Keywords: Autonomous Decentralized Systems
Abstract: Multi-scale approaches to the control of large-scale multi-agent systems, in which design is carried out across the microscopic agent-level dynamics and the macroscopic density of the population, have produced scalable strategies with provable guarantees. These methods are increasingly relevant to applications ranging from swarm robotics and multi-robot coordination to crowd guidance, environmental monitoring, and distributed surveillance. Yet the deployment of these strategies in realistic settings remains challenged by a set of idealizing assumptions (homogeneity of the agents, absence of safety constraints, availability of many actuated units, static and obstacle-free environments) that rarely hold in practice. In this talk, I will present a coherent body of recent work from my group aimed at relaxing these assumptions and bringing multi-scale multi-agent control closer to real-world deployment. I will first discuss results on the direct control of a single population, presenting robust macroscopic density control of heterogeneous agents and the role of optimal transport as a unifying approach for time-varying multi-agent coverage. I will then turn to the indirect control setting, in which a population of agents is steered through the action of a separate population of controllers. Here I will present multi-scale leader-follower architectures with robust macroscopic guarantees and mean-field control barrier functions as a scalable framework for enforcing safety in stochastic multi-agent systems. I will then address the limits of analytical design by presenting reinforcement-learning-based strategies for sparse leader-follower control, where only a small number of controllers must steer a much larger population, possibly in cluttered environments, together with experimental validation on real robotic platforms. I will close by outlining open challenges in the robust, safe, and sparse multi-scale control of complex collectives.

Keywords: Signal and/or Image Processing, Robotic and Automation Systems
Abstract: Existing open-vocabulary 3D scene graph (3DSG) methods typically require RGB-D inputs or point clouds and are rarely evaluated under overhead or other non-frontal camera views. We present a monocular RGB pipeline that combines promptable segmentation (SAM 3), monocular metric geometry estimation (MoGe-2), and rule-based relation inference, and focuses the quantitative study on context-aware object discovery. The proposed ambiguity filter assigns keep/pending/suppress decisions from scene priors and detection confidence while preserving an audit trail. On four CMU Panoptic scenes, a compact 15-label scene-agnostic prompt bank recovers substantially more of the useful object vocabulary than a minimal 5-label list. Adding the ambiguity filter improves object precision from 0.741 to 0.765 and predicate F1 from 0.366 to 0.379 without lowering recall (ObjR = 0.847). As a secondary finding, a matched-detection relation comparison suggests that 3D-native inference achieves higher precision than 2D heuristics on both dome-camera and frontal views (0.324 vs. 0.220 on CMU Panoptic, 0.530 vs. 0.464 on SUN RGB-D).

Keywords: Advanced Computics, Computational Intelligence, Robotic and Automation Systems
Abstract: Vision-language-action (VLA) models provide generalist robot policies but remain difficult to deploy on edge hardware because their weight memory limits the number of policies that can coexist. Although low-bit post-training quantization offers substantial theoretical compression, common library implementations often reconstruct floating-point weights during execution and therefore provide little persistent memory reduction. We adapt TurboQuant, originally proposed for key-value cache compression, to static model weights. The resulting TurboQuant on Weights (TQ-W) applies a randomized Hadamard rotation and Lloyd-Max scalar quantization independently to each output channel, storing only packed indices, per-row norms, and one shared sign vector. On OpenVLA-7B, TQ-W reduces measured persistent VRAM from 15.13GB to 4.60 GB,correspondingto 3.54× deployable compression. Cached decoding preserves baseline inference speed, while a fused CUDA implementation maintains the low memory footprint without materializing decoded weights. These results support rotation-based quantization as a lightweight path toward multi-policy VLA deployment on edge robots.

Keywords: Intelligent Systems, Robotic and Automation Systems, Intelligent Control
Abstract: While Vision-Language-Action (VLA) models advance end-to-end robotic control, they remain constrained by training distributions. Conversely, frontier Vision-Language Models (VLMs) offer broad reasoning capabilities, but their inference latency precludes high-frequency closed-loop control. This paper proposes a hierarchical framework utilizing VLMs as autonomous teleoperators to generate bimanual manipulation training data without human intervention. Deployed in a physics-simulated environment, VLMs process visual feedback to select symbolic actions—such as specific motor primitives, active arms, and target objects. A shared primitive executor then handles inverse kinematics and physics rollouts. We evaluate three frontier models (Gemini Robotics 1.6, Claude Opus 4.7, and Kimi 2.6) as expert demonstrators across dual-arm tasks. Successful teacher trajectories are extracted to distill Behavioral Cloning (BC) student policies that predict this same symbolic action space. Results demonstrate that student policies successfully achieve real-time reactive control for fundamental tasks at low latencies though they reveal limitations in mimicking the long-horizon sequential reasoning of VLMs for multi-object scenarios. Ultimately, this framework offers a scalable alternative to human teleoperation, providing a benchmark for distilling complex VLM reasoning into deployable controllers.

Keywords: Process Control, Manufacturing Systems, Robust Control
Abstract: This paper addresses robust speed control of conveyor transport under parametric uncertainty and external disturbances. The model predictive control using disturbance observer (MPC-DO) algorithms is based on nominal system models and does not explicitly provide stability for changing parameters. Hence, a robust MPC framework that utilizes a polytopic uncertainty representation with an LMI-based state feedback design is proposed to overcome this deficiency. The feedback gain is then computed using a Lyapunov stability condition to assure closed-loop stability for all admissible uncertainty realizations. In addition, an augmented-state observer is used to estimate unknown disturbances in real-time. The control architecture combines nominal MPC optimization, disturbance compensation, and robust feedback correction. It allows for simultaneous handling of both model uncertainty and external disturbances without explicit invariant set computation, as well as requiring complex constraint tightening. The simulation also shows a feasible solution to track accuracy, stability in all uncertainty scenarios, and reliability in estimating disturbances in a range of disturbance scenarios.

Keywords: Adaptive and Optimal Control, Nonlinear Control, Robust Control
Abstract: This paper explores predictive control strategies for nonlinear systems without accurate mathematical models. We design a Data-Driven Controller (DDC) for an inverted pendulum on a cart. Using Willems’ Fundamental Lemma, the DDC predicts future behavior via a Hankel matrix from input-output data and optimizes control input via Quadratic Programming. We compare the performance of DDC against the conventional Linear Quadratic Regulator (LQR). Evaluated across three scenarios—robustness to model uncertainties, constraint handling, and output noise rejection, the numerical results show that the DDC outperforms the LQR in managing model uncertainties and constraint on control input. However, DDC is more sensitive to output noise.

Keywords: Transportation Systems, Adaptive and Optimal Control, Modeling, System Identification and Estimation
Abstract: At urban intersections, traffic flow is often disrupted by uncooperative driving behavior, particularly at four-way signalized intersections lacking a dedicated turn lane or protected signal phase, leading to congestion, fuel waste, and emissions. In left-hand traffic systems (e.g., Japan), right-turning vehicles must yield to oncoming traffic during the green phase; when no safe gap is available, the turning vehicle stops, potentially blocking following vehicles and causing queue spillback. This paper proposes a Vehicle-to-Everything (V2X) coordination framework with two components. First, a Bayesian Gaussian Process Regression (GPR) model estimates free-flow arrival times of approaching vehicles at the stop line, feeding a slot-search optimizer that identifies the least-delay gap in oncoming traffic. Second, a finite-horizon Quadratic Programming Model Predictive Controller (QP-MPC) controls each Connected and Automated Vehicle (CAV) to precisely meet its assigned crossing time under acceleration and gap safety constraints. A factorial ablation over both components, under identical stochastic traffic conditions, shows the fuel benefit comes primarily from pairing QP-MPC with RSU-assigned crossing times. The proposed system achieves the lowest fuel consumption across all demand levels, a 37% reduction over uncoordinated driving, at the cost of a gap-waiting time longer than IDM-coordinated driving, though still precisely tracked rather than oscillatory.

Keywords: Mechanical Systems Control, Intelligent Control
Abstract: This paper proposes a hybrid control approach for a flexible-link manipulator by integrating actor-critic reinforcement learning (RL) with conventional proportional-derivative-strain (PDS) control. The total control command is determined as the simple sum of the outputs from the actor-critic agent and the PDS controller, aiming to enhance joint positioning performance and vibration suppression. This integration seeks to leverage the advantages of both approaches: the PDS component ensures stability, while the actor-critic agent explores optimal actions that outperform standalone PDS control. The RL state space is constructed using the same feedback variables as the PDS law to incorporate its inherent advantages, such as spillover avoidance and low computational complexity. Numerical simulations were conducted to evaluate the hybrid RL-PDS controller against standalone RL and conventional PDS methods. The results revealed that the hybrid performance was heavily dominated by the PDS component under the evaluated configuration, yielding only marginal improvements. These findings highlight the challenge of optimizing the configurations of the RL and PDS components within an additive hybrid architecture to achieve effective synergy.

Keywords: Mechanical Systems Control, Nonlinear Control, Adaptive and Optimal Control
Abstract: Abstract: This paper investigates the effectiveness of cooperative design of a controller and a static quantizer for an on–off valve driven hydraulic cylinder position control system. In such systems, the control input becomes discrete-valued due to the on–off valve characteristics, and quantization effects may degrade tracking performance. In this paper, the switching thresholds of the on–off valves are modeled as parameters of a static quantizer and treated as design variables together with the controller gains. A nonlinear optimization problem is formulated, and numerical simulations are conducted to compare the cooperative design approach with a conventional controller-only optimization. The results show that, under the considered conditions, both approaches achieve identical control performance, indicating that the cooperative design does not provide additional benefits in this problem setting.

Keywords: Mechanical Systems Control, Nonlinear Control
Abstract: Recently, high-speed and high-precision positioning control for an ultrasonic motor (USM) has been proposed by combining the minimum-time control and the safety-critical control based on a control barrier function. The minimum-time positioning control law is improved using a control barrier function by treating the singular arc as the boundary of a safe set. However, the controller still has discontinuity because the state achieves the singular arc, and the discontinuity is an obstacle for high-precision control. To avoid achieving the singular arc, using an almost-sure control barrier function is effective. Therefore, in this paper, we propose a new quasi-minimum-time control law using an almost-sure zeroing control barrier function. The high-speed, high-precision, and chatter-free positioning properties of the proposed controller are confirmed by simulations and experiments.

Keywords: Mechanical Systems Control, Computer Aided Design, Mechatronics Systems
Abstract: This paper presents an experimental validation of the D4Q algorithm, a data-driven design method for dynamic quantizers, using a wheeled inverted pendulum driven by a brushless DC motor. Since D4Q was originally developed for SISO systems, it cannot be directly applied to a control system with multiple outputs. To address this issue, we investigate two approaches for applying D4Q: one selects a single output from the multiple system outputs, and the other uses the controller output. Then, dynamic quantizers was designed from experimentally collected input-output data and evaluated through control experiments.

Keywords: Mechanical Systems Control, Modeling, System Identification and Estimation, Analytical Measurement
Abstract: This study focuses on small satellites and aims to develop a docking method utilizing an extendable mast. Extendable masts have garnered attention as mechanisms for constructing space structures due to their high extension performance and storage efficiency. However, because extendable masts exhibit non-uniformity in the longitudinal and aperture directions, it is believed that bending and torsional vibrations are expected to occur as a result. Therefore, in this study, the experimental setup enabled the measurement of both bending and torsional vibrations of the extendable mast, and vibration tests were conducted using impulse excitation and laser displacement sensors. The results demonstrate that low-frequency bending and torsional vibrations can be captured by reducing the influence of mechanical friction under a simulated microgravity environment using air bearings. Based on the obtained frequency response data, second-order transfer functions were identified, which inherently incorporate the dynamical influences of the tip mass.

Keywords: Mechanical Systems Control, Robotic and Automation Systems, Intelligent Systems
Abstract: Planetary exploration rovers require high traversability to operate in harsh, unknown environments, such as uneven ground and soft ground on planetary surfaces. This study investigates the flipper (fin-shaped propulsion mechanism) contact angle as a key design parameter governing locomotion on deformable terrain. Through combined single-flipper traction experiments and quadrupedal walking tests, we show that mobility efficiency is maximized at contact angles of 20–30 degrees, where the lift–drag balance is optimized. These results establish a quantitative design guideline for variable-contact-area mechanisms in planetary rovers.

Keywords: Mechanical Systems Control, Robust Control, Mechatronics Systems
Abstract: This paper presents a dual-channel 128-PSK communication system based on a Spectrum Observer (SpO) and implemented on an STM32G474 (NUCLEO-G474RE) microcontroller (MCU). The transmitter precomputes a 128- symbol loop and replays a reference carrier and a phase-modulated carrier by dual-DAC circular DMA at 1 MHz. The receiver samples the two channels synchronously, applies two SpOs, forms a relative phase vector, and exports the quantized QCAP vector as a CSV capture through a PC-side capture interface. The raw phase angle is computed from the captured vector for recording, while final symbol-window evaluation is performed offline. With a 10 kHz carrier, 100 samples per symbol, and a 1000-sample capture, the experimental results indicate sufficient relative-phase accuracy for 128-PSK discrimination under the tested condition.

Keywords: Mechanical Systems Control, Computational Intelligence
Abstract: Embodiment is an important concept in embodied AI, yet the influence of passive body dynamics on learning remains insufficiently explored. This study investigates how passive properties of the body affect model-based deep reinforcement learning for bipedal locomotion. We compared two simulation models: a ``passive model'' incorporating elastic elements and high backdrivability, and a ``torque model'' representing conventional humanoids. Our results show that while the passive model required more time to maximize rewards due to the influence of physical attractors, it achieved significantly faster convergence to stable limit cycles. Furthermore, the obtained locomotion in the passive model was more energy-efficient and robust against terrain variations, such as slopes, compared to the torque model. These findings demonstrate that integrating passive properties allows robots to exploit stable limit cycles for high-performance locomotion, highlighting the importance of hardware-software co-design in future embodied AI.

Keywords: Mechanical Systems Control, Signal and/or Image Processing, Identification and Estimation
Abstract: Acoustic levitation of objects larger than the acoustic wavelength using a circulating single-focus field is susceptible to positional oscillations. This paper proposes a feedback-controlled system for stabilizing an expanded polystyrene sphere approximately 50 mm in diameter. Nine airborne ultrasound phased arrays generate a bowl-shaped acoustic field by spatio-temporal modulation. Two high-speed cameras estimate the three-dimensional position and velocity of the sphere, and a PID controller with delay compensation updates the center of the focal trajectory in real time. Experiments comparing the proposed system with a conventional fixed-field system showed that the standard deviations in the x, y, and z directions were reduced to 0.74, 0.66, and 0.58 times those of the fixed-field system, respectively. The results demonstrate improved levitation stability, particularly in the vertical direction.

Keywords: Mechanical Systems Control
Abstract: Inverse Reinforcement Learning (IRL) aims to infer reward functions from expert demonstrations, but conventional IRL methods often suffer from high computational cost in continuous control tasks. In particular, Maximum Entropy IRL requires repeated policy optimization and high dimensional reward weight estimation, resulting in low computational efficiency. In this paper, we propose a lightweight IRL method based on feature selection. Instead of estimating reward weights over multiple features, the proposed method constructs the reward function using only the most important feature selected by statistical evaluation. This simplification eliminates iterative weight estimation and reduces computational complexity. We evaluate the proposed method on the Halfcheetah task in Gymnasium and compare it with a conventional feature-selection-based IRL method. Experimental results show that the proposed method achieves higher discounted return while significantly reducing computation time. In particular, the proposed method reduces computation time to less than one-tenth of that required by the conventional method. These results suggest that, in tasks where a dominant feature exists, a simple single-feature reward function can effectively capture the essential task structure. The proposed method provides an efficient alternative to conventional IRL approaches and offers a new perspective on low-dimensional reward modeling.

Keywords: Adaptive and Optimal Control, Intelligent Control, Identification and Estimation
Abstract: This paper focuses on data-driven model matching problem, and discuss the structure selection of feedback controller.

Most existing methods on data-driven model matching consider a parameter tuning with the controller structure fixed a priori. However, an appropriate selection of controller structure is not trivial. This paper proposes a combination of Factorized VRFT and model reduction to select an appropriate controller structure. The effectiveness of the proposed method is demonstrated through a practical experiment with a DC motor.

Keywords: Adaptive and Optimal Control, Guidance and Flight Control, Computational Intelligence
Abstract: This study investigates parameter tuning of a candidate-based ship collision avoidance algorithm by multi-objective evolutionary optimization. The parameter set includes preference coefficients for course and speed changes, directional ship-domain parameters, the TCPA risk window, and route-recovery parameters. The tuning problem is formulated as a bi-objective minimization of the worst scenario safety objective and the mean scenario efficiency objective over the 22 Imazu benchmark problems. MOEA/D is applied to search for non-dominated parameter sets from five optimization runs. The optimized solutions are evaluated against a manually designed baseline parameter set. The results show that the five runs converged to a similar trade-off region. The representative MOEA solution removed the goal-offset violations observed in the manual baseline, increased the worst minimum separation, and reduced the mean efficiency objective under the present Imazu evaluation setting.

Keywords: Adaptive and Optimal Control, Mechanical Systems Control, Nonlinear Control
Abstract: Extremum seeking control (ESC) is a method for real-time optimization of an unknown function. However, when the target system has dynamics, increased phase lag due to deterioration may degrade or prevent proper ESC operation. This study proposes an ESC method that regulates the phase difference using a phase comparator and adjusts the perturbation frequency accordingly. The proposed method adaptively shifts the operation to a frequency range where ESC remains effective, thereby maintaining convergence performance even under increased phase lag.

Keywords: Adaptive and Optimal Control, Multivariable Control
Abstract: In this paper, we propose an improved optimality level certification method for parallel model predictive control (MPC). The proposed method achieves both an optimality certificate that guarantees a given level of optimality for an intermediate solution (optimization vector) and a reduction of computational load. Unlike prior methods that parallelized both primal and dual problems, our approach parallelizes only the primal problem and efficiently constructs a feasible dual solution to perform optimality certificate. An experiment using micro:bit V2 devices demonstrates that the proposed method works correctly, giving an intermediate solution that satisfies a given level of accuracy.

Keywords: Adaptive and Optimal Control, Mechanical Systems Control, Modeling, System Identification and Estimation
Abstract: This paper proposes a data-driven design method for a gain-scheduled I-PD controller for vehicle yaw-rate tracking. Vehicle lateral dynamics exhibit nonlinear characteristics and depend on driving conditions, making it difficult to obtain an accurate mathematical model. Furthermore, from a drivability perspective, rapid fluctuations in the control input are undesirable. To address these issues, a virtual reference feedback tuning (VRFT)-based design method for a gain-scheduled I-PD controller is introduced. In the proposed approach, the controller gains are scheduled according to vehicle velocity, and the gain-scheduling parameters are tuned without explicit plant model identification. In the I-PD structure, the integral action is applied to the tracking error, while the proportional and derivative actions are applied to the output signal. This structure suppresses abrupt changes in the control input induced by variations in the reference signal. Numerical simulations are conducted using a CarSim vehicle model. The proposed gain-scheduled I-PD controller is compared with a gain-scheduled PID controller. The results demonstrate that the proposed controller suppresses sharp peaks in the steering input while maintaining good tracking performance with respect to the desired response.

Keywords: Adaptive and Optimal Control, Intelligent Control, Transportation Systems
Abstract: This paper proposes a hierarchical safety control method for vehicle obstacle avoidance problems using model predictive control (MPC) and control barrier functions (CBFs). In the proposed method, a reference trajectory is first generated by solving an MPC problem with a relaxed barrier function using iterative linear quadratic regulator (iLQR). Then, a discrete-time CBF-based optimization problem is solved to modify the control input for safety. To reduce computational complexity, the system dynamics and the CBF constraints are linearized around the reference trajectory, and the resulting problem is formulated as a quadratic programming (QP) problem with linear constraints.

The effectiveness of the proposed method is verified through simulations of a highway overtaking scenario using a front-wheel steering vehicle model.

Keywords: Adaptive and Optimal Control, Simulation of Large Systems, Safety, Environment and Eco-Systems
Abstract: Guide frames developed for tunnel wall inspection can automatically change their shape, allowing them to pass through obstacles in tunnels. However, in emergencies or when the frame shape needs to be changed on the spot depending on the shape and placement of obstacles, a method for easily and quickly controlling the frame shape was required. Therefore, a method for manually changing the shape of the guide frame according to the shape of the obstacle was devised and verified through analysis and experimentation. The guide frame is an arch structure in which variable-shape trusses (VGTs) consisting of linear actuators and hinges are connected in a straight line. A manual operation method using feedback control was devised in which the frame shape changes according to the force acting on a puller installed on the underside of the guide frame. The target shape for avoiding obstacles was performed using spline interpolation, and the shape of the guide frame was changed according to that target shape. This control method allows for easy manual deformation of the guide frame and is applicable to various shapes.

Keywords: Robust Control, Robotic and Automation Systems, Nonlinear Control
Abstract: This paper presents a path following control method for tracked vehicles based on feedback linearization with explicit consideration of track velocity dynamics. Tracked vehicles achieve turning by generating a speed difference between the left and right tracks, and the dynamics of the individual track velocities must be treated explicitly for precise control. We adopt a dynamic model of the left and right track velocities driven by their respective driving forces with speed-dependent friction, and formulate the path following problem in the Serret--Frenet frame. The relative degree from the path following error to the left--right driving force difference is shown to be three, and a nonlinear control law based on feedback linearization is derived to achieve asymptotic convergence of the error. A notable structural feature is that the derivative of the path curvature appears explicitly in the control law through the feedback linearization procedure, which may be beneficial for transient response on paths with rapidly varying curvature. Furthermore, the force allocation for constant-speed maintenance and the force difference for path convergence are algebraically separated. A numerical simulation illustrates the feasibility of the proposed method.

Keywords: Robust Control, Nonlinear Control, Social Systems
Abstract: This paper investigates mean-field games in large-scale stochastic delay systems with infinitely many decision-makers. The stabilization problem is formulated under memoryless state-feedback strategies depending on individual and mean-field states. Upper-bound Nash equilibrium strategies are obtained by minimizing an upper bound of the cost function through the Karush-Kuhn-Tucker conditions, yielding coupled Sylvester-type matrix equations (CSMEs). To reduce memory requirements for large populations, we develop two decentralized reduced-order strategy sets based on the population mean and the sample mean of the player-group size. By introducing new coefficient matrix parameters, we reveal a trade-off between the influence of state delay and the existence of a strategy set in the infinite-population limit. Numerical experiments confirm the effectiveness of the proposed method for sufficiently large populations.

Keywords: Robust Control, Guidance and Flight Control, Nonlinear Control
Abstract: This paper proposes an improved sliding mode control (SMC) method for quadrotor systems by incorporating a disturbance observer (DOB) to enhance robustness against external disturbances and modeling uncertainties. In many industrial applications, proportional–integral–derivative (PID) control is widely used for drone control due to its simplicity. However, PID controllers often require parameter retuning when disturbances occur or when modeling errors become significant. SMC is known as a robust control method that can handle such uncertainties, and its stability is guaranteed based on Lyapunov theory. Nevertheless, conventional SMC design tends to be conservative because it relies on the assumed upper bound of disturbances. To address this issue, the proposed method integrates a DOB into the SMC framework to estimate disturbances in real time and directly compensate for them in the control input. Simulation results demonstrate that the proposed method achieves superior control performance compared to PID control and conventional SMC in terms of tracking accuracy and disturbance rejection.

Keywords: Robust Control, Adaptive and Optimal Control
Abstract: This paper proposes a vibration suppression method for two stage control using Linear Active Disturbance Rejection Controller (LADRC). Two stage control is a control strategy that switches position control and velocity control depending on the magnitude of the position error between the reference and the current position. However, the conventional method exhibits vibrations near the reference position in experiments with an actual system. This vibration may be caused by the quantization error of the encoder that measures the position in actual system. To suppress this vibration, we introduce a dead zone into the error between the Extended State Observer (ESO)-estimated position and the encoder-measured position. Experimental results using an actual electric actuator confirm that the proposed method effectively reduces steady-state vibration without degrading control performance.

Keywords: Robust Control
Abstract: This paper presents a synthesis framework for low-gain, strongly alpha-stabilizing controllers in single-input-single-output (SISO) linear time-invariant (LTI) systems. By requiring the controller itself to be stable and forcing all closed-loop poles to have real parts strictly less than -alpha (alpha>0), this paradigm guarantees enhanced transient performance, including faster rise times. However, traditional synthesis methods for such controllers often yield excessive control gains, inevitably increasing the risk of actuator saturation. To circumvent this limitation, we propose a parameterized second-order band-pass control architecture. Formulating the synthesis as a constrained nonlinear optimization problem, we employ sequential quadratic programming (SQP) method to execute a dynamic "margin-trading" strategy. This approach systematically minimizes the controller's H-infinity norm by actively exploiting the residual phase change rate (PCR) budget. A numerical example validates the proposed framework, demonstrating superior gain reduction that becomes increasingly pronounced even under stringent convergence rate constraints.

Keywords: Robust Control, Computer Aided Design
Abstract: This paper presents an input partitioning approach to refine the safety verification of neural network controllers. In semidefinite programming (SDP) based verification frameworks, the abstraction of non-linear activation functions via overarching quadratic constraints inherently introduces conservatism, which may result in inconclusive results. To overcome this limitation, we first provide a rigorous mathematical proof establishing that partitioning the input space monotonically refines the reachable set approximation when utilizing local quadratic constraints. Motivated by this theoretical guarantee, we propose two systematic input partitioning algorithms that dynamically refine the input domain to reduce the conservatism. Finally, numerical experiments demonstrate that the proposed methods substantially tighten the overapproximated reachable sets, effectively mitigating inconclusive verifications and enhancing the overall reliability of the verification process.

Keywords: Robust Control, Nonlinear Control, Adaptive and Optimal Control
Abstract: This paper reveals a stability gap in DOB gain design for cascade MPC servo systems: the linear stability boundary far exceeds the nonlinear safe boundary, rendering windup invisible to conventional analysis. When MPC replaces PI, integrator anti-windup protection disappears, and the saturation error Δi couples unchecked into the Q-filter, triggering DOB windup. The augmented state matrix of the DOB is found to have a rank-one structure, and the matrix determinant lemma yields an exact factorization of the characteristic polynomial: det(sI − Acl(α)) = Δ0(s)·(1 − αh(s)), where h(0) = 0. This factorization provides: (1) a critical gain αcrit = 1/‖h‖∞ as an explicit Nyquist-type boundary; (2) an ℒ2 contraction-mapping condition α < 1 guaranteeing anti-windup stability; (3) a stability gap αcrit ≫ 1 quantifying the invisibility of windup to linear analysis. As the Q-filter bandwidth g increases, αcrit → 1⁺ and the gap shrinks, exposing a fundamental speed-robustness trade-off. Monte Carlo experiments with 500 parameter sets show a 50−74% instability rate at α = 1 and 2−19% at α = 0.9; within the robustly stable region α ≤ 0.8, α ≈ 0.7 never underperforms the no-DOB baseline across all five step angles (e.g., 14.8% reduction at 90°), and avoids the compensation degradation observed for α = 0.8 at 360°, making it the best overall compromise.

Keywords: Robust Control, Process Control, Nonlinear Control
Abstract: It is important to have a control system robustly stable for plants with uncertainty. Various researchers consider a condition to design a control system with robust stability for plants with uncertainty. A lot of researches are assumed to make the uncertainty have no number of unstable poles. Thus, these researches cannot be applied to plants with varying number of unstable poles such as the large flexble structure or nano air vehicle. Some researches provides a necessary and sufficient condition that a control system is robustly stable for a plant with varying number of unstable poles. However, these design method of a control system for plants with varying number of unstable poles often provide a conservative controller. To overcome these problems, it is important clarifying a condition that a high-gain feedback controller is designed for a class of plant with varying number of unstable poles. In this paper, it is clarified a condition of a feedback controller that is designed for a class of plants with varying number of unstable poles.

Keywords: Guidance and Flight Control, Robotic and Automation Systems
Abstract: This paper addresses the autonomous landing of unmanned aircraft and proposes a vision-based autopilot system using cockpit-view images. In our previous work, an aircraft was able to land autonomously on airport runways by following instructed flight trajectories; however, it is difficult to recover once the aircraft deviates from the trajectory. To address this issue, flight states associated with trajectory deviations and aircraft attitude are first estimated from cockpit-view RGB images and then used as inputs to the flight controller. Both the flight state estimator and the flight controller are implemented using convolutional neural networks (CNNs). Furthermore, the flight controller is trained via imitation learning using DAgger. The simulation results demonstrate that the aircraft equipped with the proposed autopilot system can effectively correct trajectory deviations and successfully land on a runway.

Keywords: Multivariable Control, Modeling, System Identification and Estimation
Abstract: Data-driven control has attracted attention as a rational approach that does not rely on explicit model construction, and as a remedy in model-based control when system identification is difficult. The concept of the data-informativity has enabled analysis and design directly from data. However, existing approaches often assume that the data dimension matches the system order, which limits their applicability in the analysis and design problems. This paper considers an algebraic generalization of informativity for stabilization and deadbeat control that is independent of the data dimension. Based on the geometric approach and algebraic techniques, we derive generalized conditions for informativity-based design.

Keywords: Multivariable Control
Abstract: This paper is concerned with the output-feedback controller synthesis for linear infinite-dimensional systems on Hilbert spaces via linear operator inequalities. We are intended to extend the LMI-based approaches to synthesis output-feedback controllers based on the change-of-variables methods for finite-dimensional systems to infinite-dimensional systems on Hilbert spaces. The exponential stabilization with a specified decay rate is considered in this paper and a linear operator inequality condition is provided. A crucial difference from finite-dimensional systems is that infinite-dimensional systems involve an unbounded state operator in general, which is then included in the operator inequality. Moreover, the controller constructed from the solution to the operator inequality needs to satisfy certain `well-posedness' condition so that the state operator of the controller generates a C_0 semigroup. We provide a sufficient condition for the well-posedness in this sense. The proposed linear operator inequality is applied to a PDE--ODE system with a numerical example.

Keywords: Multivariable Control, Mechanical Systems Control, Robotic and Automation Systems
Abstract: This paper proposes a geometric parameter design method for coverage control in multi-leader-follower systems. Leader agents perform Voronoi-based coverage control, whereas follower agents support coverage through geometric relationships with the assigned leader and the other followers. The control parameters are designed from a desired follower configuration and a representative inter-agent distance. Numerical simulations in non-weighted and weighted domains show that the proposed method maintains the desired follower formation and substantially reduces the computational cost, at the cost of a moderate loss of coverage accuracy compared with conventional Voronoi-based coverage control.

Keywords: Process Control, Flow Measurement and Control
Abstract: We have developed a concentration adjustment system that regulates hydrogen concentration at demand sites according to specific requirements within a mixed gas grid where hydrogen concentration fluctuates. This paper presents a control system capable of maintaining hydrogen concentration within the desired range even during transient changes in the grid’s hydrogen concentration. The system achieves this by switching piping configurations through a hydrogen separation device and four valve operations. Experimental validation using a test facility demonstrated that the system effectively controls the mixed gas concentration supplied to the demand site within a target variation of ±1.4%, despite low supply gas flow rates and hydrogen concentration fluctuations. These results confirm the effectiveness of the proposed mixed gas concentration control method.

Keywords: Process Control, Robust Control, Process Automation
Abstract: This paper presents a hierarchical fault-tolerant control framework to mitigate the impact of faults in subsea seawater injection systems. The approach integrates a supervisory real-time optimization (RTO) layer that determines setpoints based on desired performance and safety constraints, and a stochastic offset-free nonlinear model predictive control (NMPC) layer that compensates for faults using estimates from an LMI-based nonlinear observer. Numerical simulations demonstrate the effectiveness of the proposed framework and highlight its potential for real subsea applications.

Keywords: Process Control, Modeling, System Identification and Estimation, Mechanical Systems Control
Abstract: This paper investigates the water level control systems for flash chamber and evaporator in a hybrid hot spring thermal energy conversion (H-STEC) plant. In the model construction of water levels, the static and dynamic characteristics of the discharge mass flow rate are taken into account. For the plant models, considering the valve characteristics, a PID controller is designed for the flash chamber and a PI controller is designed for the evaporator, respectively. The anti-windup compensation scheme is also introduced to suppress the overshoot of the water levels. The effectiveness of the proposed control systems was confirmed by numerical simulation.

Keywords: Computer Aided Design, Adaptive and Optimal Control
Abstract: This paper presents a constrained optimization framework for tuning fractional-order PID (FOPID) controllers. Given a prescribed phase margin, the design is formulated as the maximization of the gain crossover frequency subject to frequency-domain loop-shaping constraints, including unity open-loop gain and a specified phase margin at the crossover frequency, together with a phase-flatness condition to reduce sensitivity to gain perturbations. The controller parameters, comprising proportional, integral, and derivative gains as well as fractional orders, are obtained by solving a nonlinear, nonconvex optimization problem. Fractional operators are implemented via Oustaloup’s band-limited rational approximation, introducing additional considerations on approximation bandwidth. To address nonconvexity and local minima, a multi-start strategy with bounded decision variables is employed, and a lower bound on the integral order is imposed to preserve low-frequency gain. Comparative simulations with a representative analytical tuning method show that the proposed approach achieves higher gain crossover frequencies under identical phase-margin specifications, resulting in faster closed-loop responses while satisfying robustness constraints. These results demonstrate the effectiveness of direct numerical optimization for FOPID tuning without restrictive structural assumptions.

Keywords: Computer Aided Design, Adaptive and Optimal Control, Multivariable Control
Abstract: A MATLAB-based toolbox for parallel model predictive control (MPC) is developed to facilitate MPC implementation on resource-constrained computational devices. It features parallel MPC simulation and code generation for deployment on micro:bits. The generated code is developed in-house and does not require external math/optimization libraries.

Keywords: Power Systems Control
Abstract: Recently, with the spread of renewable energy resources, power flow and voltage fluctuations have increased, leading to power quality issues such as grid congestion and voltage deviations. These problems are localized phenomena, requiring countermeasures at each location. In response, discussions have begun regarding reactive power markets, which define the market value for the voltage control function of distributed energy resources. This study expands the reactive power market mechanism and evaluates cases of (i) reactive power trading and (ii) parameters trading of a voltage controller (Volt-Var control function) that adjusts reactive power according to local voltage deviations. Simulations using a simplified model, verified the effective mechanism of the market and the improvement in social welfare through increased renewable energy utilization rates.

Keywords: System Engineering, Relationality-Oriented Systems Design, System Integration
Abstract: This paper introduces a tunable inerter whose inertance can be adjusted via an epicyclic gear train. The inerter is discovered from the mechanical–electrical system analogy and has been adopted to improve the performance of various mechanical systems in a fully passive manner. Since the optimal inertance may vary significantly with operating conditions, a fixed inertance often limits achievable performance. To address this issue, we develop a tunable inerter based on an epicyclic gearing mechanism, allowing the effective inertance to be adjusted to match varying system requirements. In this paper, we derive the ideal inertance and verify the proposed inerter’s properties by experiments, thereby demonstrating its potential for practical vibration and disturbance mitigation.

Keywords: Robotic and Automation Systems, Mechanical Systems Control, Sensors and Transducers
Abstract: This study develops a remote welding system aimed at reducing the physical burden and inherent dangers faced by skilled workers in welding environments. Previous research on automatic welding systems recognizes all weldable locations, making it difficult to automatically identify and isolate only the specific areas for welding. In contrast, our proposed system allows the operator to intuitively and manually specify only the target positions they want to weld directly on an RGB image. The system is built around a 6-DOF robot arm equipped with an RGB-D camera. Initially, the specified path on the RGB image is converted into 3D coordinates utilizing depth information. Since it’s difficult to capture complex object shapes from a single viewpoint, our method can also generate welding paths by integrating point clouds acquired from multiple viewpoints. Furthermore, to improve welding accuracy, ranging errors are corrected using a touch sensor that utilizes the welding wire. Additionally, to minimize the robot’s total travel distance, the generated welding path is optimized based on the Traveling Salesman Problem (TSP). As a result, the proposed method enables highly accurate path generation even for complex shapes that cannot be fully captured from a single viewpoint.

Keywords: Robotic and Automation Systems, Mechatronics Systems, Mechanical Systems Control
Abstract: This study presents a flexible leg mechanism for multi-legged robots intended to move in real-world disaster environments. The mechanism consists of serially connected triangular-prism links, a central silicone tube, and three strings driven by servo motors, enabling the leg to bend in six directions. The use of flexibility and redundancy is expected to help the robot adapt to contact with rubble and uneven surfaces. Forward kinematics were derived to estimate the foot-tip position, and a prototype was fabricated for motion experiments. The measured foot-tip positions differed from the geometric model because of string tension loss, friction between parts, and gravity, indicating that mechanical factors should be included in future modeling. In addition, a two-leg experimental robot using the proposed mechanisms was tested on different floor conditions. Although it did not move forward without casters, forward motion was confirmed when casters were attached, especially on a rubber sheet. These results clarify the basic motion characteristics and remaining issues for improving the flexible leg mechanism.

Keywords: Guidance and Flight Control, Robotic and Automation Systems, System Integration
Abstract: This research presents a human-following control method for UAVs using onboard camera images. In this work, YOLOv8‑Pose.engine is used for human pose estimation. The distance between the UAV and the target person is calculated from the torso length obtained from the pose model. The UAVs is controlled to track the target while maintaining the desired distance. Conventional approaches using Haar Cascade or OpenPose for face‑based recognition often suffer from misidentification depending on the target orientation. Furthermore, bounding-box based following methods, which estimate the distance based on the rectangular region enclosing the target from an image, are highly sensitive to variations in the bounding box size caused by the target’s posture, such as when the target spreads their arms or walking sideways. In this work, we propose a method that utilizes body keypoints obtained from the YOLOv8‑Pose.engine to achieve more accurate distance and real‑time human tracking. In tracking control, an Exponential Moving Average (EMA) filter is employed to reduce control errors. The proposed method is demonstrated through experimental tests conducted in indoor environments.

Keywords: Nonlinear Control, Mechanical Systems Control
Abstract: Wireless-based control systems are widely used in industrial and robotic applications because of their flexibility and ease of implementation. However, when such systems are exposed to cyberattacks, malicious manipulation of communication signals may cause unintended actuator behavior and potentially lead to serious injury to users. To improve cybersecurity, encrypted control has attracted attention as a method for protecting control signals and system parameters from unauthorized access and tampering. In encrypted control systems, appropriate cryptographic key-length selection is important. Although a longer key length generally improves security against brute-force attacks, excessively large key lengths increase computational cost and communication delay, which may negatively affect real-time control performance. Therefore, evaluating the influence of key length on system behavior is necessary for practical applications. This poster investigates the effects of excessive key length on control performance using an antagonistic actuator driven by McKibben-type pneumatic artificial muscles. The control system employs the ElGamal cryptosystem for encrypted control computation and communication. Experimental results demonstrate that excessively large key lengths degrade tracking performance because of increased computational overhead in the encrypted control process. These results highlight the trade-off between cybersecurity and control performance in encrypted control systems.

Keywords: Robotic and Automation Systems, Manufacturing Systems, System Engineering
Abstract: Robotic cellular manufacturing systems require efficient layout and manipulator path planning to reduce cycle time and improve productivity. Existing geometric-constraint-based methods can jointly optimize facility layout and manipulator paths. However, such optimization processes often depend on randomly generated initial layouts, and many trials may be required to obtain a high-quality solution. This paper presents a topology-guided initial layout generation method for robotic cellular manufacturing systems. The proposed method represents the task sequence and facility relationships as a layout topology graph and generates structured initial layouts by considering adjacency, task order, and geometric symmetry in facility arrangements. These initial layouts are then used as initial candidates for a subsequent joint optimization process, where facility positions and manipulator paths are refined under collision and workspace constraints. Compared with purely random initialization, the proposed method is expected to reduce unpromising initial layouts and improve the efficiency of the optimization process. Numerical experiments on robotic cellular manufacturing systems are conducted to compare the proposed initialization with random initialization in terms of optimization time, final cycle time, and solution stability. The results are expected to demonstrate the potential of topology-guided initialization for efficient layout and motion planning optimization.

Keywords: Mechatronics Systems, Mechanical Systems Control
Abstract: Because of their highly compact body structure, ultra-compact mobility vehicles readily transmit vertical and pitch directional vibrations originating from road surface irregularities to the occupants, thereby leading to a significant deterioration in overall ride comfort. To improve the ride quality, the authors have proposed a two-degree-of-freedom active seat suspension system driven by a voice coil motor (VCM). A VCM generates thrust when an electric current flows through its coil; however, it inherently possesses specific characteristics, such as fluctuations in the thrust constant and operational voltage limitations. Additionally, the target object to be controlled exhibits strong nonlinearity. These combined factors are considered to substantially affect the system's vibration suppression performance. Therefore, in this report, we investigated the influence of these specific factors on the relationship between the controller and the vibration suppression effectiveness through comprehensive analyses assuming realistic driving environments.

Keywords: Robotic and Automation Systems
Abstract: River channel blockage caused by landslides poses serious flood risks requiring immediate drainage response. This paper presents i-CentiPot-Aqua, a flexible multi-legged robot based on the “open design” concept for rapid disaster deployment. The robot is compact and lightweight during transport, but grows by pumping water into flat hoses as structural material, acquiring sufficient mass and volume for field operation. A key challenge was achieving balanced locomotion in both land and water environments. We developed a slider-crank leg mechanism that orients the flexible fin-shaped legs vertically during the power stroke and horizontally during the recovery phase, suppressing unwanted propulsive drag in submerged conditions. The improved prototype achieved locomotion speeds of 0.131 m/s on land and 0.056 m/s in water — the water speed reaching 42% of land speed, a significant improvement over the initial prototype (18%). The robot also demonstrated successful transition from a water environment onto a 20-degree slope, validating the feasibility of swarm-based drainage deployment for river channel blockage response.

Keywords: Robotic and Automation Systems, Molecular Robotics
Abstract: This paper reports on the ongoing development of BRAINS (Base containeR with open design Adaptive INtegration System), a general-purpose transport container platform that enables on-site physical reconfiguration of heterogeneous robot modules for disaster response. BRAINS is a general-purpose transport container platform that enables on-site physical reconfiguration of heterogeneous robot modules for disaster response, grounded in the “open design” philosophy that separates robots into interchangeable work machine modules (upper body) and mobile machine modules (lower body). In this work, we present three major advances over the previous prototype. First, a wire-suspended lift mechanism incorporating horizontal compliance enables robust docking of heavy modules on uneven terrain without relying on high-precision positioning. Second, AR marker-based auto-docking control reduces operator workload during module exchange. Third, an improved robot carrier featuring a motorized ramp enables autonomous loading and unloading of mobile sub-units. Additional validation experiments conducted at Osaka Institute of Technology facilities confirmed improved automation reliability and reduced operator intervention. The complete module reconfiguration sequence from wide-area search to drainage phases was demonstrated successfully, further substantiating the practical utility of the open design approach for adaptive disaster robotics.

Keywords: Autonomous Decentralized Systems, Robotic and Automation Systems, Intelligent Control
Abstract: In the field of mobile robotics, source seeking refers to designing a controller that enables a robot to find the location of the maximum concentration in a two- or three-dimensional concentration field, by repeatedly measuring the concentration at its current position and moving accordingly. Here, we address cooperative approaches for source seeking by two mobile robots that exchange their measurement information. Deployed in a two-dimensional environment, the robots seek the location of the maximum concentration using a bacterial run-and-tumble controller based on local concentration observations. In this poster, we assume that the two robots exchange their local concentration measurements at each step, and we evaluate the performance of this information exchange strategy by comparing it with the single-robot case. Simulation results in a concentration field expanding concentrically from the origin indicate that this cooperative approach reduces the time to reach the vicinity of the maximum concentration by approximately 5.5% compared with the single-robot case.

Keywords: Mechanical Systems Control, Mechatronics Systems
Abstract: This study proposes a design framework for an active controller to be embedded in a lever-type mechanism with inertia amplification. The analytical control model for the lever-type mechanism printed by a 3D printer with inertia amplification is presented. The real printed model has several advantages, such as easy maintenance, high reliability, scalability, and high performance in vibration isolation with a wide stopband. However, the lever-type mechanism has the potential to cause undesirable vibrations in motion dominated by the lower frequency range because it possesses the property that the resonance frequency range is lower than the anti-resonance frequency range. Proper design of the feedback (FB) and feedforward (FF) control enables the shifting of resonance and anti-resonance frequency. The main contribution of the study is demonstrating the effective range of parameters FB and FF through numerical simulations.

Keywords: Robotic and Automation Systems, Mechatronics Systems, Mechanical Systems Control
Abstract: The effects of body length, flexibility, and leg phase difference on the path-following performance and straightness of a super-long multilegged robot were investigated using a three-dimensional physical simulation model. The robot consists of multiple units connected by inter-segment joints, with straight locomotion controlled by angular phase differences between the fore and hind legs of each unit. Path-following performance was evaluated based on the trajectory error of each trailing unit relative to the leading unit, while straightness was evaluated by lateral deviation. Simulations were performed in MuJoCo with varying numbers of units, joint stiffness values, and fore–hind leg phase differences. The following error increased with robot length, especially under low-stiffness conditions. Higher joint stiffness generally improved both following performance and straightness. However, for certain phase differences, following performance deteriorated even under high-stiffness conditions, suggesting that stiffness alone cannot guarantee stable locomotion. These findings indicate that appropriate joint stiffness (i.e., adequate body flexibility) and phase-difference control are critical for stable locomotion in super-long multilegged robots. Future work will investigate the mechanisms by which phase difference affects following performance and establish design and control guidelines based on the interplay between body flexibility and leg motion.

Keywords: Mechatronics Systems, Mechanical Systems Control
Abstract: Flexible steel plates are widely utilized in modern industries, particularly in the automotive sector. However, contact-based transport systems using rollers, which are currently mainstream on factory production lines, present challenges such as the deterioration of surface quality and the peeling of plating. To resolve these issues, a non-contact gripping and transport technology utilizing magnetic levitation via electromagnetic force has been proposed. It has been demonstrated that when using E-shaped core electromagnets, the orientation relationship between the flexible steel plate and the electromagnet significantly impacts levitation stability. Furthermore, the authors have previously clarified the steady vibration characteristics of the flexible steel plate in a levitated state. However, the levitation stability when simulating actual transport has not yet been clarified. Therefore, in this paper, we experimentally investigated the levitation stability of the steel plate to further enhance the practicality of this technology.

Keywords: Mechatronics Systems, Mechanical Systems Control
Abstract: In the transport process for thin steel plates used in the manufacture of automobiles and electrical appliances, the use of rollers causes deterioration in surface quality. Furthermore, roller-based transport is limited in the directions in which it can move. In contrast, systems that use magnetic levitation for non-contact transport have been proposed, but the freedom of movement in the transport direction remains limited. Therefore, the authors propose a magnetic levitation transport system that incorporates an electromagnet into the end effector of a robotic arm. This system aims to simultaneously maintain surface quality and provide high freedom of movement by driving the robotic arm while holding the object in a non-contact manner. In this paper, we experimentally investigated the effect of transport velocity on levitation stability when an object in a magnetically levitated state is transported vertically by a robotic arm.

Keywords: Mechatronics Systems, Mechanical Systems Control
Abstract: In the factory transport process of flexible steel plates used for automobile bodies and home appliances, surface quality deterioration due to contact with rollers is a problem. To solve this problem, magnetic levitation transport using electromagnets has been proposed. However, due to their flexibility, flexible steel plates deflect in a complex manner during levitation, and the resulting vibrations cause the plates to fall. The authors proposed a magnetic levitation system that achieves the simplification of deflection during levitation by limiting the arrangement of electromagnets to the sides of the flexible steel plate. In this system, a control system considering the motion in the direction susceptible to vibration was constructed, and it has been confirmed that levitation can be maintained. In this paper, the influence of differences in levitation conditions and plate thickness on the vibration characteristics during levitation was investigated using the constructed control system.

Keywords: Mechatronics Systems, Mechanical Systems Control
Abstract: In the manufacturing process of flexible steel plates, transport using rollers is common. However, a major issue is the deterioration of surface quality caused by contact with the rollers during transport. Furthermore, from the perspective of factory automation, there is a need to increase the degree of freedom the transport lines. Consequently, non-contact transport using magnetic levitation has been proposed. However, conventional levitation methods face the challenge that the flexible steel plate sags during levitation, causing complex elastic vibrations that reduce levitation stability. Therefore, the authors proposed a new levitation system that suppresses sagging by suspending the steel plate from one edge, thereby utilizing its own weight. However, achieving more stable levitation is necessary for practical application. Therefore, in this report, we experimentally investigated the effects on levitation characteristics by applying disturbances simulating inertia to the proposed system.

Keywords: Mechanical Systems Control, Transportation Systems
Abstract: The seat design of vehicles contributes to enhancing comfort of drivers and passengers against external shaking motion. The study conducted random vertical vibration tests for a seat with backrest and collected acceleration data using acceleration sensors attached at the seat and backrest. The experimental modal analysis was employed for vibration analysis of the coupled seat and backrest. The analysis clarified the influence of the backrest on the random vertical vibration. The study provides the design clue of the seat and backrest to suppress external vibration.

Keywords: Mechatronics Systems, Mechanical Systems Control
Abstract: Roller conveyors are typically used to transport flexible steel plates in industrial manufacturing. However, this method causes surface quality degradation due to contact and is limited to unidirectional transportation. Non-contact transportation systems using magnetic levitation technology have been proposed, but their transportation flexibility remains low. Therefore, the authors propose a system to transport thin steel plates without contact and with high flexibility by attaching an electromagnet to a robot arm's end-effector. In constructing this system, it is necessary to clarify how disturbances, such as inertial forces generated during multi-directional transportation and acceleration/deceleration, affect the magnetic levitation system. In this report, to evaluate only the impact of transportation-induced disturbances, we transported a rigid steel ball to eliminate elastic deformation and investigated the levitation characteristics during the transportation of the steel ball by the robot arm.

Keywords: SCADA and Network Systems, Manufacturing Systems, Factory Automation
Abstract: As cyberattacks against critical infrastructure become increasingly sophisticated, strengthening Governance, Risk, and Compliance (GRC) for operational technology (OT) environments is critical. Although IEC 62443 provides a widely adopted security framework, dynamic risk identification from operational security data and its linkage to Security Requirements (SRs) remain largely unestablished. Furthermore, individual intrusion-detection systems (IDSs) provide only partial observations because of sensor-specific blind spots. To address this challenge, we propose the Cross-Sensor Chained Diamond Model (CS-CDM), which normalizes alerts from OT-network IDSs, host logs, and application logs into a six-tuple Diamond Model annotated with IEC 62443 SRs and MITRE ATT&CK® for ICS techniques. CS-CDM preserves per-sensor detectability information and correlates observations across heterogeneous data sources to reconstruct attack progression and SR traceability records. Evaluation on an ICS testbed under multiple attack scenarios, including independent third-party red-team assessments, shows that correlating heterogeneous sensors enables CS-CDM to reconstruct attack progression more completely than any single sensor alone and to identify SRs associated with observed attack activities across IEC 62443 foundational requirement groups. These results demonstrate that dynamic, requirement-level risk identification can bridge operational security monitoring and IEC 62443-based GRC.

Keywords: Robotic and Automation Systems, Safety, Environment and Eco-Systems
Abstract: This study proposes an insect-based mobile robot for the efficient inspection of aging pipelines. In underground pipeline systems, the exact routes and depths of pipelines are not always accurately documented, creating a risk of accidental damage during maintenance and construction work. In addition, identifying visually inaccessible issues such as small leaks requires considerable labor and cost. To address these challenges, we evaluated the feasibility of using an insect-based mobile robot using a Madagascar hissing cockroach as a low-power and highly maneuverable inspection platform. A radio electronic board was mounted on the insect for wireless control of its walking, including forward acceleration and directional turning which were elicited through electrical stimulation of the antennae and cerci of the insect. We demonstrated successful turning behavior inside PVC pipes and vertical movement within electrical conduits. Future work will include load tests simulating camera installation to evaluate the feasibility of real-time visual monitoring within pipeline environments. Furthermore, the integration of cameras and local positioning technology could enable the acquisition of positional and visual information in complex pipeline environments. These results suggest that insect-based mobile robots have the potential to improve the efficiency and reduce the cost of infrastructure inspection and maintenance.

Keywords: Robotic and Automation Systems
Abstract: We consider the shepherding problem in which the sheep robots alternately perform turning and straight‑line motions based on local sensing and inter‑agent interactions, with a view to real‑world implementation in robots. A major problem of this model is that, in large swarms, robots located near the goal tend to form a wall‑like structure that obstructs further guidance, causing the guidance time to escalate substantially and resulting in a marked decline in steering efficiency. To address this issue, we modify the sheep model by introducing an upper bound on the turning angle and altering the interaction pattern among sheep robots. Numerical simulations demonstrate that this modification reduces the increase in guidance time observed in large swarms.

Keywords: Robotic and Automation Systems, Transportation Systems, Mechanical Systems Control
Abstract: Autonomous mobile robots for indoor use are becoming increasingly widespread, ranging from delivery services in factories and warehouses (as AGV) to serving robots in stores and floor cleaning robots in public spaces. Self-localization and navigation are performed using LiDAR-SLAM. The purpose of this research is extension of the application of such LiDAR-SLAM-based autonomous mobile robots to outdoor environments, identifies challenges, and proposes solutions. By applying a light-resistant 3D-LiDAR, 3D point cloud map data is able to be acquired even under sunlight. The mobile robot can also travel on pathways that include slopes to overcome steps. By adding 9-axis IMU measurements, odometry errors are corrected even on slightly uneven paved surfaces, tiled or block-paved areas, and pathways with steps. This robot system is confirmed to be able to construct maps of continuous routes both indoors and outdoors. As a demonstration of continuous autonomous navigation, it is shown that it is possible to navigate routes that pass through building exit. Navigating through building exit requires adjustments such as detecting transparent glass and ramps to avoid steps. A solution is presented as adding the correction information to the map data.

Keywords: Adaptive and Optimal Control
Abstract: Autonomous driving technologies have attracted significant attention for improving traffic efficiency and safety. At intersections with bidirectional traffic flow, right-turning vehicles are often forced to wait for oncoming straight-going vehicles, resulting in traffic congestion and increased vehicle delays. This study proposes an optimal traffic flow control method that considers right-turning vehicles in a bidirectional traffic environment. First, vehicles are sorted according to their positions relative to the intersection. Then, the optimal arrival time of each vehicle at the merging zone is determined using mixed integer programming. To ensure safety, collision avoidance constraints are introduced for both consecutive vehicles and conflicts between right-turning and oncoming straight-going vehicles. The order in which right-turning vehicles and oncoming vehicles proceeding straight ahead pass each other is automatically determined using the Big-M method and 0-1 variables to avoid collisions. Based on the calculated arrival times, an optimal control algorithm is applied to determine vehicle accelerations by minimizing the sum of squared accelerations. Simulation results demonstrate that the proposed method can improve traffic flow smoothness, reduce vehicle delays, decrease CO₂ emissions, and maintain safe vehicle operations at intersections.

Keywords: Adaptive and Optimal Control, Intelligent Control, Modeling, System Identification and Estimation
Abstract: This paper proposes a controller updating method by online Virtual Reference Feedback Tuning (VRFT) in the encrypted domain. The method updates controller parameters sequentially from input-output data while keeping the data private. For this purpose, online VRFT is combined with the CKKS homomorphic encryption scheme. CKKS allows approximate calculations for real numbers on encrypted data. In the proposed method, the regression vector, the estimation error, the controller update, and the matrix update are computed in the encrypted domain. Therefore, the controller parameters can be updated without using the plant model and without exposing plaintext data during the computation. This paper also points out that approximation errors caused by encryption and decryption may affect convergence. The proposed method is a basic step toward privacy-preserving adaptive data-driven control.

Keywords: Adaptive and Optimal Control, Intelligent Control
Abstract: In this paper, we propose a robust adaptive output feedback control method for continuous-time linear systems using neural networks (NNs). In almost strictly positive real (ASPR)-based adaptive control, the augmented system with a parallel feedforward compensator (PFC) exhibits a problem: the actual output does not track the desired reference signal. To address this problem, we propose a feedforward (FF) input design method based on NN approximation of the plant's inverse system. Furthermore, we propose a disturbance compensator that reuses the NN employed in the FF input design to estimate the actual input, thereby ensuring robustness against disturbances. The stability of the proposed method is theoretically analyzed using Lyapunov analysis, and the boundedness of all signals within the control system is shown under the assumption that a well-trained NN exists.

Keywords: Adaptive and Optimal Control, Nonlinear Control
Abstract: This paper proposes an event-triggered high-gain adaptive output-feedback controller that employs a constant variation of the control input. By introducing a constant input variation that surpasses the event-triggering threshold, the controller is expected to achieve performance improvements analogous to those obtained with high-gain control. Numerical simulations on a magnetic levitation system confirm that the proposed method enhances output tracking performance while mitigating abrupt changes in the control input.

Keywords: Adaptive and Optimal Control, Modeling, System Identification and Estimation
Abstract: This paper proposes a null-space exploration method for discrete-time systems with two redundant inputs using an adaptive allocator. In the proposed method, the control input is parameterized by a unit-norm vector, and its direction is updated sequentially based on measurable signals. In particular, by utilizing the difference signal between successive states, the component corresponding to the direction of the unknown input coefficient vector is directly extracted. It is shown that the adaptive allocator asymptotically converges to the null space of the unknown input coefficient vector. Furthermore, the convergence property is established using a Lyapunov-based analysis, and numerical simulations demonstrate the effectiveness of the proposed method.

Keywords: Adaptive and Optimal Control, Modeling, System Identification and Estimation, Nonlinear Control
Abstract: Adaptive control consists of two types of loops. One is the main loop, which provides feedback on the plant output, and the other is the adaptive loop, which adjusts the feedback gain. Most of the existing event-triggered adaptive control systems introduce event-triggering mechanisms (ETMs) into the main loop. However, the control performance in the event interval becomes conservative. To cope with this issue, this paper presents a novel event-triggered adaptive high-gain output feedback control system that networks the adaptive loop rather than the main loop and incorporates two ETMs.

Keywords: Adaptive and Optimal Control
Abstract: The study proposes a predictive optimal iterative Learning Control (ILC) for single-input, single-output, discrete-time, time-invariant linear systems. The predictive optimal ILC updates the control input for the next iteration by optimizing a cost related to performance over future iterations. In the predictive ILC, the learning gain matrix is derived using the solution of a Riccati equation based on information from the system matrices of the controlled plant. The main feature of the present study is to introduce a modified gradient Q-learning method for deriving the learning gain without help from the information of system matrices. Since the modified Q-learning method enables us to realize a moderate learning process compared to existing Q-learning, it is expected that the predictive optimal ILC using the modified Q-learning will show rather robust and smoother performance in the transient learning process.

Keywords: System Integration, Network System Integration, Intelligent Control
Abstract: This study proposes an encrypted visual feedback control algorithm for regulating a one-dimensional stage using Ring Learning With Errors (RLWE) encryption. The proposed algorithm performs both feature extraction and controller computations directly on encrypted images, ensuring that sensitive visual data remain protected throughout the entire control process. Furthermore, an image captured by the camera is encrypted into a single ciphertext leveraging the message packing technique of RLWE encryption, thereby reducing computational cost. The effectiveness of the proposed framework is demonstrated through numerical simulations.

Keywords: Robust Control, Networked Sensor System
Abstract: This study analyzes dynamic-key Learning-with-Errors (LWE)-based encrypted state-feedback control systems with time-varying encoders and decoders. Using a Lyapunov-based approach, we derive conditions on the time-varying encoder and decoder parameters that ensure both asymptotic stability and numerical safety by preventing overflow. The validity of the derived conditions is confirmed through numerical examples.

Keywords: Nonlinear Control, Modeling, System Identification and Estimation
Abstract: This paper presents HEK-Control, an encrypted control framework for nonlinear systems that integrates Koopman-based lifted linear modeling with CKKS homomorphic encryption. The proposed framework transforms nonlinear control into a lifted-space linear feedback problem, so that the outsourced online computation can be expressed as encrypted linear algebraic operations. In this architecture, nonlinear lifting is performed locally, while the cloud computes the feedback input without accessing plaintext state information. Numerical simulations on representative nonlinear systems demonstrate that the encrypted controller can reproduce the behavior of the corresponding plaintext controller with limited performance degradation. The results indicate the feasibility of encrypted outsourced nonlinear control and highlight the importance of lifted-model accuracy, encryption noise, and computational overhead in practical implementations.

Keywords: Modeling, System Identification and Estimation, Human-Machine Systems, Robotic and Automation Systems
Abstract: Research has been conducted on encrypting control computations using homomorphic encryption. Due to the properties of homomorphic encryption, conventional encrypted control has difficulty being applied to control systems that include nonlinearities such as saturation. In this study, we propose a method to address such control systems by representing the saturation function using polynomial approximation and transforming it into a form that can be implemented under homomorphic encryption. In designing the approximation polynomial, constraint conditions are introduced to maintain control stability, and a polynomial satisfying these conditions is constructed. Furthermore, to ensure numerical stability in encrypted control, the appropriate range of cryptographic parameters is theoretically derived. Finally, the proposed method is implemented in a bilateral control system using a pneumatic cylinder, and experimental results confirm that the approximated system exhibits no degradation in performance in terms of position/force tracking and control delay.

Keywords: Modeling, System Identification and Estimation, Mechanical Systems Control, Networked Sensor System
Abstract: This study aims to analyze the impact of quantization errors on the passivity of encrypted bilateral control systems. In recent years, advances in communication technology have accelerated the research and development of teleoperation systems. As a countermeasure against cyberattacks on such networked control systems, encrypted control has been proposed to enable the concealment of internal information and the detection of tampering. However, encrypted control inevitably introduces quantization errors when converting real-valued control signals into plaintexts. Although several previous studies have investigated stability guarantees for control systems with errors, few studies have addressed the impact of quantization errors on the stability of encrypted bilateral control systems.This study aims to analyze the nature of the impact theoretically, and this paper presents the approach as an initial conceptual step toward the goal.

Keywords: Modeling, System Identification and Estimation, Networked Sensor System, Network System Integration
Abstract: Encrypted control employs homomorphic encryption (HE) to protect both the computation and communication stages, making it a promising approach for secure networked control systems. Most existing results pre-design a controller in the plaintext domain and then implement it over encrypted data. However, this can be problematic because HE induces non-negligible communication and computation delays that typically increase with the security level, potentially degrading control performance and even destabilizing the closed-loop system. To address this issue, we propose a co-design framework for cryptographic parameters and delay-aware feedback gain. We first characterize an upper bound of the encryption-induced delay as a function of the cryptographic parameters. Then, for a given set of cryptographic parameters and feedback gain, we derive a sufficient condition under which the closed-loop system remains stable for any admissible delays, expressed as a finite set of linear matrix inequalities. This leads to a tractable outer-inner design procedure: the outer loop selects cryptographic parameters satisfying the desired security level, while the inner loop seeks a stabilizing delay-aware feedback gain.

Keywords: Intelligent Systems, Computational Intelligence, Medical and Welfare Systems
Abstract: Medical cast structure optimization is important for improving patient comfort by decreasing weight while maintaining or enhancing stiffness. Conventional casts are typically heavy and low ventilation, motivating the need for lightweight structural design without trade-off stiffness. This study examines the use of the Double Deep Q-Network (DDQN) for optimal placement lattice structures in the design of a custom medical arm cast. The arm cast model was obtained through 3D scanning and was divided into 80 parts. Each part is defined to either solid or lattice material. The aim is to minimize the strain energy under mass constrain, since the reduction of strain energy corresponds to the increase in stiffness of the system. Each design iteration was evaluated using finite element analysis (FEA) with LS-DYNA to determine the effect of the design changes on the structural stiffness. Each element of the DDQN agent iteratively changed the material distribution based on the reward associated with the reduction of strain energy and the satisfaction of the mass constrain. The optimized design places solid material along load paths or high-stress areas while placing lattice material in low-stress areas. The results demonstrate the potential of deep reinforcement learning for structural optimization in medical cast applications.

Keywords: Intelligent Systems, System Integration, Mechatronics Systems
Abstract: This paper presents a fault detection framework for a 6-DOF robotic manipulator based on supervised machine learning models. To ensure reliable operation, we propose a data-driven approach that classifies sensor anomalies such as noise, spikes and bias that are induced within the joint control feedback loop. The methodology is structured in two stages: first, a comprehensive benchmark is conducted across multiple classification models, including Decision Tress, k-Neares Neighbors, Suport Vector Machines, and Artificial Neural Networks. Subsequently, an optimization of hyperparameters is performed for the most accurate architecture to maximize classification accuracy. Experimental results demonstrate the effectiveness of this refined approach in identifying sensor faults in real-time, providing a scalable and high-performance solution for the advanced diagnostics of robotic control systems.

Keywords: Intelligent Systems
Abstract: Reinforcement learning is a framework in which an agent learns a control policy through trial-and-error interaction with an environment. However, it generally requires a large number of interactions to achieve good performance, and this problem becomes more serious in sparse-reward tasks. To improve learning efficiency, previous studies have used teacher policies that guide the agent's exploration. More recently, large language models (LLMs) have been used to construct teacher policies by selecting high-level skills from natural-language descriptions of states. Although this approach can reduce the need for manually designed teacher policies, the teacher policy generated by an LLM may include errors. Such errors can misguide the agent and reduce learning efficiency. In this paper, we propose a method for correcting an LLM-based teacher policy using feedback from the environment. The LLM output is used as an initial teacher policy, and its option-selection probabilities are updated according to whether the selected option successfully leads to the next subgoal in a given sequence of subgoals. The corrected teacher policy is then used to support reinforcement learning. Experiments on the SimpleDoorKey task in MiniGrid show that the proposed method achieves better sample efficiency than LLM4Teach when the teacher policy contains errors, suggesting that it is robust to an increasing number of teacher-policy errors.

Keywords: Intelligent Systems
Abstract: LLM-based GUI agents generate GUI actions from natural-language instructions, enabling the operation of general-purpose computer systems. One of the critical challenges of these GUI agents is the grounding accuracy: the conversion of a textual action plan into on-screen locations. To address this challenge, we incorporate visual changes, such as hover highlights and cursor-shape changes, into the GUI agent. This idea is inspired by how users manually adjust the pointer: first detect GUI visual change, then finely adjust the pointer position. We reflect the above concept in our post-hoc method, which refines the pointer position based on plausible visual changes, supported by a previously constructed knowledge base. In our experiments to evaluate the success rates of grounding tasks, our post-hoc refinement improved accuracy on tasks requiring precise alignment.

Keywords: Intelligent Systems, Computational Intelligence, Biological and Physiological Engineering
Abstract: Explainable artificial intelligence (XAI) is important because modern machine learning models often achieve high predictive performance while remaining difficult to interpret. To bridge the gap between global and local explanations, this study proposes an intermediate approach that extracts human-readable if-then rules from relevant search regions. We apply supervised learning-based classifier systems (UCS) to a protein solubility prediction model in a large search space. We compare UCS trained on mutation sets derived from genetic algorithm (GA) search histories with UCS trained on randomly generated mutation sets, and validate the extracted rules on a target protein. The results show that UCS can derive interpretable and reusable mutation rules that capture solubility-improving patterns learned by the prediction model.

Keywords: Intelligent Systems, Intelligent Control
Abstract: Velocity tracking is one of the most fundamental tasks in reinforcement learning for quadruped robots. However, a policy that merely tracks the commanded velocity does not necessarily account sufficiently for the robot’s bodily state, and may therefore result in unstable locomotion that is prone to falling. In this study, we formulate quadruped locomotion using a reduced-order model focused on the center of mass, and perform learning that incorporates the controllability of center-of-mass motion with ground reaction forces as inputs. The results show that the fall rate is reduced. These results suggest that locomotion stability can be improved not only by learning body motion from joint torques, but also by learning how ground reaction forces arising from body motion influence center-of-mass dynamics.

Keywords: Computational Intelligence, Innovative Systems Approach for Realizing Smarter World, Power Systems Control
Abstract: This paper proposes a particle swarm optimization (PSO) method with improved computational efficiency for the simultaneous optimization of the placement and capacity of battery energy storage systems (BESSs) in power distribution networks. The charging and discharging schedules of a BESS are treated as optimization variables, and two strategies are incorporated into the PSO framework. Namely, the dimensionality of particles, i.e., candidate solutions, is gradually increased and the number of particles is reduced as the search progresses in order to improve the modeling accuracy of the schedule while reducing computational cost. Simulation results using the IEEE 33-bus distribution system confirm that, compared with conventional PSO, the proposed method reduces the total system cost and shortens computation time while maintaining solution quality.

Keywords: Computational Intelligence, Signal and/or Image Processing, Sensors and Transducers
Abstract: Tool wear significantly affects machining quality, process stability, and maintenance planning in modern manufacturing. However, conventional data-driven monitoring methods usually require large labeled datasets and often show poor generalization under different machining conditions. To address this issue, this study proposes a tool wear monitoring framework integrating Artificial Neural Networks (ANN) with transfer learning to improve diagnostic performance under limited data conditions. A multi-sensor platform was established to collect vibration, acoustic, and current signals related to tool wear. Time-domain and frequency-domain features were extracted, and correlation analysis was used to identify representative indicators for diagnosis. The ANN hyperparameters were optimized through experiments, and transfer learning was applied based on the selected model configuration. Results show that when only 13% of the target dataset is available, the transfer learning model achieves 83% accuracy, compared with 66% for the model trained from scratch. With 40% of the target data, the accuracy further increases to about 94%. These results demonstrate that transfer learning can alleviate data scarcity and improve diagnostic performance under different machining conditions.

Keywords: Computational Intelligence, Robotic and Automation Systems, Mechatronics Systems
Abstract: As the labor population declines, the manufacturing sector increasingly requires Human-Robot Collaboration(HRC) to overcome full automation constraints in Small and Medium-sized Enterprises (SMEs). However, implementing Reinforcement Learning (RL) in shared workspaces remains challenging due to the unpredictability of human movements and unexpected obstacles. To ensure safety, this study proposes a method for robots to acquire physical awareness by referencing the body schema, a cognitive psychology concept representing body configuration. By integrating the body schemas of both the human and the robot into a unified 3D representation, the system enables the robot to perceive realtime spatial relationships and relative positions. This approach facilitates safe workspace sharing by allowing the robot to adapt its motions to human postural changes, thereby mitigating the risk of industrial accidents during autonomous task learning.

Keywords: Human-Machine Systems, Intelligent Systems, Innovative Systems Approach for Realizing Smarter World
Abstract: In recent years, various sports have been introducing systems that utilize IT technologies to enhance the quality of training and to perform evaluations. In table tennis, there is a problem that technical instruction is difficult, and skills are often acquired intuitively. Hence, in this study, we propose a system for table tennis practice that visualizes both the player's posture and the racket movement in three dimensions. This system utilizes technology for estimating the 3D human posture and motion sensor for measuring the racket 3D movement. To estimate the player's posture, we use AlphaPose and MotionBERT to estimate the player's 3D posture from a single camera movie. To estimate racket posture, a motion sensor, Movella DOT is attached to the tip of the racket grip to acquire posture angle data at each time point. By integrating these two methods, We realize visualization of the player pose and racket movement in three-dimensional space in table tennis.

Keywords: Human-Machine Systems, Intelligent Systems, Medical and Welfare Systems
Abstract: This paper considers patient fatigue detection in tele-rehabilitation system with electrical stimulation using a machine learning. The maximum values of control input and knee joint angle are used in knee joint flexion and extension exercises with electrical stimulation as feature data. First, the feature data is classified into fatigued data and non-fatigued data using the K-means method. In order to normalize the fatigued data and non-fatigued data, the ratio between the average control input and average knee joint angle is obtained. The ratio is multiplied by the knee joint angle. Next, determining fatigue based on distance from the center point of cluster is considered. Finally, we verify the experimental results when test subjects are changed.

Keywords: Human-Machine Systems, Robotic and Automation Systems, Mechanical Systems Control
Abstract: This paper presents an active-assistive control framework for an upper-limb rehabilitation exoskeleton. To reduce the gravity-related component in the measured interaction torque, relaxed-state force/torque (F/T) data are collected before training and used to identify a subject-specific gravity compensation term. Based on the resulting active torque estimate, a variable admittance controller and a desired assistive velocity generator are combined to provide user-responsive assistance within predefined motion ranges. In this framework, the exoskeleton follows the user more directly when voluntary effort is sufficient and provides supplemental assistance when the user’s effort decreases during the task. Experiments on shoulder and elbow flexion/extension with healthy subjects show that the proposed framework, together with the subject-specific gravity compensation term, improves intention-consistent assistance compared with the case without gravity compensation. Surface electromyography (sEMG) observations further suggest that the proposed framework can reduce excessive effort during flexion while encouraging more active contribution during extension. These results provide an initial validation of the proposed framework for single-joint rehabilitation tasks.

Keywords: Human-Machine Systems, Signal and/or Image Processing, System Engineering
Abstract: This study explores the cognitive human behaviors involved in voice resonance through an experimental phonetics analysis of the vowel sound /ai/. By examining the acoustic signals of high school participants, the research focuses on the behavior of formants F1 and F2 to determine if vowel sounds remain consistent across different utterances. Results demonstrate that /ai/ does not maintain a unified resonant identity; specifically, the second formant (F2) shows significant dramatic shifts depending on the phonetic context and following consonants, such as /m/, /l/, /h/, and /k/. Analysis through Praat spectral illustrations reveals that while some utterances remain stable, others—such as /ai/ followed by /h/ in the phrase "I have"—exhibit dramatic frequency drops. These findings suggest that vowel production is a nonstationary, non-unified signal heavily dependent on individual human behavior and vocal tract articulation, challenging the assumption of vowel consistency in experimental phonetics.

Keywords: Human-Machine Systems, Human Interfaces, Intelligent Systems
Abstract: 近年、運転支援用の音声ナビゲーションシステムが広く採用されています。しかし、「ここは事故が多い地域です。「運転には注意してください」と言われ、ドライバーに苛立ちや過度な介入の印象を与える可能性があります。本研究では、2台のロボット間の会話型相互作用を用いた新しいリスクアラート法「D-Bones」を提案します。さらに、実際の車両実験を通じて、ロボット間の会話的なやり取りが単一ロボットの発

Keywords: System Engineering, Mechatronics Systems
Abstract: In this paper, we proposed a thermoelectric generation structure with an enclosing arrangement of Peltier elements around a vortex tube to recover energy from exhausted compressed air. In the proposed structure, Peltier elements were placed around the vortex tube to form separate hot air and cold air spaces, thereby shortening the heat transfer path between the separated airflow and elements. The structure also enabled the utilization of thermal loss from the hot-side housing of the vortex tube. To further enhance heat transfer, the heat sink arrangement was optimized by increasing the heat sink size toward the downstream side, where heat carried by airflow away from the element surface became crucial. Experiments were conducted to investigate the effects of the heat sink configuration and number of enclosing Peltier elements. The results demonstrated that the proposed structure achieved a higher energy recovery rate compared with the conventional structure. The maximum energy recovery rate was 0.277%, corresponding to approximately 3.6 times that obtained in a previous study. These findings verify the effectiveness of the proposed structure and heat sink arrangement for improving thermoelectric energy recovery from exhausted compressed air.

Keywords: Autonomous Decentralized Systems, Social Systems
Abstract: In this paper, we propose a method for identifying detectable nodes within dynamical network systems, utilizing degree centrality as the foundational metric. Degree centrality quantifies a node's prominence in a network by assessing the number of connections. Scale-free networks, which are prevalent in social networks and power grids, exhibit skewed degree distributions that may be leveraged for node identification. In this paper, we extend the application of degree centrality to the analysis of dynamic systems, facilitating the exploration of detectable nodes. The number of detectable nodes is minimized using a greedy algorithm that iteratively calculates the degree and selects the node with the highest degree.

Keywords: Autonomous Decentralized Systems, Mechatronics Systems
Abstract: Abstract: Swarms designed with swarm robotics in mind are inexpensive yet possess system robustness, task flexibility, and scalability. However, there is a reality gap between the swarm behavior obtained through simulation and its actual implementation on physical robots. In this study, we improved the robot hardware based on the challenges identified in previous research involving physical robots. To reduce the reality gap, we conducted reinforcement learning experiments in which random noise was added to the inputs and outputs during the emergence of swarm behavior. As a result, while the system gained some resistance to noise in simulations, no reduction in the reality gap was observed in real-world experiments; therefore, we discuss the causes and potential solutions.

Keywords: Autonomous Decentralized Systems, Factory Automation, Intelligent Systems
Abstract: In this research, we propose a task allocation method for multi-agent systems using a distance-based grouping algorithm. The proposed algorithm generates task groups consisting only of nearby packages. Our grouping method reduces the computational cost, which enables the system to be applied to large-scale problems. Also, the operating time of agents is one of the major concerns in multi-agent systems. We consider the energy consumption of each agent in the cost function of the auction phase to prevent battery depletion. Finally, we conduct simulations to show the effectiveness of the proposed allocation method in terms of computation time and energy consumption.

Keywords: Social Systems, Modeling, System Identification and Estimation
Abstract: In crowd accidents at event venues, children are frequently among the victims, facing a higher risk due to their underdeveloped physical and cognitive characteristics. Therefore, safe space design requires models capable of predicting children's behaviour in crowds. This study constructs a mathematical model of pedestrian flow which reflects the characteristics of children aged 3 to 5 years. Furthermore, the differences in behavior and mobility compared to adults are investigated based on the proposed model with model predictive control.

Keywords: Molecular Robotics, Modeling, System Identification and Estimation
Abstract: Molecular circuits based on DNA strand displacement exhibit high programmability. However, they face the challenge that the same system is difficult to reuse repeatedly. In this study, we propose a design principle for an output-resettable DSD system that can repeatedly respond to variations in input concentrations. Through mathematical modeling and simulations, we confirm that the system exhibits transient responses to input changes and returns to a steady state independent of the input. Furthermore, we experimentally implement part of the circuit and verify its correct operation. This work provides a new design guideline for resettable DNA circuits and is expected to contribute to the realization of molecular control systems that repeatedly operate in response to environmental changes.

Keywords: Intelligent Control, Guidance and Flight Control, Mechanical Systems Control
Abstract: This paper presents a guaranteed-cost design method for a fuzzy-model-based disturbance observer and applies it to elevator trim estimation for a flying-wing unmanned aerial vehicle (UAV). The elevator trim varies significantly owing to payload-induced shifts in the center of gravity and errors in airspeed measurement. For elevator-trim estimation, a nonlinear state-space model that incorporates the unknown elevator trim as a disturbance is transformed into an equivalent fuzzy model, based on which a fuzzy disturbance observer is designed. New linear matrix inequality (LMI) conditions are derived for minimizing a cost function that captures the trade-off between the convergence rate of the estimation error and the magnitude of the observer gains. The proposed method provides a systematic procedure for synthesizing observer gains while explicitly accounting for this trade-off. Finally, the effectiveness of the proposed method is examined using flight-test data.

Keywords: Intelligent Control, Intelligent Systems, Mechanical Systems Control
Abstract: This study evaluates the use of Gemini Robotics-ER 1.6 for zero-shot object localization in a physical motor control system. The estimated object position was converted into a camera rotation command without task-specific training. Experiments with two target objects confirmed that the system reduced the horizontal image error and moved the objects closer to the image center. The results demonstrate the feasibility of object-centering control while revealing limitations for continuous visual feedback.

Keywords: Intelligent Control
Abstract: In this paper, a learning-based self-triggered control method is proposed for discrete-time linear systems with disturbances. The proposed controller determines both the inter-execution interval and the held control input. The uniform ultimate boundedness (UUB) of the closed-loop system is guaranteed by using a given controller only when it is judged that the learnt controller is inappropriate. Numerical results show that the proposed method reduces control updates while preserving UUB.

Keywords: Intelligent Control, Power Systems Control, Human-Machine Systems
Abstract: Geothermal power generation is highly anticipated as a key contributor to achieving carbon neutrality. However, automating control systems for geothermal fluids remains a challenge due to the difficulty of accurately identifying physical models of reservoir and wellbore systems and tracking their temporal changes. Consequently, valve operations have traditionally relied on the empirical knowledge of on-site operators. Previous research proposed a new control framework based on a deep learning model that estimates the Instability Level (IL) from time-series wellhead pressure data. Nevertheless, providing specific manipulated variables for full automation remained an unresolved issue. In this study, we developed an integrated system that combines the IL output model with a waveform prediction model. This system features both an instability detection function and a guidance function that suggests the polarity of operational adjustments (valve opening) required to reduce the IL. Through evaluation using real-world data, we report on the accuracy of the instability detection and the overall utility of the proposed guidance system.

Keywords: Intelligent Control, Intelligent Systems, Nonlinear Control
Abstract: We present a new observer-based output feedback controller for a nonlinear system described in descriptor form that can be transformed into a Takagi-Sugeno fuzzy system. When the premise variables are unmeasurable, the analysis of the resulting estimation error system becomes significantly more challenging. In this paper, we analyze the stability of an augmented system that takes into account both a controller and an observer by exploiting the Lipschitz condition for the error dynamics. Based on the proposed augmented system, we design a controller and an observer independently, and the corresponding design conditions are expressed in terms of linear matrix inequalities. To demonstrate the validity of our proposed design conditions through comparison with existing results, a numerical example is presented.

Keywords: Intelligent Control, Process Control, Computer Aided Design
Abstract: Data-driven predictive control involves a trade-off between noise robustness and computational efficiency. While rich datasets can mitigate the effects of noise, they also increase the online computational burden. Building on our previous work using LASSO for frequency-domain basis selection, this paper investigates the application of dictionary learning to further enhance data compressibility for an electromagnetic molding machine. However, numerical evaluations reveal that dictionary learning yields inferior control performance compared to selecting the original frequency-domain bases. This finding suggests that preserving the structural integrity of physical basis functions is more effective for data reduction than purely adaptive basis learning. These results indicate the importance of the system's spectrum for efficient data-driven control design.

Keywords: Intelligent Control, Nonlinear Control, Adaptive and Optimal Control
Abstract: This study considers decentralized priority assignment for two robots, where the robots autonomously determine which one passes first to avoid collisions. The motivation for considering this task is to study mutual predictive control, i.e., control with temporary cooperation based on predicting each other's behavior, which can be applied to on-demand cooperation among high-end robots. To achieve the decentralized priority assignment, we focus on distributed constrained control using control barrier functions. We show a theoretical result that clarifies a difficulty in directly applying an existing method. By modifying the existing method based on the theoretical result, a decentralized control method for the priority assignment is presented. The performance of our method is demonstrated through simulation.

Keywords: Adaptive and Optimal Control, Modeling, System Identification and Estimation, Multivariable Control
Abstract: In this study, we extend indirect policy gradient adaptive control (PGAC) to large-scale discrete-time systems. Indirect PGAC is an online algorithm that consists of two stages: estimating the system model and improving the feedback controller using a policy gradient method. However, when the state dimension is large, both model estimation and gradient computation become computationally expensive and can be numerically ill-conditioned. To overcome this difficulty, we introduce a reachability-based projection onto a low-dimensional feature space. The projection matrix is constructed via singular value decomposition of the initial data matrix and is then used as a common preconditioner throughout the subsequent PGAC procedure. Numerical experiments on 100-dimensional systems compare the full-order and reduced-order PGAC algorithms in terms of the true linear quadratic regulator (LQR) cost and online computation time. The results show that the proposed method achieves almost the same closed-loop performance as the full-order method while substantially reducing online computational cost.

Keywords: Nonlinear Control, Robust Control, Guidance and Flight Control
Abstract: This study investigates sliding mode control for low-thrust orbital deployment toward small-satellite constellation missions. Recent advances in electric propulsion have made continuous low-thrust maneuvers feasible for small satellites, enabling precise observation missions such as interferometric synthetic aperture radar. However, accurate constellation deployment requires robust orbit tracking under perturbations, model uncertainties, and thrust limitations. To address this problem, an orbital-element-based control framework is formulated using Gauss’ planetary equations. The semi-major axis, eccentricity, and inclination are selected as controlled variables, allowing in-plane and out-of-plane orbital motion to be treated separately. A sliding variable is defined from normalized orbital-element errors, and a sliding mode controller is designed to drive the satellite toward the desired orbital state while maintaining robustness against bounded matched disturbances. Numerical simulations considering perturbation forces and input constraints demonstrate that the proposed framework achieves the target orbital elements with a feasible control effort.

Keywords: Adaptive and Optimal Control, Nonlinear Control
Abstract: This study addresses the application of the Newton method for the optimization problem in the iterative feedback tuning (IFT) and explores the positive definiteness of its Hessian. One of the issues with the Newton method is that the descent condition on the objective function cannot be satisfied under non-positive definite Hessian, meaning the Hessian has zero or negative eigenvalues. Based on the background above, this study examines the positive definiteness of the Hessian and an approximated Hessian used in the Gauss--Newton method. More precisely, this study numerically shows the following insights: 1. the Hessian matrix is not positive definite in large domain; 2. the approximated Hessian matrix is always positive semidefinite by the definition but can be singular in a specific domain close to the stability limit. Additionally, a numerical example that the optimal solution equals the target response, so-called a matched case, shows larger domain where the Hessian matrix is positive definite than that in an unmatched case.

Keywords: Intelligent Control
Abstract: Controlling the dynamic behavior of gene regulatory networks (GRNs) is a crucial challenge in systems biology. Boolean networks (BNs) are widely used to model GRNs, and ``edge removal control,'' which temporarily blocks interactions between genes, has attracted attention as a novel intervention method. In this paper, we propose a reinforcement learning-based method to set stabilization for BNs by edge removal control. The set stabilization problem is to find a controller such that the state converges to the target set and is frequently studied in control of BNs. In addition, Signal Temporal Logic (STL) specifications are imposed as a constraint on edge removal control. Using STL, we can consider important constraints from a biological viewpoint. We introduce flag states to expand the state space, transform the history-dependent control problem into a Markov Decision Process (MDP), and obtain the control policy using Q-learning.

Keywords: Adaptive and Optimal Control, Mechanical Systems Control, Intelligent Control
Abstract: This study presents an application of Iterative Learning Control (ILC) combined with gradient descent Q-learning for anti-sway control of a rotary pendulum system. In the proposed framework, ILC enables improved anti-sway performance through repeated learning iterations, while Q-learning adaptively updates the control policy through a modified gradient descent-based learning process. The main objective is to reduce the oscillation of movements and improve stabilization performance without relying on precise system models. The effectiveness of the proposed method is evaluated using a rotary pendulum system as an underactuated benchmark platform. It is also intended as an initial step toward investigating the feasibility of applying learning-based algorithms to underactuated systems using measured system data. The results demonstrate a gradual improvement in sway suppression and learning performance in successive iterations, indicating the potential of the proposed framework for future learning-based control applications.

Keywords: Intelligent Control, Adaptive and Optimal Control, Process Control
Abstract: This study proposes a novel database-driven control framework to address the limitations of an offline learning approach using a database (DD-FRIT). While many approaches provide good control performance for nonlinear systems, they struggle to adapt to variations in unknown systems. The proposed framework improves the control performance of the conventional DD-FRIT method by incorporating a novel mechanism based on a control performance assessment method and a data similarity evaluation approach using a Self-Organizing Map (SOM). By integrating these structures, the proposed control system can detect the control performance degradation caused by nonlinear variations in the system characteristics, and an appropriate database can then be selected and updated to adapt to the current system. The effectiveness of the proposed method is verified through a nonlinear numerical simulation example and quantitatively compared with the conventional DD-FRIT method. The results demonstrate that the proposed SOM-integrated framework effectively improves the control performance under variations in system characteristics.

Keywords: Nonlinear Control, Adaptive and Optimal Control, Computer Aided Design
Abstract: The stable manifold method computes nonlinear optimal feedback by extending solution trajectories backward in time from a small sphere around the origin. As the trajectories are extended, nonlinearity deforms the iso-time spheres non-uniformly, so that the trajectory ensemble spreads unevenly across the state space. We propose a geometric characterization of this evolution based on dynamic mode decomposition (DMD). The linear operator relating the point clouds on two adjacent iso-time spheres is identified as a finite-dimensional approximation of the underlying Koopman operator, and its singular values and determinant quantify the local directions of stretching and the density of the trajectory ensemble. Under the assumption that the projection of the trajectories onto the state space is locally one-to-one, the characterization is carried out in the state space itself. This geometric characterization is intended to serve as a basis for adaptively refining the trajectory ensemble.

Keywords: Intelligent Control, Robust Control, Mechanical Systems Control
Abstract: In multi-agent shepherding, managing anomalous behaviors such as individual stagnation and flock splitting remains a key challenge. Traditional rule-based controllers rely on fixed logical thresholds, which often lack the flexibility to handle concurrent anomalies in dynamic environments. To address this issue, we propose a dual-layer control architecture that integrates a Large Language Model (LLM) as a slow, high-level decision-maker, alongside a rule-based kinematic vector controller as a fast, low-level executor. The fast layer monitors the environment, detects anomalies, and executes context-specific physical recovery modes tailored to various swarm states. Upon detecting an anomaly, the system translates environmental states into text prompts for the LLM. Based on this contextual information, the LLM is designed to select an appropriate intervention mode and issue commands to the fast layer. This asynchronous integration aims to maintain real-time physical stability while utilizing the LLM’s adaptability to complex situations. This framework is expected to improve the robustness of swarm guidance against concurrent anomalies, offering a practical approach to anomaly resolution in autonomous robotic swarms.

Keywords: Robust Control, Modeling, System Identification and Estimation, Nonlinear Control
Abstract: This poster presents a unified output feedback control framework called receding horizon total control (RHTC), which integrates model predictive control (MPC) and receding horizon estimation (RHE) into a single quadratic program (QP). The analysis derives the closed-loop transfer function matrix of RHTC explicitly from plant, MPC, and RHE parameters, and casts the closed-loop system as a Lur’e system whose nonlinear part is the QP solution map. The circle criterion then provides a direct asymptotic stability guarantee when the normalized closed-loop transfer function matrix is strongly positive real. A numerical example confirms the framework.

Keywords: Nonlinear Control, Adaptive and Optimal Control, Robust Control
Abstract: Safety-critical cyber-physical systems must stay inside a prescribed safe set even when the actuator capacity decreases over time due to wear, faults, or a changing environment. Under such time-varying input constraints, a standard control barrier function (CBF) safety filter can become infeasible. Designing for the initial capacity is then unsafe, while designing for the worst-case capacity is conservative. We study this problem with an adaptive backup CBF framework. The available input bound is treated as a state, and its worst-case decay is predicted together with the physical dynamics. On this extended model, we construct an adaptive recovery law and certificates for return to the backup set and for safety along the recovery path. We prove a viability result: every certified state can be driven to a known invariant backup set while respecting the predicted loss of authority and remaining in the safe set. We also show that the adaptive law gives no smaller progress toward the backup set than a design that holds the capacity at its worst-case value, so it can certify states that the static design cannot. A cart-pole example under decaying actuation supports these results.

Keywords: Modeling, System Identification and Estimation, Nonlinear Control, Adaptive and Optimal Control
Abstract: Collective locomotion in nature, such as ducks swimming in a single-file formation and migratory birds flying in a V-formation, achieves remarkable energy efficiency through hydrodynamic or aerodynamic interactions. To harness such efficiency in engineering systems, we propose a novel control model for swarms interacting through dynamic fields. Rather than explicitly modeling agent-agent interactions, we instead employ PDE models of dynamic fields and kernel-based models of agent-field interactions. These components are integrated within a port-based modeling framework to yield a coupled ODE-PDE system. A distinguishing feature of this model is that the ODE and PDE subsystems are coupled at moving interior points, in contrast to the boundary coupling typical of existing formulations. To demonstrate the framework, we develop a model of floating mobile agents governed by the shallow water equations and conduct numerical simulations. The results confirm that the proposed model reproduces energy-saving effects consistent with those associated with known formation locomotion phenomena.

Keywords: Process Control, Intelligent Control, Computer Aided Design
Abstract: In recent years, the rapid advancement of artificial intelligence has accelerated the development of data-driven control methods for manufacturing systems. Virtual Reference Feedback Tuning (VRFT) has emerged as a promising approach for directly tuning controllers from data without explicit model identification. However, conventional VRFT designs typically employ globally defined linear controllers, which limits their ability to handle time-varying system dynamics while preserving interpretability. This study investigates a method referred to as Model Tree-Based VRFT Design of PID Controllers for Time-Varying Systems, which integrates model trees with VRFT to address these limitations. In the proposed framework, the linear regression models in a model tree are replaced with PID controllers, whose parameters are directly tuned using VRFT. The model tree partitions the input space according to system variations, and each leaf node yields a locally optimized controller corresponding to a specific operating condition. This approach enables piecewise linear control that adapts to time-varying dynamics while maintaining a high level of interpretability. The hierarchical structure of the model tree explicitly represents system variation factors, allowing both reliable control performance and transparent interpretation. Consequently, the proposed method provides an effective and interpretable framework for time-varying systems.

Keywords: Adaptive and Optimal Control
Abstract: The study considers convergence analysis on data-driven iterative learning control (ILC) for a compact form iterative dynamic linearization (CFIDL) method. The data-driven ILC based on CFIDL formulates ILC algorithm consisting of single parameter updating and input signal updating at each discrete-time instant. The algorithm enables us to achieve asymptotically tracking the given reference signal by repeating trials. The existing study ensures convergence on the assumption of the sign of the parameters. However, the parameter depends on the past input signals as well as the controlled system. Hence, it encounters difficulty to derive the condition for holding the assumption. Therefore, the study considers the condition for holding the assumption. The numerical simulation experiments support the analysis and discussions on the study.

Keywords: Adaptive and Optimal Control
Abstract: For the horizontal movement of a quadcopter, the cascade control structure is required, where the outer feedback controllers generate the desired attitude angles using the information of current positions as feedback signals and the inner feedback controllers generate desired input signal to follow the desired attitude. The attitude angles are determined by the roll and pitch angles Therefore, the study introduces a two-input, two-output PID controller for inner feedback loop, and proposes virtual reference feedback tuning (VRFT) for PID gain tuning. The PID controller is also introduced into the outer feedback loop, and the PID gains are tuned by the VRFT approach. The effectiveness of the proposed method is demonstrated by numerical simulation experiments.

Keywords: Adaptive and Optimal Control
Abstract: The study proposes a design method for dual-stage iterative learning control (ILC) with Markov parameters. The dual-stage ILC formulates ILC algorithm consisting of single parameter updating and input signal updating at each discrete-time instant. The approach is based on compact form iterative dynamic linearization (CFIDL) method and allows us to apply nonlinear systems. However, the approach assumes that the sign of the parameters are known. Since the parameters depend on the past input data, it encounters difficulty to ensure the assumption even when the controlled systems are single-input, single-output, linear, time-invariant systems. Therefore, the study modifies the dual-stage ILC to introduce Markov parameters to represent controlled systems for single-input, single-output, linear, time-invariant systems. The proposed approach requires updating law for multiple parameters but allows a deeper convergence analysis. The study shows numerical experimental results to show the effectiveness of the proposed method.

Keywords: Adaptive and Optimal Control
Abstract: The study considers a method for the data-drive iterative control (ILC) method for arm angular position control in a rotational pendulum system. The data-driven ILC method updates control law and system parameters at every iteration and achieves asymptotically tracking to a prescribed target reference signal by repeating iteration. The study applies the data-driven ILC method to arm angular position control in a rotational pendulum system. In the controlled system, the motion of the pendulum interferes with the motion of the arm as disturbance signals. Hence, the purpose of the study is to show that the data-driven ILC enables us to realize desirable tracking properties even for nonlinear systems together with the presence of disturbances. The effectiveness of the data-driven ILC is shown through numerical simulation experiments.

Keywords: Adaptive and Optimal Control
Abstract: The study considers applying the virtual reference feedback tuning (VRFT) method for active vibration control. The VRFT allows us to update the control parameters using collected data from the measured data instead of using the system model. The existing VRFT method updates controller parameters using measured time series data and the reference time series data to follow. However, vibration control usually specifies the specification of the controller in the frequency domain. Therefore, it is desirable that the VRFT for active vibration control could be formulated in the frequency domain. The study introduces modal analysis approach to active vibration control. Then, a modified VRFT approach in frequency domain is introduced. The effectiveness of the proposed approach is shown through a numerical experimental result.

Keywords: Adaptive and Optimal Control, Nonlinear Control, Computer Aided Design
Abstract: In this research, we consider how to calculate an upper bound of the number of iteration in the gradient method in Model Predictive Control (MPC) optimization. A considered model is a Lagrange system, which is a class of nonlinear systems. In MPC, optimization problems need to be solved repeatedly at each time step. Therefore, the difficulty of solving the optimization problems depend on the current state, and the increase in the computational cost is a significant issue to be considered. Under a stopping criteria using the gradient norm, the upper bound of the number of iterations is approximately calculated and compared with the theoretical upper bound depending on a Lipschitz smoothing parameter. In nonlinear systems, it is difficult to obtain the Lipschitz smoothing parameter globally, so a local Lipschitz smoothing parameter is calculated by the linear approximation of the system. A numerical example compares the proposed and theoretical number of the iteration and examines the proposed method.

Keywords: Robotic and Automation Systems, Human-Machine Systems, Modeling, System Identification and Estimation
Abstract: In walking assist suits using physical reservoir computing for state estimation with pneumatic artificial rubber muscles, a challenge remains in continuously controlling the force of driving artificial muscles in accordance with walking movements. To properly assist the wearer's walking using artificial muscles, appropriate control tailored to the wearer's body movements is required. In this report, toward the design of a control method that automatically exerts an assist force synchronized with the wearer's muscle activity, we estimated the wearer's surface electromyography (EMG) from the internal pressure of the artificial muscles while wearing the assist suit and conducted an actual device driving test based on this estimation. Experimental results confirmed that the EMG estimated from the internal pressure of the sensor artificial muscles exhibited waveforms similar to the measured EMG. In the driving test, we continuously generated target pressure command values in real time from the artificial muscle internal pressure during walking, confirming that assist control synchronized with walking movements is feasible.

Keywords: Robotic and Automation Systems, Mechatronics Systems
Abstract: In this study, we propose a flexible sheet-type inverter device using a Fluidic Sheet Transistor (FST). The proposed inverter induces traveling waves using a single-line pneumatic input, thereby enabling the deployment of mobile robots in environments such as (i) confined spaces where tubing is prone to entanglement, (ii) environments sensitive to electromagnetic noise, and (iii) conditions in which electronic devices are easily damaged. As an application example, by combining it with a sheet-type actuator, we demonstrate an electric-free soft mobile robot system capable of load transportation using only a DC pneumatic source as input.

Keywords: Process Control
Abstract: In this paper, we consider data-driven control of odd-function typed Hammerstein nonlinear systems. This nonlinear system model is used to denote large class of non-linear systems such as Twisted and Coiled Polymer Actuator and so on. Here, we address the internal model control architecture to control this class of nonlinear systems by using data and without mathematical model As a result, the proposed method yield not only suitable control performance but also nonlinear mathematical model

Keywords: Robotic and Automation Systems, Modeling, System Identification and Estimation, Mechanical Systems Control
Abstract: McKibben pneumatic actuators (MPAs) are widely used in soft robotics due to their flexibility and high power-to-weight ratio. Length estimation from pressure and tension signals, without relying on external displacement sensors, would broaden the applicability of MPAs in practical robotic systems where sensor installation is difficult or undesirable. While physical reservoir computing (PRC), which exploits the MPA's own dynamics as a computational resource, has been proposed for this task, its advantages over alternative approaches have not been systematically evaluated. In contrast to prior PRC studies that required specially fabricated actuators, our approach achieves length estimation using a standard MPA constructed from off-the-shelf materials. In this paper, we compare three approaches on the same dataset: PRC, Echo State Network (ESN) as an artificial reservoir approach, and ARX model as a linear baseline. Results show that PRC achieves superior generalization performance with far fewer parameters than ESN, and outperforms the linear ARX model despite sharing the same parameter count. These results demonstrate the practical value of exploiting physical dynamics as a computational resource for compact and robust soft sensing.

Keywords: Modeling, System Identification and Estimation, Medical and Welfare Systems, Robotic and Automation Systems
Abstract: Pneumatic soft exosuits require the ability to estimate and control the wearer's state while maintaining lightness and flexibility. For the design of pneumatic reservoir structures suitable for such systems, evaluating multiple structural conditions only by hardware experiments is costly and time-consuming. This study therefore aims to establish a simulation-based framework for evaluating pneumatic reservoir structures for soft exosuits. First, a simulator of a pneumatic reservoir system driven by pneumatic artificial muscle (PAM) length deformation was constructed, and its validity was examined by comparing simulated and experimental pressure responses. Next, reservoir-computing-related metrics were evaluated using causally embedded pressure-state vectors constructed from a single closed-end pressure signal, so as to preserve online implementability without using future samples. These results, obtained under a single representative pipeline and orifice configuration, provide a basic validation step toward future simulation-based structural evaluation of pneumatic reservoirs for soft exosuits.

Keywords: Human-Machine Systems, Medical and Welfare Systems, Modeling, System Identification and Estimation
Abstract: Soft wearable suits actuated by pneumatic artificial muscles (PAMs) are suitable for human motion assistance and provide lightweight, comfortable assistance. However, existing PAM suits utilizing antagonistic actuation mechanisms suffer from slack periods that prevent assistance and opposing tension that hinders movement. This study develops a slack-free walking assist suit employing constant-tension control. We attached movable pulleys to the lower-leg PAMs to increase their displacement. To reduce the user’s physical burden while maintaining constant tension, we connect tension sensors to the PAMs and continuously control their internal pressures. Experiments confirmed the elimination of PAM slack periods, despite limited performance in tension tracking. Furthermore, preventing PAM slack inherently improved the accuracy of posture estimation via reservoir computing.

Keywords: Identification and Estimation, Human-Machine Systems, Medical and Welfare Systems
Abstract: Accurate and robust estimation of human intention is one of the most important criteria for successful human-machine interaction. In the field of lower-limb exoskeletons, this robustness is required when gait conditions vary due to environmental disturbances or human error. Existing approaches often rely on narrow operating ranges of walking speeds and environmental conditions, which limit their applicability in real-world scenarios. In this work, we introduce a method for human intention recognition on the joint level that maintains consistent performance over different levels of dynamics by leveraging principles of physical reservoir computing. Additional robustness is achieved by compensating for fluctuations in sensor measurements due to external disturbances. The proposed system uses internal pressure measurements from a Pneumatic Artificial Muscle (PAM) to estimate the wearer’s joint angular velocity.

Keywords: Robotic and Automation Systems, Autonomous Decentralized Systems, Intelligent Systems
Abstract: This paper improves the behavioral rule extraction of evolutionary robotic swarms using output-discretizing neuroevolution. Swarm robotics, which is the research of decentralized collective robotics, discusses how to design the controller of each robot. Neuroevolution is a promising approach to automatically design robot controllers. However, designed neural networks generally become a black box for the obtained behavioral rules. This paper proposes an improvement approach to behavioral rule extraction from neuro-evolutionary robotic swarms. In a previous work, the evolved behavioral rules were extracted by reconstructing the controller using human-readable decision trees. In this paper, the output-discretizing neuroevolution was conducted to make the robots achieve the task with a finite number of output values of a controller; this makes the reconstruction a simpler classification problem. The proposed method was examined by the path formation task with a hundred robots. The results showed that the output-discretizing neuroevolution achieved similar task performance to the original ANN that uses continuous outputs. In addition, the results also showed that the proposed method successfully improved the controller reconstruction.

Keywords: Autonomous Decentralized Systems, Robotic and Automation Systems, Intelligent Systems
Abstract: Embodied evolution is a decentralized evolutionary robotics paradigm in which robots continuously evolve their controllers through local interactions in their environment. Previous work showed that incorporating Mutation-Based Evolving Artificial Neural Network (MBEANN) improves a simple embodied evolution framework. This paper presents an ablation study that disables key components of MBEANN to assess their impact. MBEANN and its ablated variants are evaluated on a two-target navigation task performed by a robotic swarm in simulation. Results show that removing key components slightly degrades performance, indicating their positive contribution to the embodied evolution framework.

Keywords: Biological and Physiological Engineering, Mechanical Systems Control
Abstract: Crocodilian hindlimbs support massive bodies and enable terrestrial locomotion. Previous studies have suggested that the musculotendinous branching structure in the hindlimb of crocodilians contributes to supporting a semi-erect posture. In this study, we developed a three-dimensional musculoskeletal model incorporating this branching structure and conducted dynamic simulations. The model maintained a semi-erect standing posture via active contraction of the caudofemoralis longus muscle, together with passive elastic forces of associated tendons. The resulting motion showed inter-joint coordination among the hip, knee, and ankle, consistent with previous two-dimensional and robotic studies. These findings suggest that the branching structure contributes to passive inter-joint coordination in crocodilian limb support. This framework provides a foundation for biomechanical analyses of crocodilian limb movements and informs bioinspired robotic design, and the reconstruction of locomotor function in extinct archosaurs.

Keywords: Robotic and Automation Systems, Autonomous Decentralized Systems, Intelligent Systems
Abstract: Outdoor navigation methods often adopt approaches based on environment map generation and self-localization using exteroceptive visual sensors such as cameras and LiDAR. However, real outdoor environments are unspecified environments that change from moment to moment in accordance with physical laws. Therefore, there are cases in which only the exteroceptive sensors of robots cannot recognize regions that are traversable. To address this problem, this study proposes a primitive swarm robot navigation method that utilizes robots that have become stuck in unspecified environments. The proposed method generates, in a self-organized manner, paths that avoid stuck regions by means of random behavior of robots and the generation of attractive regions. Numerical simulations demonstrate that the proposed method achieves a higher traversal success rate than conventional methods, and that randomness is a major factor in avoiding trap regions.

Keywords: Intelligent Systems, Social Systems
Abstract: Real-world relationships between individuals, corporations, and nations dynamically shift between cooperation and non-cooperation as circumstances evolve. While the infinitely repeated n-player Prisoner's Dilemma explains cooperation maintenance via trigger strategies, it fails to account for proactive strategic shifts driven by changing conditions. Conversely, traditional Minority Games focus almost exclusively on non-cooperative behavior.This study proposes a novel Minority Game where player states evolve, creating a biphasic dynamic: an early stage favoring cooperation for collective scoring and a late stage favoring non-cooperation as players approach the 10-point victory condition. Using Counterfactual Regret Minimization (CFR), we analyze the optimal strategy for this extensive-form game with imperfect information to elucidate the specific timing and mechanisms behind the transition from cooperation to non-cooperation.

Keywords: Autonomous Decentralized Systems, Robotic and Automation Systems, System Engineering
Abstract: This paper proposes a new framework to address the Sim-to-Real gap, i.e., the performance degradation that occurs when deploying swarm robots trained in simulation into real-world environments. Unlike conventional domain randomization, which typically relies on trial-and-error, applying SHAP analysis as an Explainable AI method enables the quantitative identification of specific sensor inputs that contribute to the discrepancy with reality. By performing data-driven domain randomization targeted specifically at these identified critical elements, our experiments successfully improved the reproducibility of collective behavior and narrowed the gap between simulation and reality to a statistically negligible level. This approach enables systematic, explainable robot control design that does not rely on heuristics, providing a vital foundation for constructing highly reliable swarm robotic systems in complex environments.

Keywords: Signal and/or Image Processing, Computational Intelligence, Entertainment Systems
Abstract: In this study, we develop a material classification framework of a puppet head using X-ray CT images for digital archiving of “Ningyo Joruri” which is a traditional Japanese puppet theater. A method using U-Net model is first applied for five-class segmentation, followed by a two-stage correction framework consisting of a 3D CNN for misclassified location detection and an lmKAN model for material relabeling. Experimental results show that our proposed method improves the accuracy for all materials, particularly reducing errors caused by metal artifacts and clarifying material boundaries.

Keywords: Sensors and Transducers
Abstract: Photoacoustic effect is a promising modality for investigating the inside of a sample by generating ultrasound in a non-contact manner via optical excitation. While conventional photoacoustic excitation has relied on near-infrared wavelengths, recent studies have used sub-terahertz (sub-THz) waves. However, previous studies of sub-THz photoacoustics have focused only on transmission-mode detection, in which the generated ultrasound is detected on the opposite side of sub-THz irradiation, thus being subject to geometric constraints. In this study, we construct an experimental platform for reflection-mode sub-THz photoacoustic sensing. In our experiment, we use a water-filled PET container as a test object. To attain reflection-mode sensing, modulated sub-THz waves for ultrasound generation and infrared light from a laser Doppler vibrometer (LDV) for detection are irradiated on the same point on the surface. A key to implement this configuration is to adopt oblique incidence for the sub-THz irradiation, which allows for coaxial detection of the ultrasound by the LDV. We demonstrate that after acquiring ultrasonic spectra via frequency sweep, we can extract the corresponding time-of-flight (ToF) data by means of inverse Fourier transform, which clearly captures multiple reflections in the container in the time-domain. This demonstration shows the potential of the reflection-mode non-contact ultrasonic sensing based on the sub-THz photoacoustic effect.

Keywords: Signal and/or Image Processing, Sensors and Transducers, Opto-Electronic Measurement
Abstract: Hyperspectral imaging provides rich spectral cues for material recognition; however, the high dimensionality of spectral bands not only increases computational and memory costs but also demands a large amount of training data for effective learning. To address these challenges, we explore the design of optical filters that reduce spectral dimensionality while preserving discriminative information for waste segmentation. Specifically, we propose a method to design optical filters for SWIR-based imaging systems using hyperspectral data, enabling efficient waste segmentation with fewer spectral bands. We adopt a two-stage framework. In Stage 1, the spectral transmission curves of K optical filters are optimized end-to-end from sampled pixel spectra using a lightweight spectral network, allowing efficient use of limited training data under non-negativity, smoothness, and orthogonality constraints. In Stage 2, the learned filters are fixed and applied to hyperspectral data to obtain a K-channel representation, which is then used for semantic segmentation with SegFormer-B0. Experiments on public SpectralWaste dataset show higher foreground mIoU with three learned filters than with the full-band and PCA-3 baselines under same SegFormer-B0 backbone for Stage 2 segmentation. These results suggest that the proposed method provides an effective and compact front-end for SWIR-based waste segmentation, enabling practical deployment without requiring full hyperspectral acquisition.