigh-only TCN to 3.570◦(p = 0.0386 versus the thigh-only model). Separate condition-shift tests evaluated assisted, 1.25-m/s, and stop-and-go walking without using the corresponding test condition for model training or validation. Overall MAE was 35.3–52.9% lower than that of the uncorrected robot-derived angle. During robot-connected streaming on a Jetson Orin NX, mean algorithm latency was 2.642 ms (99th percentile 3.380 ms; maximum 5.522 ms), and all 15,000 timed samples were below the 10-ms controller period.

Keywords: Exoskeleton Robot, Human-Robot Interaction, Rehabilitation Robot
Abstract: This paper proposes a lightweight k-Nearest Neighbors (k-NN) based real-time gait phase detection framework tailored for resource-constrained embedded environments in wearable hip-assistive robots. Although deep learning algorithms can provide accurate gait detection, their computational and memory requirements often limit deployment on low-level microcontrollers. To address this limitation, the proposed framework utilizes only embedded sensor data from the hip-assistive robot and employs a compressed reference dataset generated using the Grid-based Decision Boundary Sampling (GDBS) algorithm. The proposed framework achieved an accuracy of 93.85%±0.5% using only 100 reference points, while maintaining performance comparable to that obtained with the full 550,000 sample baseline pool. This data compression strategy reduced execution delays and RAM consumption, enabling stable operation within a 1 ms loop on a low-power microcontroller unit for reliable continuous gait phase detection.

Keywords: Exoskeleton Robot, Human-Robot Interaction, Robotic Applications
Abstract: Adaptive ankle assistance requires assistive output to vary according to the user’s activity intensity. This study investigates whether braking-phase energy-related quantities can serve as a compact descriptor of activity intensity and determine push-off assistance. A control framework was developed in which absorbed braking energy is used to estimate assistance magnitude and define the push-off energy budget. Rather than prescribing a time-driven or gait-phase-driven torque trajectory, the proposed method determines assistance from absorbed braking energy, regulates torque according to the remaining energy budget, and generates the assistance profile through interaction with the device. The framework was implemented on an ankle assistive device through hopping experiments involving four participants performing 40-second repeated hopping tasks with varying movement intensities. The results confirmed the intended braking-to-push-off mapping, with strong torque correlation (r=0.957,R^2=0.916) and an energy recovery ratio consistent with the prescribed energy budget (η_fit=0.878,R^2=0.975). Residual deviations from ideal linearity and repeatable torque profiles across cycles indicate practical fidelity for adaptive energy-based assistance. This pilot study will be extended to evaluate physiological benefits and to other locomotor tasks.

Keywords: Exoskeleton Robot, Robot Mechanism and Control
Abstract: More than 100,000 individuals in Japan are estimated to suffer from spinal cord injuries, and patients with severe lower-limb paralysis often rely on wheelchairs for mobility. To support future exoskeleton-type walking assistance systems for such users, a human-scale postural control platform was developed. The proposed platform employs rocker-shaped feet instead of actively controlled ankle joints and consists of four degrees of freedom corresponding to the hip and knee joints of both legs. Standing stabilization is achieved through knee joint torque control combined with gravity compensation. Numerical simulations and hardware experiments were conducted to evaluate the effectiveness of the proposed method. The results demonstrated that the platform could maintain a stable upright posture and recover from external disturbances without active ankle control. These findings confirm the feasibility of the proposed approach as a preliminary step toward the development of autonomous balance control for exoskeleton-type walking assistance systems.

Keywords: Human-Robot Interaction
Abstract: This paper introduces a VAD-based kinematic evaluation framework for generated gestures in companion robots, where VAD represents valence, arousal, and dominance. The proposed system generates robot gesture sequences from emotion-dominant idiomatic expressions using a large language model and evaluates whether the generated motions show affective tendencies consistent with the predicted emotion profile. The framework maps measurable kinematic features of each gesture into a VAD space. Arousal is estimated from motion velocity, acceleration, and hold ratio. Dominance is computed from range of motion and vertical upward motion tendency, while valence is estimated from jerkiness, irregularity, and burstiness. The resulting motion-derived VAD values are compared with reference VAD values obtained from the emotion profile classified by the language model. These reference values are used as affective tendency points rather than definitive ground-truth labels. For preliminary quantitative evaluation, a controlled dataset was constructed using eight emotion categories, five idioms per emotion, and five generated attempts per idiom, resulting in 200 gesture samples. The proposed framework provides a measurable basis for analyzing affective tendencies in generated gestures and supports future reinforcement-learning-based motion refinement.

Keywords: Sensors and Signal Processing, Robotic Applications
Abstract: Reliable wireless communication is essential for teleoperated mobile robots operating in indoor disaster environments. However, walls, shelves, corridors, and non-line-of-sight propagation can cause rapid degradation of the received signal strength indicator (RSSI), resulting in communication blackout and unstable teleoperation. This paper proposes a learned communication-blackout-aware model predictive control (MPC) framework for mobile relay robot deployment. A lightweight blackout-risk estimator is trained using simulated RSSI transition samples and predicts the probability of future blackout from safe RSSI, RSSI rate, inter-node distance, obstacle blocking count, and RSSI deficit. The estimated blackout risk is incorporated into the MPC cost function together with RSSI violation and safety-guard terms, and an RSSI-safeguarded action selection limits excessive instantaneous RSSI degradation caused by the learned risk term. Simulation results in a warehouse-type indoor environment show that the proposed method achieves the highest average worst-link RSSI, the lowest RSSI violation count, and the lowest mean blackout risk compared with distance-based deployment, connectivity-aware reactive control, and RSSI-constrained MPC.