Keywords: Artificial intelligence, Energy Systems, Industrial applications
Abstract: Day-ahead energy consumption forecasting re-quires periodic model updates to track site-specific and seasonal distribution shifts. Local retraining enables such adaptation without cloud infrastructure, but the suitability of deep learning architectures for memory-constrained retraining remains understudied. We propose FP-CNN, a resource-aware architecture combining a Fourier series-based periodic prior with a dilated 1D-CNN encoder, evaluated across three building sites covering summer and winter conditions. FP-CNN achieves 10× lower peak training memory and 43× fewer inference FLOPs than an attention-based baseline, converges up to 8.5× faster than a re-current encoder, and maintains competitive forecasting accu-racy. Ablation studies confirm that the Fourier prior plays a central role in improving accuracy when combined with the CNN encoder. These results demonstrate that FP-CNN offers a practical balance between forecasting accuracy and resource-aware training efficiency for site-adaptive building energy fore-casting.

Keywords: Computational intelligence, Robotics and swarm intelligence, Artificial intelligence
Abstract: Meta-heuristic algorithms are indispensable for solving complex optimization problems, while balancing exploration and exploitation stands as a pivotal challenge in their design. Inspired by the adaptive behavior of biological systems that maintain exploratory drives under external pressures, a Pressure-Driven Adaptive Optimizer (PDAO) is proposed. It dynamically regulates search strategies by simulating the antagonism and adaptation between intrinsic exploratory drive and environmental pressure, thereby achieving an effective equilibrium. The performance of PDAO is validated through experiments on the CEC 2022 benchmark suite and a real-world engineering design problem, compared with five well-established meta-heuristic algorithms. Results demonstrate that PDAO achieves superior overall performance, exhibiting both stability and effectiveness.

Keywords: Control theories, Cyber-physical systems and security
Abstract: This paper establishes a new lifting-based framework for computing the finite-horizon L_∞-gain of sampled-data systems with aperiodic and uncertain sampling instants. An extended version of the conventional lifting technique is proposed to address aperiodic sampled-data systems. Based on a linear parameter-varying representation derived through the extended lifting technique, we finally propose a gridding-based approach to the L_∞-gain analysis, together with a convergence analysis that ensures an asymptotically accurate computation.

Keywords: Control theories, Cyber-physical systems and security, Intelligent control
Abstract: This paper proposes an optimal event-triggered control (ETC) method using Model Predictive Control (MPC) to mitigate network congestion in Networked Control Systems (NCSs). By predicting future plant states, the controller simultaneously optimizes control inputs and transmission timing to suppress unnecessary communication. Experimental evaluations using a rotary inverted pendulum demonstrate that the proposed method reduces data transmission by 29.6 % while improving state performance by 28.7 % compared to conventional time-triggered control. This approach effectively balances communication efficiency and control precision.

Keywords: Energy Systems, Artificial intelligence, Intelligent control
Abstract: The transition toward sustainable energy systems is increasingly shaped by distributed energy resources, prosumers, flexible demand, and digital intelligence. Reinforcement learning (RL) has emerged as a promising tool for adaptive control in such environments, especially in decentralized settings. However, most existing learning-based control frameworks still treat humans as exogenous disturbances, static preference holders, or passive recipients of automated decisions. This position paper argues that such a view is increasingly inadequate. As energy systems become more distributed and participatory, future control architectures must explicitly account for human preferences, trust, feedback, and interaction costs. We position human-centric reinforcement learning as a key direction for next-generation energy systems and argue that large language models (LLMs) can serve as semantic interfaces between human intent and formal control layers. Building on recent developments in human-in-the-loop energy management, preference elicitation, and LLM-assisted decision support, this paper identifies key research gaps and outlines a research agenda for trustworthy, scalable, and human-aware energy system control.

Keywords: Industrial applications, Artificial intelligence, Intelligent control
Abstract: Chemical mechanical polishing (CMP) is a critical planarization process in advanced semiconductor manufacturing, where precise control of the material removal rate (MRR) is essential for high-quality wafer production. Traditional run-to-run (R2R) control methods, including exponentially weighted moving average (EWMA), neural network-based approaches, and reinforcement learning (RL), often struggle to achieve both robustness and efficiency under dynamic disturbances. In this study, we propose an N-step RL framework for CMP control, which extends the agent’s foresight by accumulating rewards over multiple time steps. Integrated with an Actor-Critic architecture, the method enables more accurate, stable, and efficient updates compared to single-step approaches. Simulation results demonstrate that the proposed N-step RL algorithm outperforms existing methods in terms of control performance, convergence speed, and robustness to both static and dynamic disturbances.

Keywords: Intelligent control, Cyber-physical systems and security, Control theories
Abstract: This paper proposes a tracking control method utilizing Gaussian process regression to maintain a moving target within a UAV’s field of view. By leveraging estimation variance to enforce distance constraints and combining model predictive reference generation with PD control, the method prevents excessive proximity and input saturation. Simulations and experiments demonstrate stable tracking and FOV maintenance even under observation loss and disturbances.

Keywords: Intelligent control, Cyber-physical systems and security, Control theories
Abstract: This paper proposes a seamless controller redesign framework for networked control systems (NCSs) with unknown Markovian data dropouts. The method determines optimal update timing by detecting information gain saturation via Kullback-Leibler (KL) divergence, without relying on prior knowledge. This enables data-driven adaptation based solely on observed communication logs. Experiments using a rotary inverted pendulum validate that the approach mitigates performance degradation caused by update delays and maintains robustness under uncertain network conditions.

Keywords: Industrial applications, Artificial intelligence, Intelligent control
Abstract: Achieving wafer-level temperature uniformity is critical in advanced semiconductor manufacturing processes involving multizone heated bake plates. This paper proposes a drift-aware distributional reinforcement learning (DRL) framework with improved experience replay (IER) for run-to-run (R2R) temperature uniformity control. A fuzzy-regularized radial basis function (RBF) neural network approximates the nonlinear mapping from heater control inputs to the wafer temperature distribution. The batch control task is reformulated as a Markov decision process (MDP), embedding process drift estimation via exponentially weighted moving average (EWMA) into the RL state representation. A distributional soft actor-critic (DSAC) algorithm combined with a dual-buffer IER mechanism learns a control policy that minimizes temperature deviation and mitigates inter-batch disturbance accumulation. CFD simulation results under both ideal and drift conditions demonstrate that the proposed DSAC-IER strategy outperforms baseline methods in temperature uniformity and robustness to drift.

Keywords: Industrial applications, Control devices, sensors and actuators, Energy Systems
Abstract: Modern large-scale research facility, such as the European Spallation Source (ESS), require highly distributed yet strictly coordinated safety architectures to protect personnel from ionizing radiation hazards. This paper presents the design and implementation of the Nexus Personnel Safety System (Nexus PSS), a centralized supervisory interlink designed to orchestrate safety logic across multiple independent domains, including the accelerator, target station, bunker, and neutron instruments. The Nexus PSS functions as a high-integrity coordination hub that manages the “Ready for Beam” permit chain. The paper describes a logic orchestration strategy that aggregates safety data from distributed failsafe PLCs without compromising the autonomy of local safety functions. This orchestration is executed over a redundant fiber-optic ring network to ensure robustness, scalability, and maximum availability. The paper details the fail-safe communication architecture, the mapping of global interlock logic, and the strategies used to maintain safety integrity during partial facility operation.

Keywords: Industrial applications, Control theories, System identification and modelling
Abstract: The closed-loop automated control of coal gasifiers is a prevailing trend in the industry. Although combustion chamber temperature governs both product yield and safety, real-time measurement is precluded by the harsh internal environment of high heat and slag. Coupled with complex physics, significant time delays, and inherent difficulties in modeling the gasification processes, the current level of automation remains limited. To address these challenges and achieve optimal operation, a closed-loop temperature control system for the coal gasifier is presented, based on a data-driven model predictive control framework integrated with the input-mapping method. First, given that accurate temperature data cannot be obtained in real-time from the combustion chamber, the radiant waste heat boiler temperature is adopted as a surrogate control indicator. A dynamic mathematical model relating the temperature of the coal gasifier radiation waste pot to the oxygen flow rate is derived via parameter identification using historical plant data. This model is extended to explicitly incorporate the system delay. To compensate for the inevitable model-process mismatch and the persistent delay effects, the input-mapping method is introduced. This method leverages continuously updated historical data for online correction. The effectiveness of the proposed method is validated through dynamic simulations performed in Aspen Plus.

Keywords: Industrial applications, Measurement and instrumentation
Abstract: Aiming at the challenge of rapid gas-plume discovery in unknown three-dimensional environments, this paper proposes an olfaction-feedback-based wide-area adaptive spiral search algorithm, termed A3D-SSOF. A three-dimensional Archimedean spiral combined with an online threshold calibration mechanism is employed to enable robust target detection in noisy environments. Multiple sensing features, including concentration magnitude, concentration gradient, effective hit density, and information entropy, are fused through a softmax-based adaptive weighting strategy to dynamically regulate the radial step size and axial pitch. Furthermore, a distance- and memory-constrained restart mechanism and a joint termination mechanism are designed. Comparative, ablation, and real-world flight experiments demonstrate improved search efficiency and robustness under the tested conditions.

Keywords: Industrial applications, Nonlinear control and applications, System identification and modelling
Abstract: Post-combustion CO₂ capture using monoethanolamine (MEA) absorption must operate effectively under dynamic loads from renewable energy integration, requiring accurate dynamic models for control system design. This paper presents a comprehensive rate-based dynamic model of an MEA-based CO2 absorber implemented in MATLAB for control-oriented studies. The model incorporates material and energy balances, two-film mass transfer theory with chemical reaction enhancement, and comprehensive physical property correlations. Steady-state column profiles demonstrate physically consistent behavior with CO2 capture efficiency on baseline conditions. Open-loop dynamic simulations reveal substantial nonlinear and asymmetric responses to step changes in gas flowrate, liquid flowrate, and inlet temperatures, with capture efficiency variations up to 3 percentage points. Several responses exhibit non-minimum phase characteristics and oscillatory behavior, highlighting control challenges. Closed-loop performance is evaluated using a PI controller designed through relay autotuning, demonstrating successful disturbance rejection with fast recovery to setpoint and minimal steady-state error. The developed model provides a practical platform for investigating dynamic behavior and designing control strategies, with applications to nonlinear control development and integrated absorber-stripper control systems.

Keywords: Industrial applications, Robotics and swarm intelligence, Control devices, sensors and actuators
Abstract: Rock-sampling robots are essential equipment for the intelligence of field geological exploration. However, in actual operations, robots face severe challenges such as unstructured field environments, intense illumination fluctuations, and highly variable rock lithology. Therefore, relying on a self-developed field rock-sampling robot platform, this paper proposes a multimodal fusion perception method for field environments to achieve high-precision segmentation of field rock regions. Addressing the issue of irregular terrain distribution, a multi-sensor fusion mapping method is employed to construct a 3D map with accurate geometric structures and rich texture colors. Based on this map, by integrating the geometric structures of point clouds with visual texture information, a multimodal collaborative constrained unsupervised rock segmentation algorithm is designed. This algorithm leverages physical priors to effectively overcome interference from drastic light changes and resolves the problem of insufficient annotated data caused by dynamic lithological variations. Field experiments demonstrate that the rock segmentation accuracy reaches 89.57%, a significant improvement over the traditional region-growing method.

Keywords: Industrial applications, System identification and modelling, Nonlinear control and applications
Abstract: An efficient inventory system plays a vital role in improving supply chain performance and profitability. This study develops a decision-support framework for a cooperative multi-warehouse inventory system that integrates supplier selection under uncertain operating conditions. The proposed approach models coordination among warehouses so they can jointly satisfy overall demand and control the inventory level to a predefined safety point while sharing resources and inventory decisions. To explicitly address uncertainty in demand and supply variability, the problem is formulated as a chance-constrained optimization model, ensuring that key operational and service requirements are satisfied with a predefined probability level. The resulting stochastic mathematical program is implemented in Python and solved using the Gurobi solver. Numerical experiments demonstrate that the proposed model can produce cost-effective and reliable decisions while maintaining inventory and target service levels. The results provide practical managerial insights and a useful analytical tool for supply chain practitioners seeking more resilient and risk-aware inventory planning.

Keywords: Medical and financial systems, Artificial intelligence, Deep learning and machine learning
Abstract: This study explored the feasibility of using magnetocardiography (MCG) combined with machine learning for early non-invasive screening of aortic stenosis (AS) with concomitant mitral regurgitation (MR), a common multivalvular disease (MVD). A total of 119 patients with AS and MR and 89 healthy controls were retrospectively enrolled. MCG data were acquired using a 36/64-channel spin-exchange relaxation-free MCG system. After preprocessing, morphological, statistical, and textural features were extracted from one-dimensional signals and two-dimensional images, with 61 significant features selected via univariate analysis. Six machine learning models (LR, RF, LASSO, SVM, RFE, GB) were trained and evaluated. The Random Forest (RF) model achieved the best performance, with an AUC of 0.981, accuracy of 93.3%, sensitivity of 87.9%, and specificity of 100%. Feature importance analysis indicated that T-wave, R-wave, and QS-wave related features were most contributory, consistent with known cardiac electrophysiological abnormalities. This study preliminarily demonstrate that an MCG-based machine learning screening model shows high accuracy and potential for identifying AS with MR, offering a new approach for early non-invasive screening of MVD. Future work should involve larger cohorts, integration of deep learning, and multicenter validation for further optimization and translation.

Keywords: Nonlinear control and applications, Control theories
Abstract: This paper considers nonlinear input-affine systems with a fixed input delay and presents a framework to ensure the closed-loop system is passive under safety-critical control feedback. The passivity property is crucial in many physical systems, as it offers inherent stability and robustness against passive unmodeled dynamics, while safety enforces reliable behavior with trajectories remaining within a desired safe set. However, the presence of input delay in the system dynamics may disrupt this balance by introducing a lag in the control action, compromising both passivity and safety. To address these problems, a predictor-based feedback approach has been presented. Further, the passivity notion has been extended to the time-delay system. A set of conditions is then formulated to ensure the preservation of passivity of the nonlinear closed-loop system when the input-delayed system is subjected to control barrier function (CBF)-based safety filtering. A stricter sufficient condition is also provided, which connects the choice of the overall storage function to the CBF while guaranteeing passivity preservation and safety. Finally, a simple example is presented to illustrate the effectiveness of the proposed approach.

Keywords: Nonlinear control and applications, Control theories
Abstract: This paper pursues stabilizing control design for nonlinear systems estimating the largest domain guarantee for the largest admissible uncertainty. Recently, the state space domain and the uncertainty bound were formulated by two merging curves computed from regionally input-to-state stable control Lyapunov functions (ISS-CLFs), which are independent of controllers. The preceding study did not give computational methods other than brute force computations in the entire space of disturbance and state variables. This paper shows how to compute the curves providing the stability and robustness domain estimates efficiently in a recursive manner. The development also provides a way to construct or modify ISS-CLFs avoiding excessive flattening of the domain estimates.

Keywords: Nonlinear control and applications, Control theories, Autonomous vehicles
Abstract: Control Barrier Functions (CBFs) ensure safety for control systems by rendering a prescribed safe set forward invariant. For safety constraints with relative degree greater than one, standard zeroing CBF (ZCBF) class is not directly applicable, and existing ZCBF constructions may not guarantee forward completeness, allowing finite escape times. This paper proposes a novel ZCBF design method for systems with high relative degree constraints obtained via input-output linearization. The resulting ZCBF has an explicit closed form and yields a safety filter that guarantees forward invariance together with forward completeness on a domain containing the safe set. Under locally bounded inputs, making the entire safe set forward invariant is generally impossible.Therefore, we explicitly characterize a non-empty strict forward-invariant subset under the proposed ZCBF mechanism. Simulations demonstrate effectiveness of the resulting ZCBF.

Keywords: Nonlinear control and applications, Control theories, Cyber-physical systems and security
Abstract: This paper investigates parameter-dependent H∞ filtering for networked control systems with measurement quantization and packet dropouts. Logarithmic quantization error is represented by sector-bounded uncertainty, and packet loss is modeled by a Bernoulli process. A full-order filter with feed-through term Df is considered, and parameter-dependent Lyapunov functions with slack variables are used to derive less conservative LMI conditions. Two simulation examples demonstrate improved attenuation performance compared with existing methods.

Keywords: Nonlinear control and applications, Control theories, Intelligent control
Abstract: This paper addresses the nonfragile static output feedback control problem for continuous-time nonlinear systems represented by Takagi-Sugeno (T-S) fuzzy models. First, the integral Lyapunov function is developed to substantially reduce the conservatism associated with conventional Lyapunov approaches. Second, multiplicative gain variations are incorporated to account for inevitable implementation errors in practical systems, thereby modeling controller fragility. Third, the nonfragile controller design is formulated directly as strict linear matrix inequalities (LMIs), eliminating the need for complex iterative algorithms. This computationally efficient approach ensures both asymptotic stability of the closed-loop system and the prescribed H∞ performance level. Finally, the effectiveness of the proposed control scheme is validated through two simulations.

Keywords: Nonlinear control and applications, Control theories, Motion and vibration control
Abstract: In recent years, with the increasing prevalence of drones, obstacle avoidance control has become crucial. To resolve the inherent challenges of conventional avoidance methods, such as Artificial Potential Fields (APF) and Model Predictive Control (MPC), and to theoretically guarantee system safety, Control Barrier Functions (CBFs) have attracted significant attention. Furthermore, while CBF designs utilizing the backstepping method have been proposed to reduce computational load, previous studies have focused solely on the attitude dynamics of the drone; the entire system dynamics from position to attitude, which are essential for obstacle avoidance, were not considered. Therefore, this paper proposes an extended CBF design method that integrates both the rigid-body position and attitude dynamics of a drone for the entire system using the backstepping method. Our approach theoretically guarantees the safety of the entire system by progressively expanding position-related constraints via the backstepping process. We construct an integrated control law combining an assist input for obstacle avoidance, derived from the designed extended CBF, with a stabilization input based on a Control Lyapunov Function (CLF). Finally, numerical simulations demonstrate that the proposed method successfully achieves safe obstacle avoidance and convergence to the target state simultaneously.

Keywords: Nonlinear control and applications, Industrial applications, Deep learning and machine learning
Abstract: With the increasing integration of renewable energy sources and the proliferation of non-linear power electronic loads, modern power grids are confronting significant challenges to Power Quality (PQ). Harmonic pollution, in particular, exhibits high dynamicity and non-stationary characteristics. Traditional harmonic analysis methods, which rely on steady-state assumptions, often fail to accurately identify the precise start and end times of transient harmonic events—a challenge known as the Change Point Detection (CPD) problem. Existing solutions encounter a ``Local-Global'' feature dilemma: Convolutional Neural Networks (CNNs) lack the ability to model long-range dependencies, whereas standard Transformer architectures are hampered by high computational complexity and insensitivity to subtle local variations. To bridge this scientific gap, this paper proposes a harmonic voltage change point detection framework based on a textbf{Local Information Enhanced Neural Network (LIENN)}. The architecture innovatively integrates an textbf{Information Enhanced Neural Network} for morphological feature extraction with a textbf{Shifted Window Self-Attention (SWSA)} mechanism, enabling the capture of global harmonic mode evolution with low computational overhead.

Keywords: Nonlinear control and applications, System identification and modelling, Autonomous vehicles
Abstract: This study investigates an autonomous navigation control method for a quadruped robot. Global exploration is an important task for mobile robots, and Monte Carlo Model Predictive Control (MCMPC) is used to achieve it. However, legged robots exhibit body oscillations during locomotion. To address this issue, smoothing is applied to the discontinuous MCMPC solutions. Gradient-based MPC is then used to refine the MCMPC solution using it as the initial value. This enables smoother locomotion. Physics simulations were conducted to validate the proposed method. In scenarios involving both local and global optima, gradient-based MPC and MCMPC alone either became trapped in a local optimum or exhibited unstable behavior, whereas the proposed method successfully　transitioned to the global optimum. Furthermore, in the scenario with obstacles, the proposed method enabled the robot to reach the target position without stagnation while avoiding obstacles. These results confirm that the proposed method enables global navigation with continuous locomotion for quadruped robots despite body oscillations during walking.

Keywords: System identification and modelling
Abstract: This paper studies the control-oriented recursive identification of finite impulse response systems with binary-valued observations. Inspired by the Maximum Likelihood method, a novel recursive algorithm is proposed using the statistical property of system noises and observations. Unlike existing research, the gradient of the proposed algorithm is derived from the local likelihood function, which has not been previously considered. The core advantage of the algorithm is the adaptation of the recursive weight term, and especially, it has an accelerating effect when the estimated value deviates far from the true value. The proposed algorithm is proved to be convergent in both almost sure and mean square sense. Furthermore, the almost sure and mean square convergence rates are also obtained under some mild conditions. A simulation is presented to demonstrate the advantage of convergence rate over existing algorithm.

Keywords: Adaptive systems, Autonomous vehicles, Intelligent control
Abstract: When targeting autonomous operation in a wide range of scenarios, imposing predefined structures on the uncertainty that a controlled system will face can be restrictive. On the other hand, predefined structures are often needed to obtain stability guarantees. By taking autopilots of autonomous vehicles as a case study, we present a design where stability guarantees can be obtained without imposing predefined structures on the uncertainty: only general system properties are used, valid for several systems under the Euler-Lagrange framework. Such unstructured uncertainty can be compensated by adaptive laws, easily embeddable in the autopilot architecture to extend the vehicle autonomy.

Keywords: Control theories, Deep learning and machine learning
Abstract: In this work, we propose a data-efficient learning strategy tailored for discrete-time linear zero-sum games, eliminating the need for prior knowledge of system models. We first demonstrate that model-based policy iteration is equivalent to Newton’s method for solving the Game Algebraic Riccati Equation (GARE), ensuring quadratic convergence. To address unknown dynamics, unlike existing least-squares Q-learning methods that typically require extensive samples, we utilize Willems’ Fundamental Lemma to transform this Newton process into a fully datadriven form. Theoretical analysis confirms that the proposed algorithm inherits the quadratic convergence rate of Newton’s method. Simulations on an F-16 autopilot model verify its rapid and accurate convergence to the Nash equilibrium using limited observed data.

Keywords: Robotics and swarm intelligence, Computational intelligence, Industrial applications
Abstract: Robotic mobile fulfillment systems (RMFS) employ mobile robots to deliver movable racks to workstations for order picking. Their efficiency depends on coupled decisions on order allocation, order sequencing, and rack service scheduling, yet joint optimization is computationally expensive at scale. This paper proposes a two-stage heuristic framework to minimize total rack-to-workstation travel distance. During stage I, orders are allocated by SKU-set similarity and candidate racks are filtered to shrink the search space. Then during stage II workstation order sequences are improved via simulated annealing, beam search and lightweight post-optimization. Experiments on 30 synthetic instances of varying scales demonstrate that the proposed method reduces travel distance by 36.2% and runtime by 79.3% on average compared with a state-of-the-art MIP–heuristic baseline, validating strong solution quality and scalability.

Keywords: Control theories, Nonlinear control and applications
Abstract: This paper presents two recent results on optimal control with state constraints for fractional differential equations. The first problem considers the finite-dimensional setup, where the state dynamics is the mathbb{R}^n-valued (left) Caputo-type fractional differential equation with order alpha in (0,1). The objective functional is formulated by the left Riemann-Liouville fractional integral. In addition, there are terminal and running state constraints; while the former is described by initial and final states within a convex set, the latter is given by an explicit instantaneous inequality state constraint. The second problem is formulated in the infinite-dimensional setup, where the state equation is modeled by the mathsf{X}-valued left Caputo fractional evolution equation with the analytic semigroup, where mathsf{X} is a Banach space. The objective functional is given by the Bolza form, expressed in terms of the left Riemann-Liouville (RL) fractional integral running and initial/terminal costs. The endpoint state constraint is described by initial and terminal state values within convex subsets of mathsf{X}. For both problems, we state the Pontryagin-type maximum principles and provide their important remarks.

Keywords: Robotics and swarm intelligence, Motion and vibration control, Autonomous vehicles
Abstract: Safety is a critical requirement in multi-robot navigation. Significant progress has been made in collision avoidance among robots and with static or moving obstacles, enabling safe operation in many structured and dynamic environments. However, standard dynamic obstacle models often assume fixed geometry and do not capture hazards whose occupied region expands over time. This behavior is common in hazardous environments where fire fronts, gas plumes, or contamination zones spread and continuously shrink the safe region, imposing deadline-like constraints on motion planning. In this paper, we study decentralized multi-robot navigation with growing obstacles modeled as time-varying expanding sets. We formulate reach-avoid navigation under a time-varying admissible set and compute a Hamilton-Jacobi (HJ) reachability value function using a double-obstacle variational inequality to encode goal reaching while avoiding expanding hazards. For inter-robot safety, we incorporate pairwise HJ safety levels and coordinate avoidance responsibility via a mixed-integer assignment mechanism, allowing each robot to switch between nominal goal-seeking control and provably safe avoidance actions. We further provide safety guarantees under assumptions for both obstacle avoidance and inter-robot collision avoidance. Simulations of multi-robots demonstrate collision-free goal reaching in expanding-obstacle environments. Project page: https://putraaryawinata.github.io/MRGO/.

Keywords: Control theories
Abstract: Data-driven predictive control (DDPC) enables control without relying on explicit system models, making it suitable for applications where model parameters are uncertain. However, its applicability is constrained because it is theoretically designed under the assumption that the collected data are noise-free. This paper develops a theoretical foundation for DDPC, named TLSPC, to address these challenges. Reduced Hankel constraints are introduced to relax perturbation-related components and reduce decision variables, ensuring consistent control outcomes. System states are estimated using a total least squares approach, maintaining robustness despite data corruption and measurement noise. TLSPC leverages offline computations to enable fast online control without requiring parameter tuning. Benchmark examples demonstrate the effectiveness of the proposed method, highlighting potential for practical applications.

Keywords: Control theories, Cyber-physical systems and security
Abstract: This paper addresses the problem of achieving consensus in Multi-Agent Systems (MASs) in the presence of Denial-of-Service (DoS) attacks by developing a resilient control strategy. With the growing complexity and frequency of cyber-attacks, maintaining reliable coordination among agents has become increasingly important. To tackle this issue, two types of controllers are proposed: a distributed state-feedback controller and an observer-based controller. The proposed method allows the controller gains to be adaptively adjusted based on the real-time conditions of each communication channel affected by DoS attacks. The effectiveness and robustness of the proposed approach are validated through a numerical simulation example.

Keywords: Control theories, Industrial applications, Control devices, sensors and actuators
Abstract: This paper presents a compact discrete-time disturbance observer framework equipped with a robust optimal generalized sampler. The robust optimal generalized sampler constructs a new discrete-time output by combining multiple sensor samples within each control period. The sampler gain vector is obtained by solving an optimization problem subject to strict-positive-realness-based linear matrix inequality constraints and additional linear matrix inequality conditions that jointly capture bounded model uncertainty and the sampler gains, thereby ensuring that the induced numerator polynomial remains Schur stable for all admissible uncertainties. This robust zero relocation restores the minimum-phase property required by inverse-model disturbance observers, enabling disturbance estimation and compensation in sampled-data systems.

Keywords: Control theories, Control devices, sensors and actuators, Nonlinear control and applications
Abstract: This paper presents a risk-aware Model Predictive Control (MPC) framework that integrates Equivalent Input Disturbance (EID) with Conditional Value-at-Risk (CVaR) constraints. By employing an EID estimator to capture external disturbances and unmodeled dynamics in real-time and aggregating them into an equivalent input signal, the proposed method provides precise statistical properties for risk assessment. By directly embedding EID-based risk constraints into the MPC optimization objective function, the controller proactively identifies and avoids high-risk scenarios during receding horizon optimization, facilitating optimization within a residual space where risk levels are significantly attenuated. Simulation results demonstrate that this strategy not only effectively enhances tracking performance under strong noise interference but also significantly reduces the system's CVaR compared to baseline methods.

Keywords: Deep learning and machine learning, Industrial applications, Artificial intelligence
Abstract: Fence intrusion (e.g., climbing or crossing over a fence/railing) poses a critical safety risk in surveillance scenarios, yet reliable warning remains challenging due to rare positive events, occlusions, and viewpoint-dependent appearance variations. Frame-wise detectors or action classifiers often yield high false alarms and limited generalization, while running pose estimation continuously is computationally expensive for realtime streams. We present an explainable event-level warning pipeline that combines person detection and multi-object tracking with geometry-constrained candidate gating and pose evidence verification. Specifically, we define a fence line and a near-fence band to select candidates, and trigger intrusion events when reliable knee keypoints cross above the fence line under a multiframe stability scheme. Experiments on a real-world surveillancestyle dataset containing 24.0 hours, 6 scenes, and 72 intrusion events show that our method improves event-F1 by 5.0 points and reduces false alarms per hour by 61% compared with competitive baselines, while running in real time at 25 FPS on 1080p streams. The proposed approach is lightweight, interpretable, and practical for deployment in streaming surveillance systems.

Keywords: Industrial applications, Artificial intelligence, Robotics and swarm intelligence
Abstract: To address safe and wall-adaptive path planning for vehicle-mounted cleaning manipulators in coal mine roadways, this paper proposes a planning framework using point cloud data. A local two-dimensional (2D) grid map is constructed from oadway point cloud data. Based on this representation, a wall distance-aware A-star Algorithm (A*) search is developed by ntroducing a penalty term that enforces deviation from the desired wall distance, and a K-dimensional-tree-based (KD-tree-based) collision checking mechanism is integrated to ensure feasibility in cluttered environments. After global path generation, a twostage refinement strategy is applied, including adaptive window smoothing under turning angle constraints and a wall-adaptive adsorption method based on Gaussian-weighted local surface fitting. Simulation results on flat-roof and dome-shaped roadway models show that the proposed method outperforms A-star (A*) search, rapidly-exploring random trees (RRT), and rapidlyexploring random tree star (RRT*) in reducing wall-distance variance and improving cleaning coverage while maintaining safe distances, and real-world experiments further confirm its cleaning effectiveness.

Keywords: Brain-computer interfaces, Artificial intelligence, Deep learning and machine learning
Abstract: Deep learning has substantially improved surface electromyography (sEMG)-based gesture recognition. Yet in high-stakes settings such as prosthetic control and rehabilitation interaction, developers care not only about predictive accuracy but also about which sensor channels and spectral components a model actually relies on. Existing local explanations mostly operate at the level of time points or temporal windows, making it difficult to align them with the signal-processing representations commonly used for multichannel sEMG. To address this gap, we propose CB-Shapley (Channel--Band Shapley), a channel--band Shapley-based post hoc attribution method for deep sEMG gesture classifiers. CB-Shapley reparameterizes the raw time-domain input into structured channel--band explanatory units and estimates the marginal contribution of each band to the target-class score through frequency-domain masking, inverse transformation, and Monte Carlo sampling. Compared with point-wise saliency maps in the time domain, CB-Shapley yields explanations that better match spectral-analysis practice and facilitate inspection of band dependence and channel-usage patterns. On synthetic tasks with known discriminative frequencies, CB-Shapley successfully recovers the target frequencies and remains stable under severe noise.

Keywords: Energy Systems, Intelligent control
Abstract: The mismatch between capacities of various devices and actual energy demands poses challenges for the coordinated operation of integrated energy systems (IESs). To address this issue, a multi-agent trading strategy for IESs considering capacity configuration is proposed. A two-stage bi-level collaborative planning and operation optimization method is established. In the first stage, a capacity configuration planning model is established to maximize the annual profit of the integrated energy system operator (IESO), and the Newton-Raphson-Based Optimizer is designed to determine the optimal capacity configuration for the IESO. In the second stage, a Stackelberg game model with the IESO as the leader and multiple prosumers as the followers is established, and a distributed iterative algorithm based on the bisection method is proposed to determine the optimal energy scheduling strategy for the IESO and the prosumers, thereby achieving the maximization of the IESO's daily operating profit and the minimization of each prosumer's cost. By iteratively alternating between the two stages, the coordinated optimization of system capacity configuration and operation dispatch is ultimately achieved. Comparisons of scenarios demonstrate that the proposed strategy not only improves the rationality of device capacity configuration and the economic benefits of each agent, but also resolves conflicts of interest among multiple agents.

Keywords: Control theories, Nonlinear control and applications
Abstract: PID controllers are widely implemented in sampleddata form, yet sampling may change the stability and regulation properties guaranteed by ideal continuous-time feedback. This paper studies sampled-data PID control for MIMO non-affine nonlinear uncertain systems. Building on the existing admissible continuous-time PID gain set (Zhao and Guo, 2022), we show that, for every admissible PID gain triple, there exists a sampling period upper bound below which global stabilization and regulation are preserved. We also derive an explicit uncertainty dependent threshold beyond which stabilization is impossible for any choice of PID gains. These results reveal both the capability of sampled-data PID control under sufficiently fast sampling and its intrinsic sampling-induced limitation.

Keywords: System identification and modelling
Abstract: In this paper, we propose a consistent estimation method for Box–Jenkins (BJ) models that is applicable under both open-loop and closed-loop data conditions. The proposed sequentially decoupling (SD) estimator {color{black}is constructed from three sequential least squares (LS) estimators:} (i) estimation of a high-order autoregressive model with exogenous inputs (ARX) model; (ii) estimation of the BJ model’s dynamic model via an output-error (OE) model; and (iii) estimation of the noise model of the BJ model using another OE model. We establish the consistency of the SD estimator under standard regularity conditions.

Keywords: System identification and modelling
Abstract: In many large-scale learning and signal processing applications, observations are generated sequentially by complex evolving systems, where both the underlying data distribution and the model parameters may change over time. Such nonstationary environments invalidate the classical i.i.d. assumption and make repeated retraining impractical, thereby calling for adaptive methods that can update in real time as new samples arrive. Ideally, these methods should operate in a single-pass manner and maintain computational and storage costs that do not grow with the length of the data stream. In this work, we study the Momentum Least Mean Squares (MLMS) algorithm from the perspective of adaptive identification, motivated by its low computational cost and suitability for online implementation. For time-varying stochastic linear systems, we establish theoretical guarantees for MLMS, including tracking-error and regret bounds under general practical conditions. In contrast to the classical LMS recursion, whose analysis relies on a first-order random vector difference equation, the incorporation of momentum creates an additional state variable and results in a second-order time-varying random dynamical system. This substantially complicates the stability analysis, since it requires handling more intricate products of random matrices.

Keywords: System identification and modelling
Abstract: Quantized data arise naturally in many practical problems spanning engineering systems and social-science applications. At the same time, learning methods built on the ell_1 loss are widely appreciated for their stronger robustness to outliers relative to ell_2-based approaches. Despite these advantages, there has been limited work on adaptive identification methods that simultaneously handle quantized observations and exploit ell_1-based optimization. To address this issue, we propose a new adaptive identification scheme tailored to quantized observation models under an ell_1-loss framework. From a theoretical perspective, we prove that the proposed method achieves global convergence of the parameter estimates to the true parameter values without requiring the persistent excitation assumptions, and we further show that its average regret converges to zero as the number of samples grows. To illustrate its practical value, we apply the method to a real-world judicial sentencing dataset, where it demonstrates improved predictive performance and clear practical relevance.

Keywords: Control theories
Abstract: Despite the widespread use of PID controllers in engineering practice, designing optimal PID parameters remains a challenging problem in both theory and practice, particularly for uncertain nonlinear dynamical systems. Building on the authors' recently established PID control theory for MIMO nonlinear uncertain systems (Zhao and Guo, 2022), which provides a concrete PID parameter set guaranteeing global stability of PID-controlled systems, this paper proposes a near-optimal PID tuning method using only input-output, or zeroth-order, performance data.

The method formulates PID tuning as a constrained optimization problem and solves it through an iterative learning algorithm that combines hysteretic random search with the Kiefer--Wolfowitz method (HRS-KW). This combination is used to take advantage of both global exploration and local gradient acceleration. The proposed approach requires no precise structural knowledge of the system dynamics, while theoretically ensuring closed-loop stability and almost sure convergence to an (epsilon)-optimal PID parameter set.

Keywords: Nonlinear control and applications, Mechatronics, Motion and vibration control
Abstract: With the increasing reliance on precise autonomous navigation in maritime transport, surveillance, environmental monitoring, and offshore inspection—including floating renewable energy systems—vulnerabilities associated with GNSS-denied environments, signal degradation, spoofing, and jamming pose significant operational risks. These challenges renew interest in alternative navigation methods such as celestial navigation. This paper presents the design and simulation of an inertial stabilized platform (ISP) for a ship-borne telescope used in autonomous celestial navigation. The primary challenge is maintaining line-of-sight (LOS) stability under nonlinear ship dynamics and stochastic sea disturbances. A nonlinear sliding mode control (SMC) strategy is proposed. The system model incorporates friction, gravitational imbalance, and ship-induced disturbances. Simulation results show that SMC achieves improved disturbance rejection compared to a proportional-integral-derivative (PID) controller. The SMC delivers significantly smaller steady-state errors and reduced overshoot. For celestial navigation, where long-term pointing stability is critical, SMC offers a more robust and reliable control solution.

Keywords: Mechatronics, Nonlinear control and applications, Industrial applications
Abstract: Gantry crane systems have evolved into advanced industrial platforms supporting applications in container terminals, shipyards, heavy manufacturing, and offshore energy infrastructure. Automating the positioning process enables simultaneous optimization of travel time and load stability. The crane dynamics are derived using the Euler–Lagrange formulation and exhibit strongly coupled nonlinear behavior, making control design challenging. This paper proposes a nonlinear control strategy based on Partial Feedback Linearization (PFL), where a composite Proportional–Derivative (PD) control law is formulated within the virtual input framework. The control law incorporates position and sway errors and their derivatives to ensure accurate tracking of the virtual output. Zero-Vibration-Derivative (ZVD) input shaping is employed to generate a desired virtual trajectory with minimal residual sway. Controller parameters are optimized using Particle Swarm Optimization (PSO) with a multi-objective cost function and compared with a baseline PID scheme. Simulation results show that while both controllers achieve comparable position tracking, the proposed PD–PFL strategy significantly improves sway suppression and reduces control effort. This improvement stems from the PFL framework’s ability to mitigate nonlinear coupling effects and enhance overall trajectory performance.

Keywords: Energy Systems, Artificial intelligence, Industrial applications
Abstract: Normal Behavior Models (NBMs) are a mainstream approach to wind turbine condition monitoring, establishing a baseline of sensor signals under normal operating conditions and using residuals to detect faults. However, existing research commonly neglects baseline drift caused by longterm equipment operation, referred to here as aging drift. This paper uses five consecutive years of Supervisory Control and Data Acquisition (SCADA) data from four turbines at a European onshore wind farm to systematically quantify aging drift in gearbox bearing temperature. Alarm logs are processed through a four-layer filtering pipeline combined with a 72-hour deduplication rule, yielding 90 independent fault events as evaluation targets. Experimental results show: (1) the annual residual mean of a static NBM increases monotonically over time, with significant inter-turbine variation in aging rate; (2) this drift causes static NBMs to generate substantial false alarms in year five, with an average false alarm rate (FAR) of 27.2%; (3) the proposed Peer-to-Peer (P2P) method, which uses inter-turbine temperature comparison to substantially reduce sensitivity to aging drift, reduces the average FAR to 2.9% (an 89% reduction) while maintaining an average early-warning lead time of 51 hours for detected faults. This study reveals the significant impact of aging drift on long-term NBM performance and highlights the importance of rigorous alarm filtering and deduplication for trustworthy evaluation.

Keywords: Nonlinear control and applications, Control theories, System identification and modelling
Abstract: Climate change is significantly influenced by the increasing concentration of carbon dioxide (CO_2) in the atmosphere, which directly affects atmospheric temperature and ecological systems. The dynamic interactions between CO_2 concentration, photosynthetic biomass, and atmospheric temperature are important to study to understand the interrelationships between these three factors and determine effective mitigation strategies. In this study, a compartmental climate change model is extended into a fractional-order framework to incorporate memory effects, allowing for a more realistic representation of long-term environmental dynamics. Furthermore, an optimal control problem is formulated to identify effective intervention strategies aimed at reducing CO_2 concentration and temperature. The control variables used in this study are restrictions on private vehicle use and reforestation efforts. Numerical simulations are performed for various fractional orders to analyze the impact of memory effects on system behavior and control performance. The results show that the fractional-order model exhibits better dynamic behavior than the classical integer-order model, particularly in terms of faster convergence and reduced oscillations for lower fractional orders. In addition, the presence of memory effects on the implementation of control strategies influences the intensity and timing of control interventions.

Keywords: System identification and modelling, Health systems, Control theories
Abstract: In the last decade, drug abuse has increased significantly among youth population at the global level, so it has become an important problem and has an impact on public health and socioeconomic stability. In this study, a mathematical model of the spread of drug users in the youth and adult age groups is developed as an effort to investigate the influence of age factors on the dynamics of drug user behavior. Based on analysis, the conditions for stability and the existence of endemic equilibrium for the spread of drug users depend on the basic reproduction number. This study also applies Pontryagin’s Maximum Principle to optimize control variables including anti-drug campaign and enhanced rehabilitation, are utilized forward-backward sweep simulations and cost-effectiveness analysis to determine the most efficient strategy.

Keywords: Control theories, Autonomous vehicles, Adaptive systems
Abstract: The reliability of unmanned surface vehicle (USV) actuators is a critical factor in ensuring safe and effective maritime operations. Actuators are susceptible to performance degradation that can escalate into faults, increasing the risk of failure and system instability. This paper proposes a fault-tolerant control framework that integrates Model Predictive Control (MPC) with an Adaptive Kalman Filter (AKF) to address actuator faults in USVs subject to stochastic dynamics. The USV is modeled by a two-degree-of-freedom linear stochastic system affected by both process and measurement noise, and the presence of noise renders the MPC problem stochastic with probabilistic state constraints. The AKF jointly estimates the system states and fault magnitudes, providing reliable predictions for the MPC stage, while the probabilistic state constraints are reformulated into deterministic inequalities using the statistical distribution of the state variables. Simulation results across three scenarios of increasing fault severity demonstrate that the proposed MPC--AKF method tracks the reference heading angle accurately while satisfying all imposed constraints, confirming its effectiveness in maintaining USV operation under actuator faults.

Keywords: Intelligent control, Artificial intelligence, Industrial applications
Abstract: Multi-modular floating structures (MMFS) are a promising solution for offshore applications, but connector loads and relative motion remain critical challenges. This paper proposes an agentic large language model (LLM)-based controller for real-time regulation of connector stiffness, operating directly on observed system states without task-specific training. The approach is evaluated on a 1x4 MMFS under a multiple regular wave excitation loads in sequence and compared against an "ideal" feed-forward rule-based controller derived from offline brute force optimisation. While the rule-based strategy achieves near-optimal performance within steady-state conditions, its lack of adaptability limits performance outside these regimes. In contrast, the LLM-based controller with the feedback loop and memory could maintain safe displacement levels and achieve comparable load performance by adapting to the observed system behaviour. These results highlight the potential of LLM-based control as a flexible and scalable approach for MMFS under uncertain offshore conditions.

Keywords: Control theories
Abstract: This paper addresses the small packet problem in discrete-time industrial control systems (DICSs) 4.0, where the small packets are modeled as an impulsive effect based on Lyapunov-like theory to promote system stability. Considering the common yet under explored issue of small packet data under TCP/IP protocol, this study first analyzes the characteristics of small packet data occurrence. To further address the challenges associated with small packet transmission, a dynamic acknowledgment (ACK) bundling mechanism is proposed, which can monitor the small packet data in real time to ensure the integrity and reliability of the communication process. Additionally, an intelligent controller designed for impulsive effects is developed, which can effectively avoid the system risks caused by the transmission of multiple packets of data. Finally, the proposed method is validated by a flight connected to an industrial network. The results show that small packets greatly enhance system stability and capacity of data transmission, improving network utilization.

Keywords: Control theories, Autonomous vehicles, Robotics and swarm intelligence
Abstract: Navigating a collision-free and optimal trajectory for a robot is a challenging task, particularly in environments with moving obstacles such as humans. We formulate this problem as a stochastic optimal control problem. Since solving the full problem is computationally demanding, we introduce a tractable approximation whose Bellman equation can be solved efficiently. The resulting value function is then incorporated as a terminal penalty in an online rollout framework. We construct a trade-off curve between safety and performance to identify an appropriate weighting between them, and compare the performance with other methods. Simulation results show that the proposed rollout approach can be tuned to reach the target in nearly the same expected time as receding horizon A^star while maintaining a larger expected minimum distance to the moving obstacle. The results also show that the proposed method outperforms the considered CBF-based methods when a larger obstacle clearance is desired, while achieving comparable performance otherwise.

Keywords: Industrial applications, Intelligent control, Artificial intelligence
Abstract: This paper evaluates the Action Chunking with Transformers (ACT) model for contact-rich industrial surface cleaning tasks using the LeRobot imitation learning platform. A single-arm leader–follower teleoperation system was used to collect 50 demonstrations of systematic sweeping motions on a stamped metal panel contaminated with coffee grounds, serving as a controlled proxy for industrial particulate contamination. The ACT model successfully learned the geometric structure and sequential ordering of multi-step sweep primitives without large-scale pre-training. However, policy execution was 2–3× slower than human teleoperation due to repetitive segment re-execution and inconsistent contact height regulation—limitations attributable to the absence of explicit force feedback. A systematic batch size analysis (8, 16, and 32) revealed a counterintuitive efficiency trade-off: batch size 8 achieved the highest task success rate and the lowest total wall-clock training time (sim9 hours) while remaining within 10 GB GPU memory constraints, making it compatible with consumer-grade hardware. Batch size 32, despite faster per-step convergence, exhibited sequence-skipping artifacts during deployment. These results demonstrate that lightweight VLA models are viable for semi-structured industrial cleaning, with smaller batch sizes offering superior cost-effectiveness over aggressive batch scaling strategies.

Keywords: Industrial applications, Measurement and instrumentation, Mechatronics
Abstract: Industrial aluminum-foil slitting is studied under coupled standard-length intervals, quota limits, bilateral defects, rework operations, and economic objectives. A defect-aware framework is presented to jointly improve waste and economic return. Backward feasible planning, recursive quota-compliant search, rework-aware reuse, neighborhood recut, waste-to-product redistribution, local dynamic programming, tail-end backtracking, non-degradation rollback, and final defect sanitization are integrated. Scrap, downgrade, rework, and packaging costs are evaluated by a unified economic model. Experiments on 30 production-style cases are conducted against three reproducible baselines and four ablation variants, showing large waste reduction, positive net-profit gain, and zero final product-defect violations.

Keywords: Intelligent control, Control theories
Abstract: This paper investigates the fixed-time bipartite consensus problem for a class of nonlinear multi-agent systems subject to input delays. To reduce communication burden and alleviate chattering, an event-triggered dynamic soft variable structure control (DSVSC) scheme is proposed. By introducing auxiliary variables, the influence of input delays is compensated for the controller design. Using Lyapunov stability theory, sufficient conditions are derived to ensure fixed-time bipartite consensus, and an explicit upper bound on the settling time is obtained. In addition, the proposed event-triggered mechanism is shown to exclude Zeno behavior. Simulation results verify the effectiveness of the proposed scheme and its superiority in terms of control smoothness.

Keywords: Intelligent control, Control theories, Computational intelligence
Abstract: This paper proposes a historical information-dependent approximate optimal control (HIDAOC) scheme for fractional-order nonlinear systems (FONSs). Unlike conventional methods that implicitly rely on Markovian assumptions, the proposed method explicitly incorporates the full historical state information into the cost function, which preserves the intrinsic memory characteristics of FONSs. First, a historical information-dependent fractional Hamilton-Jacobi-Bellman (FHJB) equation is rigorously formulated based on the definition of coinvariant differentiability. Then, a historical information-dependent policy iteration (HIDPI) algorithm is developed to iteratively solve the analytically intractable FHJB equation. To effectively capture the nonlocal dynamics, a novel critic neural network is constructed to approximate the optimal cost function, utilizing fractional-order integral polynomials of the system states as activation functions, rather than time-invariant activation functions. Theoretical analysis guarantees the stability of the closed-loop system under the derived fractional-order approximate optimal control policy. Finally, simulation results on a practical example demonstrate the validity of the proposed control scheme.

Keywords: Intelligent control, Robotics and swarm intelligence, Medical and financial systems
Abstract: This paper presents an empirical analysis of reward formulation effects in Proximal Policy Optimization (PPO) for a surgical needle-pick task using the Surgical Robot Learning (SurRoL) simulation framework. PPO is implemented in a bare-metal configuration, without demonstrations or auxiliary learning mechanisms, to isolate the impact of reward structure on learning dynamics in an actor–critic setting.

Three reward designs with increasing structural complexity—Sparse, Less-sparse, and Staged—are evaluated under identical training conditions. Learning behavior is analyzed using internal PPO optimization metrics recorded from TensorBoard, including policy gradient loss, value function loss, policy distribution standard deviation, and clipping frequency, as well as task-level evaluation metrics such as contact consistency and distance to goal.

The results show that increasing reward complexity does not necessarily improve learning outcomes. While Sparse rewards preserve task-relevant interaction patterns despite unstable optimization, Less-sparse rewards achieve more balanced and structured learning dynamics under stationary shaping. In contrast, Staged rewards introduce non-stationary learning signals that disrupt actor–critic coordination, resulting in misleading training indicators and poor task-level generalization.

Keywords: Mechatronics, Robotics and swarm intelligence, System identification and modelling
Abstract: A compact six-degrees-of-freedom (6 DoF) remotely operated unmanned underwater vehicle (ROV) is designed, developed and interfaced with a personal computer for real-time control and data acquisition. The system incorporates over the air (OTA) firmware update functionality to enable efficient deployment and maintenance. The ROV, named Marjan RI, operates at speeds of up to 2 knots and is rated for depths of 80-100 m. It features an open frame architecture integrating eight thrusters, an onboard electronic control unit (ECU), power management, communication protocols and an umbilical tether providing bidirectional data exchange with a ground station. This paper also presents the nonlinear equations of motion for the underactuated, neutrally buoyant open frame underwater vehicle. Experimental step responses results from preliminary pool tests are also reported. That is to assess the vehicle’s dynamic behaviour and ensure readiness of the vehicle for data driven model development through system identification techniques.

Keywords: Motion and vibration control, Nonlinear control and applications, Artificial intelligence
Abstract: Trajectory tracking of pneumatic actuator systems is challenged by inherent nonlinear dynamics that limit the effectiveness of conventional linear controllers. Nonlinear hyperbolic PID (NH‑PID) controllers have demonstrated improved tracking capability, particularly when embedded with fuzzy logic tuning mechanisms. However, classical multi‑input fuzzy logic controller (FLC) structures introduce increased computational complexity, which may hinder real‑time implementation on embedded platforms. This paper presents a comparative evaluation of NH‑PID control using two fuzzy embedding strategies: classical multi‑input fuzzy logic and single‑input fuzzy logic control (SIFLC). Performance comparison is conducted based on IAE and ISE indices, steady-state and transient responses, simulation execution time, control loop processing time, CPU utilization, and processor temperature. Simulation results indicate that while FLC embedding provides marginal improvements in certain transient responses, the SIFLC‑based NH‑PID significantly reduces computational demand—by up to 80.3% in simulation time and 17% in CPU utilization—along with a 9.6% reduction in processor temperature, while maintaining comparable tracking accuracy. These findings highlight a critical performance–complexity trade‑off in nonlinear fuzzy‑PID controllers and demonstrate the suitability of SIFLC embedding for real‑time trajectory tracking applications.

Keywords: Nonlinear control and applications, Robotics and swarm intelligence, Control theories
Abstract: This paper studies the resilience of prescribed performance-based matrix-weighted multi-cluster consensus in the presence of Denial-of-Service (DoS) attacks. A network of first-order agents interacting over a matrix-weighted graph is considered. An adversary capable of breaking a limited number of edges at each time instant and thereby disrupting the formation of matrix-weighted multi-cluster consensus is modeled. The attack is formulated as an optimal control problem, and the optimal attack strategy for the adversary is determined. A prescribed performance control law is then designed to guarantee that the relative position errors remain within predefined bounds despite adversarial disruptions. Numerical simulations demonstrate that the proposed control law preserves the prescribed performance constraints even under optimal DoS attacks.

Keywords: Nonlinear control and applications, Robotics and swarm intelligence, Mechatronics
Abstract: Haptic interfaces enable bidirectional transmission of motion and force information between human operators and robotic systems. This study presents the design and experimental evaluation of a bilateral haptic interface based on brushless DC (BLDC) motors controlled using the SimpleFOC algorithm on an ESP32 microcontroller. Each haptic module integrates a BLDC actuator, an AS5600 magnetic encoder, and a load cell, while digital dial indicators are used for finite-stroke displacement measurement. The system consists of two haptic units configured as producer and consumer, with motion input applied by a Dobot MG400 robot. Three experimental modes were conducted: free motion, stopper interaction with a 1 mm clearance, and full-rotation tracking over approximately 360°. The results show that the system maintains temporal synchronization and reproduces constrained interaction under stopper contact. However, large displacement mismatch remains in the free-motion and stopper tests, while the full-rotation test demonstrates much better angular tracking. These results confirm the feasibility of a low-cost bilateral haptic platform based on open-source motor-control frameworks and embedded hardware for teleoperation-oriented applications.

Keywords: Robotics and swarm intelligence, Control theories, Mechatronics
Abstract: The paper presents modelling and robust yaw control of a mini-class Remotely Operated Vehicle (ROV). The underactuated unmanned underwater vehicle employs three thrusters, enabling planar yaw and heave motion. A yaw dynamics model is derived using simplified Newtonian dynamics and refined through system identification approach. To address modelling uncertainties and environmental disturbances a robust control strategy is developed using Quantitative Feedback Theory (QFT). Real-time implementation of the plant in a Model-in-the-Loop (MIL) framework, together with the shaped controller on the Opal-RT real-time platform demonstrates the practical feasibility and robustness of the proposed control strategy.

Keywords: ROV, Quantitative Feedback Theory, robust control, Model-in-the-Loop simulation.

Keywords: Intelligent control, Artificial intelligence, Industrial applications
Abstract: Process simulators such as Aspen Plus are widely used in chemical process design, yet their reliance on platform-specific Component Object Model (COM) automation makes them difficult to integrate with cloud-hosted Large Language Model (LLM) agents. This paper presents an engineering integration wrapping Aspen Plus behind a Representational State Transfer (REST) abstraction layer so external LLM agents can drive simulations without local simulator or Python COM client installation. A FastAPI server exposes over 75 endpoints across 15 categories—session management, flowsheet discovery, stream and block property manipulation, and generic tree access—combined with ngrok tunneling for Hypertext Transfer Protocol Secure (HTTPS) access. We adopt the STR_MAIN subtree as a uniform path convention for calculated stream outputs and mitigate state corruption under concurrent COM calls with workflow-level sequential batching. A dynamic two-phase discovery pattern first enumerates simulation-dependent items before parameterized queries, letting the same Application Programming Interface (API) drive any Aspen Plus file unchanged. Reference n8n workflows demonstrate multi-step agentic operations with proper request sequencing. A methanol-water distillation case study confirms end-to-end integration with 89.97% methanol recovery and exact mass-balance closure. The framework externalizes Aspen Plus over HTTPS and documents solutions to two COM-specific failure modes.

Keywords: Robotics and swarm intelligence, Mechatronics, Autonomous vehicles
Abstract: This paper studies fixed-time distance-based formation control laws for multi-agent systems subject to disturbances. Existing distance-based methods mostly achieve only asymptotic stability or finite-time stability, leaving settling times either unspecified or dependent on initial conditions. Moreover, while some robust controllers handle bounded disturbances, none provide fixed-time convergence for either single- or double-integrator agents. To address this gap, we propose robust fixed-time distance-based formation control laws for single- and doubleintegrator agents. Both proposed controllers operate without inter-agent communication and reject time-varying disturbances. Our stability analysis establishes explicit upper bounds on the settling time that are independent of initial conditions. Simulation results are also provided to demonstrate the effectiveness of the control laws under disturbances.

Keywords: System identification and modelling
Abstract: This paper addresses the identification problem of Markov jump autoregressive systems with exogenous inputs involving an unknown transition probability matrix (TPM). A variational Bayesian inference framework is proposed to solve this problem. Firstly, unlike existing point estimation methods, a Dirichlet distribution is introduced as the conjugate prior of the TPM within a hierarchical probabilistic model, which naturally satisfies the stochastic constraints of the TPM and serves as a regularizer, thereby avoiding overfitting under short sample lengths or infrequent switching conditions. Then, based on mean-field theory, closed-form variational updates are derived for all unknown variables with a forward–backward scheme for efficient computation, enabling uncertainty quantification. Finally, simulation results demonstrate that the proposed variational Bayesian identification algorithm exhibits higher accuracy than the point estimation method of the TPM.

Keywords: System identification and modelling, Industrial applications
Abstract: Coal-water slurry gasification is a widely adopted gasification process that achieves coal conversion and chilled water circulation through a gasifier and multi-stage flash evaporation units. The coupling between the upstream and downstream units, the complex chemical reaction mechanisms and unknown parameters caused by the missing sensors, pose significant modeling challenges, thereby limiting the application of advanced control. Existing studies have achieved modeling of isolated gasifiers or flash unit based on experimental data, yet they overlook the interconnections among these units in actual production. To enhance the industrial relevance of the model and facilitate the design of advanced control strategy, this paper decomposes the coal gasification system into three subsystems and proposes a general hybrid mechanistic and data-driven modeling framework for the key variables in the gasifier and flash evaporation units. The accuracy and practical value of the proposed model are validated based on actual industrial production data.

Keywords: Control theories, Cyber-physical systems and security, Robotics and swarm intelligence
Abstract: This paper studies the robust observer-based output-feedback consensus for heterogeneous linear multi-agent systems over a directed graph in the presence of bounded disturbances. Each follower is described by distinct linear dynamics and uses only output measurements. A two-layer dynamic protocol is developed by combining a local state observer with a distributed output-based leader observer. Using the regulator equations, the closed-loop dynamics is decomposed into subsystems of local estimation, leader-estimation, and regulation error. This structure leads to three compact convex conditions gathered in a single main theorem. A three-step Lyapunov analysis establishes boundedness of all closed-loop signals and practical output consensus under bounded disturbances, while exponential output consensus is recovered in the disturbance-free case. The overall approach ensures exponential input-to-state stability (ISS) to attain fast convergence and robustness. A numerical example with five heterogeneous followers illustrates the effectiveness of the proposed design.

Keywords: Industrial applications, Autonomous vehicles, Nonlinear control and applications
Abstract: This paper presents manual and autonomous control systems for a two-wheeled drone moving along a wall. The drone's movement along the wall is controlled by rolling and yawing, while pitching is used to regulate the contact force against the wall. The attitude control system employs two independent controllers, separating the management of roll/yaw dynamics from pitch control. In manual control, a human operator moves the drone along the wall. In autonomous control, the drone is automatically guided to reach a target position on the wall. The effectiveness of the proposed control system is verified through numerical simulations under both manual and autonomous control.

Keywords: Nonlinear control and applications
Abstract: The leader-following consensus problem for multiple uncertain Euler-Lagrange (EL) systems has been extensively studied. One of the common approaches to dealing with this problem is the robust control approach, which has the advantage that every local control law is static. However, the validity of the existing robust control approaches relies on some restrictive assumptions on the communication networks and the dynamics of the EL systems. In this paper, we propose a robust control approach that is able to solve the leader-following consensus problem for multiple uncertain Euler-Lagrange systems over jointly connected switching networks by a combination of a static local control law and an adaptive distributed observer assuming that the uncertain parameters are contained in a bounded region with a known bound and the disturbances are bounded with known bounds.

Keywords: Nonlinear control and applications, Adaptive systems, System identification and modelling
Abstract: Uncertainties are unavoidable in almost all physical control systems, and maintaining the reliability of the control design under uncertainties is a fundamental challenge. Among these uncertainties, the mismatch between the actual and nominal control input gain exists in various environments, such as aircraft control systems with actuator degradation. However, the control performance using traditional control designs under nominal models will be compromised if these parametric uncertainties are too significant. In this paper, a reliable active disturbance rejection control (ADRC) design with a consistent Kalman Filter (CKF) is proposed for a class of discrete-time, uncertain systems with unknown control input gain. In detail, the control reliability is guaranteed by integrating the sufficient stability criterion via ADRC with online identification via CKF. Moreover, the effectiveness of the proposed CKF-ADRC design is demonstrated in a simulation of a typical aircraft model.

Keywords: Nonlinear control and applications, Autonomous vehicles, Control theories
Abstract: This paper proposes a control framework for obstacle avoidance in aircraft subject to non-convex input constraints. Unlike typical mobile robots, aircraft must maintain a minimum speed to prevent stalling and are limited by centripetal acceleration during turns. These physical restrictions create a non-convex admissible region in the control input space, which makes applying standard control barrier function (CBF) methods based on convex quadratic programming challenging. Furthermore, CBF approaches using the L2 distance metric often fail to initiate turning maneuvers against frontal obstacles, resulting in dangerous deceleration commands that conflict with the minimum speed constraint. To address these issues, we introduce a CBF based on the weighted L1 norm to facilitate early turning decisions. Next, we present an analytical approach to precisely determine the feasible input region formed by the intersection of the CBF-based safety constraint and the non-convex input constraint. Finally, we determine an optimal safety input that respects the nominal tracking controller via weighted L_1 norm minimization within the identified region. A numerical simulation shows that the proposed method ensures safety and enables smooth avoidance maneuvers, including turning, even when facing frontal obstacles.

Keywords: Nonlinear control and applications, Control devices, sensors and actuators, Autonomous vehicles
Abstract: This paper proposes a planar localization method for navigation of autonomous vehicles using a LiDAR-based approach, with the objective of achieving both lightweight computation and robust performance. The localization methodology is based on the unscented Kalman filter (UKF), enhanced by a nonlinear observer (NLO) used as the nonlinear transformation in the unscented transform process to improve accuracy and stability. The method incorporates a range-flow constraint for odometry estimation and a scan matching algorithm for pose updates. To reduce computational burden, a downsampling strategy is employed for intermittent measurement updates. Validation is performed through the localization of a mobile robot system in a 2D map environment, demonstrating accurate and robust performance with low computational requirements. These results show promise for the proposed LiDAR-only localization framework in real-world applications.

Keywords: Nonlinear control and applications, Control theories
Abstract: Control Barrier Functions (CBFs) provide a methodical way to guarantee safety by enforcing forward invariance of safe sets by solving a real-time constrained quadratic program. However, their performance can be severely affected by unknown disturbances and modelling uncertainties, often leading to safety constraint violations or conservatism. This article proposes a control framework that integrates a finite-time disturbance observer (FTDOB) with CBF-based control to achieve finite-time convergence of the disturbance estimate and accurate disturbance compensation. The proposed FTDOB ensures finite-time convergence of disturbance estimates, which are incorporated into the CBF framework. Rigorous theoretical analysis is provided for the proposed method. Simulation results on a nonlinear benchmark system demonstrate that the proposed method achieves faster disturbance rejection and improved safety margins of the system in finite time.

Keywords: Nonlinear control and applications, Control theories
Abstract: This paper studies the set of effective initial states that ensure system states converge to the equilibrium point with actuator constraints, namely the domain of attraction (DA). A fixed-time controller is designed for the affine nonlinear systems, and the analytical expression of the fixed-time DA is derived. An optimized parameter is then introduced to obtain a more accurate numerical boundary of the DA, which reduces the conservatism of the analytical method. Finally, simulation results validate the validity of the theorems proposed in this paper.

Keywords: Nonlinear control and applications, Control theories, Control devices, sensors and actuators
Abstract: Virtual synchronous generator based inverters have been widely applied in power electronic interfaces due to their regulation and support capabilities. However, the transient angle safety of inverter under fault ride-through conditions remains insufficiently investigated. This paper proposes a safe critical control scheme to enhance transient angle safety. The influence of virtual excitation based reactive power control on transient angle safety is analyzed. Then, a safe superlevel set based on an energy-like function is constructed for fault conditions, and a control barrier function is incorporated into the power control layer to constrain system states within the safe set. As a result, transient angle safety is ensured during and after faults. The effectiveness of the proposed scheme is verified through simulations.

Keywords: Control theories, Intelligent control, Artificial intelligence
Abstract: This paper proposes a data-driven Q-learning framework for H-infinity control of discrete-time linear systems under a static output-feedback (SOF) structure. Existing reinforcement learning (RL)-based SOF H-infinity methods address feasibility for a prescribed disturbance attenuation level, rather than directly optimizing the disturbance attenuation level. The proposed method reformulates the SOF H-infinity control problem as a Q-function-based min--max optimization problem and develops an LMI-based policy iteration algorithm in which the attenuation level is treated as a decision variable. In the proposed framework, the Q-function matrix is estimated directly from measured data through an LMI-based policy evaluation step, and the SOF gain is iteratively updated through a corresponding policy improvement step. Using measured data, the proposed framework updates the control policy without requiring explicit knowledge of the system matrices. Numerical results demonstrate that the proposed method achieves an H-infinity performance level close to the state-feedback optimal value. These results indicate that the proposed approach can effectively reduce the conservatism associated with conventional SOF design methods and provide a practical framework for data-driven SOF H-infinity control.

Keywords: Industrial applications, Adaptive systems, Intelligent control
Abstract: This paper develops a cooperative fault-tolerant control framework for heterogeneous networked linear systems subject to sensor faults and external disturbances. Each unit employs an augmented mathcal{H}_infty observer that jointly estimates the system state and sensor fault, providing disturbance-attenuated estimation guarantees. An inner state-feedback gain is synthesized through convex mathcal{H}_infty LMIs to ensure robust closed-loop stabilization, while an outer distributed integral action enables cooperative tracking of a common setpoint. The resulting cooperative tracking error is shown to satisfy an input-to-state stability (ISS) condition with respect to disturbances and estimation uncertainty, and converges exponentially to zero in the disturbance-free case. Simulations on heterogeneous DC-motor networks with star, cyclic, and path communication topologies demonstrate accurate state and fault estimation, robust tracking performance, and resilience against time-varying sensor faults and disturbances.

Keywords: Intelligent control, Adaptive systems, Deep learning and machine learning
Abstract: Morphing aircraft enhance their adaptability to diverse environments and tasks by adjusting their mechanical structures. Online reinforcement learning (RL) algorithms can generate accurate flight policies without prior knowledge. However, they require interaction with real aircraft, which can be both time-consuming and pose safety risks. This paper proposes a Decision Transformer model that combines morphology-embedding (M-DT), aiming to achieve universal flight control within the morphing space. Specifically, the flight control task is formulated as predicting the current optimal control action based on historical states, actions, and reward-to-go. The flight control sequences of the morphing aircraft are modeled via M-DT. To achieve cross-configuration generalization, the morphing configuration is incorporated as an additional mode, and configuration information is embedded into the state, action, and reward data in the trajectory sequences to regulate the output of the Transformer. The simulation results verify that large-scale joint training enables M-DT to effectively generalize to unseen dynamic characteristics, demonstrating its effectiveness and universality in the general flight control of the morphing aircraft systems.

Keywords: Intelligent control, Industrial applications, Communication
Abstract: In this paper, a novel microgrid (MG) and electric vehicle (EV) collaborative operational framework is proposed to achieve a symbiotic relationship between MG operators and EV users, while accommodating diverse EV participation modes in demand response (DR) services. Firstly, a new concept of EV Virtual Aggregator (EVVA) is introduced, which classifies EVs into distinct aggregators based on response characteristics, thereby facilitating the integration of heterogeneous EV clusters. Secondly, a comprehensive multi-objective optimization model is developed to simultaneously consider economic, environmental impact, MG safety, and EV battery degradation costs. The optimization problem is solved using an enhanced particle swarm optimization algorithm. Thirdly, a hierarchical incentive allocation mechanism is proposed, grounded in the Shapley value method. The first stage implements an initial allocation based on contribution of each EVVA, while the second stage performs a refined allocation considering schedulable capacity of individual EV. Finally, case studies demonstrate that the proposed framework significantly enhances collaborative operation of MG-EV system. The results indicate that the proposed framework not only ensures the rationality of benefit distribution among EV users but also significantly improves the economic efficiency and safe operation level of the MG, achieving a 3.2% reduction in total cost and a 58.4% reduction in safety cost.

Keywords: Intelligent control, Nonlinear control and applications, Adaptive systems
Abstract: This paper proposes a learning-based resilient control framework for nonlinear systems in the presence of denial-of-service (DoS) attacks. By introducing an exponentially increasing factor into the performance index, an optimal control problem with a prescribed convergence rate is formulated. Combining switched systems theory with the adaptive dynamic programming (ADP) method, online learning of a resilient controller is achieved using state and input data from the attacked system. It is further shown that the closed-loop system is uniformly ultimately bounded (UUB) under DoS attacks. Simulation results validate the effectiveness of the proposed framework.

Keywords: Intelligent control, Nonlinear control and applications, Artificial intelligence
Abstract: In this work, we address the problem of predictive interception and jamming of a target using an autonomous agent equipped with a directional jammer and operating under nonlinear stochastic dynamics. We formulate this task as a stochastic optimal control problem that maximizes total jamming time inside a finite rolling planning horizon while penalizing chance-constraint violations for collision avoidance. This results in a highly nonlinear, nonconvex, high-dimensional optimization problem that is challenging to solve with standard numerical solvers. To tackle this challenge, we first reformulate the stochastic optimal control problem as a sampling problem over finite-horizon control sequences, by defining an unnormalized Boltzmann distribution whose high-probability regions correspond to low-energy control plans. We then utilize a variance-exploding denoising diffusion process directly in control space to sample such plans, progressively transporting noisy control sequences toward high-probability regions of the target density. Simulation results show that the proposed method generates robust interception-and-jamming trajectories and outperforms conventional nonlinear optimization solvers.

Keywords: Control theories
Abstract: The duality between observability and reachability allows results and analysis methods for one property to be directly applied to the other, simplifying the design and understanding of control systems. This paper extends the investigation of this duality property from ordinary linear systems to surrogate singular linear switched systems in which each operational mode may be nonsingular or singular. The study focuses on the discrete-time domain. The dual system for such a class of systems is established. The results showed that the dual system cannot be derived via the ``naive'' dual used in ordinary systems. Instead, reversing the switching signal and switching the consistency space are needed, highlighting the implications for system analysis and design in the presence of mode-dependent singularities.

Keywords: Control theories
Abstract: The rapid rise of online loan and pay-later services simplifies consumer financing through installment-based purchases, but pressures providers to maintain sufficient capital for growing loan demands. This research develops a dynamic system that models the interaction between potential capital and potential credit to preserve capital availability amid high demand. Sensitivity analysis evaluates the influence of parameters and the system's stability under varying conditions. An optimal control strategy derived from Pontryagin's minimum principle and solved numerically using the forward–backward sweep method, represents educational campaigns to moderate borrowing behavior. The simulation results show that the proposed control effectively maintains fund availability within safe limits and stabilizes potential credit at a controlled level. Implementing the recommended control yields the lowest cost function to maintain potential capital. It reduces potential credit growth by 10.89% compared to the uncontrolled case, ensuring the long-term sustainability of digital lending services.

Keywords: Nonlinear control and applications, Control theories, Robotics and swarm intelligence
Abstract: Distributed control of multi-agent helicopter systems has gained considerable interest in recent years due to its applicability in executing complex tasks that cannot be achieved by a single agent. This study addresses a distributed adaptive neural network sliding mode control (D-ANNSMC) strategy for synchronized trajectory tracking (STT) of multi-agent non-linear two-degrees-of-freedom (2-DOF) helicopter systems in the presence of uncertainties and disturbances. The communication topology is directed and the system employs radial basis function (RBF) neural networks (NNs) for uncertainty approximation and disturbance estimation. A Lyapunov-based stability framework is established to guarantee the asymptotic stability of the closed-loop system, ensure the boundedness of the NN weight adaptation process, and realize asymptotic convergence of error for STT with respect to the reference. The effectiveness of the D-ANNSMC scheme is verified through simulation studies and real-time experiments on multi-agent Quanser 2-DOF helicopter setups.

Keywords: Nonlinear control and applications, Deep learning and machine learning, Control theories
Abstract: The objective of this research is to incorporate input constraints into reinforcement learning-based optimal control methods and achieve system stability while satisfying these constraints. We propose a control method with norm input constraints for nonlinear systems with unknown value functions and drift dynamics. To achieve this objective, the proposed method utilizes the control input and value function derived from a redesign method based on inverse optimal control. Reinforcement learning is employed to estimate the unknown parameters in the value function and drift dynamics and derive the optimal control input. Furthermore, simulation results confirm that the state remains bounded while satisfying the norm input constraint, verifying the effectiveness of the proposed reinforcement learning-based optimal control method. Future work will investigate whether introducing a fast model estimator can further improve the convergence properties of the state.

Keywords: Nonlinear control and applications, Robotics and swarm intelligence, Adaptive systems
Abstract: This paper studies distributed trajectory tracking for networks of nonholonomic mobile robots under adversarial information exchange. An exact global input--output feedback linearization scheme is developed to regulate planar position outputs, yielding linear error dynamics without prescribing internal state trajectories. To mitigate corrupted neighbor information, a resilient desired-signal construction is proposed that combines local redundancy with trusted in-neighbor signals, without requiring adversary detection or isolation. When sufficient redundancy is available, the method suppresses adversarial influence and recovers nominal tracking performance. If redundancy conditions are violated, adversarial effects enter as bounded disturbances and the tracking error remains ultimately bounded. Simulation results on star, cyclic, and path topologies validate the analysis and demonstrate the superior resilience of cyclic networks due to distributed information propagation.

Keywords: Nonlinear control and applications, Robotics and swarm intelligence, Autonomous vehicles
Abstract: This paper proposes an undesired equilibria-free CLBF safety control via a Circulation CLBF (Circ-CLBF) for multi-robot navigation. CLBF offers a closed-form alternative to quadratic programming-based safety control, but suffers from saddle points where robots may become stuck despite active obstacle avoidance. We integrate a circulation CBF into the CLBF control via a hybrid automaton, detecting undesired equilibria using the angle between the CLF and CBF control vectors, and activating a smooth circulation input to deflect robots away from saddle points. We also incorporate a LiDAR-based CBF for online obstacle detection and a CBF-based conflict resolution (CBF-CR) for multi-robot priority handling. We validate Circ-CLBF in MATLAB across several scenarios: single-robot navigation around non-convex and circular obstacles, multi-robot opposing-goal cases, and a 12-robot two-room warehouse simulation. The results show that Circ-CLBF successfully escapes undesired equilibria in all tested cases while achieving safe navigation outside the safety distance throughout all tests.

Keywords: Nonlinear control and applications, Robotics and swarm intelligence, Control theories
Abstract: This paper presents a dynamic modeling and an observer based robust control design for an Unmanned Aerial Manipulator (UAM) system, which comprises of an Unmanned Aerial Vehicle (UAV), an active manipulator and a gripper. The dynamics of UAM is challenging as it possess high coupling characteristics, has nonlinearities and it is underactuated as well; therefore, it requires two separate control frameworks to address the stability of the overall system. By considering the system as two subsystems (UAV and manipulator), wherein each subsystem acts as a disturbance to the other. A Lyapunov stability-based backstepping controller with disturbance observer is proposed for the hexacopter UAV and a robust controller design is opted for the manipulator to overcome internal disturbances from the UAV. The performance of the controller is validated in the Software-In-The-Loop (SITL) simulation with the help of PX4 autopilot and the Robot Operating System (ROS) with Gazebo simulator. The robustness verification is performed through two cases: 1) by continuously varying the manipulator joint angle, whereas UAV tracking time-varying trajectory, and 2) grasping a payload and performing a pick operation while hovering. These cases change the overall center of mass of the system, which has potential applications in aerial grasping and delivery.

Keywords: Motion and vibration control, Communication
Abstract: Small drones are used in an increasing number of applications. Due to their limited computation and communication capabilities, the networked control of such drones is becoming attractive. In this paper we present a methodology for designing the networked control, with a focus on determining the best sampling period, from the point of view of both quality of control and quality of network transmissions. The design is based on linear matrix inequality conditions, along with benchmark tests on different software and hardware platforms. An experimental case study on a QAV250 drone using the PX4 Autopilot software shows the effectiveness of the approach.

Keywords: Mechatronics, Control devices, sensors and actuators
Abstract: Swashplateless rotors offer a novel approach to UAV propulsion by enabling directional force control without the mechanical complexity of swashplates or the reliance on coupled thrust generation in conventional multi-rotor systems. Since their introduction in 2013, they have inspired research in actuator design, control strategies, and UAV applications. This review consolidates the current state of knowledge, highlighting the unique trade-offs of swashplateless systems compared to conventional fixed-pitch and swashplate-based rotors, and summarizing the different UAV implementations, their design choices, operational achievements, and the challenges encountered in practice. The paper provides a structured overview of existing research, enabling a systematic comparison of different approaches and offering insights into current limitations and potential research directions.

Keywords: Mechatronics, Control theories, Control devices, sensors and actuators
Abstract: This study proposes two counter-steering assist con- trol methods for a steer-by-wire bicycle to enable stable turning transitions for riders who are unable to actively lean the bicycle body. The first method is based on equilibrium point analysis. A nonlinear bicycle dynamics model is derived, and the roll angle required to maintain the balance between centrifugal and gravitational forces is determined. A steering control is designed to track the target roll angle trajectory. Simulations confirmed that counter-steering is successfully reproduced; however, riding experiments revealed that the roll angle failed to track the target trajectory, resulting in a loss of stability. The second method utilizes a virtual trail effect reproduced by the steer-by-wire mechanism. By temporarily adding a virtual roll angle opposite to the turning direction, counter-steering is naturally induced at the front wheel. Experimental results demonstrated that the bicycle successfully transitioned from straight to turning motion under all tested conditions. These results indicate that the trail-effect-based method is more suitable for practical counter-steering assistance, as it achieves stable turning transitions while maintaining compatibility with the rider’s natural steering behavior.

Keywords: Robotics and swarm intelligence, Intelligent control, Nonlinear control and applications
Abstract: This paper presents a learning-enhanced Model Predictive Control (MPC) framework for bipedal robot locomotion. We replace the conventional interior-point QP solver with a Pontryagin’s Maximum Principle (PMP)-based optimizer that analytically computes the gradient and Hessian of the cost function for the centroidal dynamics of a point-foot biped, exploiting the linear time-varying structure to derive closed-form costate recursion and superlinear control updates. To address the modelplant mismatch inherent in the simplified centroidal model, we introduce a residual learning framework that integrates a lightweight multilayer perceptron (MLP) with the PMP solver to predict one-step dynamics residuals, preserving the physicsbased nominal model as an inductive bias. Validation on the PF TRON1A bipedal robot in MuJoCo over ten independent 20 s walking trials shows that the PMP solver achieves tracking accuracy comparable to the HPIPM baseline while maintaining real-time feasibility, and the learning-enhanced variant further reduces the mean tracking error and substantially reduces the lateral position error, demonstrating significantly straighter walking with negligible computational overhead.

Keywords: Control theories, Deep learning and machine learning, Computational intelligence
Abstract: Federated Learning (FL) has emerged as a powerful method for collaborative model training without sharing raw data. However, FL faces significant challenges, including privacy risks, non-IID data distributions, and slow convergence speed. In this paper, we focus on optimizing the local update process in FL to enhance convergence speed and model performance. We propose FedOCPs, a method that incorporates simplified Optimal Control Principle (OCPs) to improve local optimization by implicitly incorporating curvature information, thereby achieving faster convergence in local optimization and improving global model performance. Experiments on the USPS benchmark demonstrate that FedOCPs outperforms baseline methods under both IID and non-IID settings. It achieves 95.71% test accuracy under the Dirichlet non-IID distribution, demonstrating its effectiveness in both convergence speed and test accuracy.

Keywords: Deep learning and machine learning, Control theories, Big data
Abstract: Adaptive first-order algorithm have become the de facto standard for training deep neural networks. However, their empirical success is largely evaluated under fine-tuning settings with pretrained models, where optimization starts from a near-optimal region of the loss landscape. In this work, we argue that such evaluation protocols may significantly underestimate the potential of second-order optimization algorithms. We present a second-order algorithm and demonstrate that its advantage becomes pronounced in training-from-scratch scenarios, where the optimization problem is highly non-convex and ill-conditioned. In contrast, under pretrained fine-tuning settings, the performance gap between first-order and second-order methods diminishes due to the proximity to local minima. Our findings suggest that current benchmarking practices may provide an incomplete picture of optimizer effectiveness. We advocate for a more comprehensive evaluation framework that accounts for different optimization regimes, particularly those involving challenging training dynamics from random initialization.

Keywords: Nonlinear control and applications
Abstract: This paper studies a class of discrete-time nonlinear optimal control problems with control constraints. By introducing a virtual control variable, the original constrained problem is transformed into an unconstrained optimal control problem while automatically satisfying box-type input bounds, including asymmetric ones. Based on this transformation, a second-order optimal control principle (OCP)-based numerical algorithm is developed. The gradient is computed through backward costate recursion, while the Hessian is obtained via auxiliary forward-backward difference equations, avoiding numerical differentiation and external optimization solvers. Numerical simulations show that the proposed method achieves fast convergence and high computational efficiency while maintaining the desired control performance.

Keywords: Control theories, Communication, Big data
Abstract: This paper studies distributed K-clustering over directed communication networks, where each agent accesses only local data and collaborates to recover the global cluster centers without sharing raw datasets. Different from existing Alternating Directions Method of Multipliers (ADMM)- or consensus-based first-order methods, we develop an optimal-control-based framework in which the fixed-assignment clustering subproblem is modeled as a consensus-coupled discrete-time optimal control problem. Based on the associated optimality system, an explicit distributed multi-step update law is derived that incorporates averaged gradient and curvature information. For the single-cluster case, the resulting algorithm achieves superlinear convergence to the global cluster center. For the multi-cluster case, a periodic assignment update mechanism is introduced, under which the assignment sets stabilize and the superlinear convergence property is preserved. Numerical results show faster convergence than existing gls{admm}-based methods and accurate recovery of the global clustering structure.

Keywords: Computational intelligence
Abstract: This paper focuses on the low-cost storage and implementation of an optimization algorithm grounded in optimal control principles. The original OCP-based algorithm features two major advantages: an almost quadratic convergence rate and stable convergence behavior. It is superior to Newton's method because it does not require Hessian definiteness and is insensitive to initial guesses. However, the original algorithm remains heavily dependent on the explicit computation and storage of the Hessian, as well as the inversion of associated matrices. Consequently, its direct application to high-dimensional or storage-constrained problems inevitably incurs significant computational burdens and memory overhead. To address these limitations, we leverage Hessian-vector products and the conjugate gradient method to circumvent the need for explicit Hessian matrices and matrix inversion. This yields a practical OCP-based optimization algorithm that preserves the original advantages and extends its scope of applicability while overcoming the aforementioned limitations.

Keywords: Nonlinear control and applications, Industrial applications, Energy Systems
Abstract: Accurate load prediction is central to electricity scheduling, but the operational objective is not simply to minimize prediction error. In power systems, over- and under-prediction have asymmetric economic consequences, and battery storage can absorb part of these errors before costly imbalance actions are required. This talk develops a battery-aware support vector regression framework in which the ε-insensitive tube has a direct physical interpretation: ε represents the usable battery capacity, while asymmetric tube widths represent stored energy available for discharge and remaining space for charging. The prediction is therefore viewed as a control action, chosen to minimize expected operational cost under battery-state dynamics rather than merely to approximate future load. The resulting formulation links asymmetric loss functions, robust statistical learning, and feedback decision rules for energy systems. In a myopic form, the optimal predictor generalizes asymmetric-loss quantile prediction; in a dynamic form, it leads naturally to a model-predictive-control interpretation with battery-state feedback. The framework provides a principled way to jointly consider forecast accuracy, imbalance prices, battery capacity, and operating cost, and illustrates how machine-learning predictors can be redesigned for decision-oriented control applications in power and industrial systems.

Keywords: Control theories, Robotics and swarm intelligence, System identification and modelling
Abstract: This paper presents a trajectory tracking framework for a two-degree-of-freedom (2-DOF) XY robot using a cascade control structure with reference feedforward compensation. The robot motion is planned in the Cartesian workspace using a modified artificial potential field (APF) method with filtering combined with a jerk-limited velocity-field trajectory generator to produce dynamically feasible reference trajectories. The generated trajectories explicitly satisfy velocity, acceleration, and jerk constraints to ensure smooth actuator motion. The proposed control system consists of an outer position loop and an inner velocity loop, where feedforward terms are introduced to improve tracking performance. The system is implemented on an ESP32-based embedded controller that drives two DC motors via ball screw mechanisms. Experimental results demonstrate that the proposed method significantly improves trajectory tracking accuracy. In particular, the mean squared tracking error is reduced by more than 20x compared with a conventional cascade controller without feedforward compensation. These results confirm the effectiveness of the proposed trajectory generation and control framework for precise motion control of multi-axis robotic systems.

Keywords: Robotics and swarm intelligence, Mechatronics
Abstract: This paper investigates the finite-time practical consensus problem for single-integrator multi-agent systems with fixed, undirected communication topologies. An inclusive, general class of self-triggered controllers is proposed and shown to converge in a finite-time with rigorous stability analysis. Zeno behaviour is proved to be excluded and an upper bound on the settling time is given. Numeric simulations of multiple examples of controllers are presented to verify the theoretical results.

Keywords: Control theories, Nonlinear control and applications, Intelligent control
Abstract: This paper investigates the stabilization problem of continuous-time Takagi-Sugeno (T-S) fuzzy systems with unknown system matrices. A switched non-parallel distributed compensation controller is designed using input-state data corrupted by measurement noise. Two data collection scenarios are considered: 1) local model data obtained from individual subsystems, and 2) data collected from the overall fuzzy system under the assumption that the membership functions (MFs) are measurable. By adopting a non-quadratic Lyapunov function and modeling measurement noise via quadratic constraints, data-driven controller design conditions are derived in the form of linear matrix inequalities for both scenarios. Simulation results from two illustrative examples demonstrate the effectiveness of the proposed method and reveal notable differences in conservatism between the two data collection strategies.

Keywords: System identification and modelling, Nonlinear control and applications, Mechatronics
Abstract: This paper presents the development and experimental validation of a highly nonlinear multibody system model for a three-degree-of-freedom (3-DOF) hovering system. This model captures the coupled rotational dynamics and kinematic constraints of the moving bodies. The paper introduced a method for estimating the nominal speed of the rotor(s) according to the asymmetric geometry of the system components, that attain the starting point of operation. This point can be utilized in the linearization process of the system. Experimental testing is conducted to evaluate the expected accuracy of the model under disturbance conditions. The comparison between simulated and measured responses demonstrates consistent agreement across roll, pitch, and yaw axes. The results confirm that the proposed model captures the dominant system dynamics with sufficient accuracy, supporting its use in subsequent nonlinear control design.

Keywords: System identification and modelling, Energy Systems, Measurement and instrumentation
Abstract: Accurate battery modelling is essential to improve the performance and reliability of battery management systems. This study investigates the voltage prediction performance of an equivalent circuit model (ECM) for a 3000 mAh Nickel Manganese Cobalt (NMC) lithium-ion battery cell (LG HG2) under dynamic operating conditions. Various ECM structures, including one, two, and three RC branches with and without hysteresis modelling, were evaluated using pulse discharge and UDDS current profiles. Model parameters were extracted from empirical data and model performance was assessed by comparing the simulated voltage against measured values. The results demonstrate that the ECM accurately reproduces battery voltage behaviour, achieving an average relative error of approximately 0.58% to 0.87%. Additionally, the model maintains a root mean square error (RMSE) ranging from 30 mV to 37 mV, which corresponds to a relative RMSE of less than 1%. Furthermore, increasing the model order beyond two RC branches provides only marginal improvements in accuracy, whereas hysteresis modelling enhances model robustness during transient conditions. These results demonstrate that the second-order ECM incorporating hysteresis modelling achieves a favourable trade-off between model accuracy and computational complexity, making it well suited for battery management system applications.

Keywords: Control theories, Nonlinear control and applications, Energy Systems
Abstract: This paper proposes a distributionally robust model predictive control (DR-MPC) framework for mixed logical dynamical (MLD) systems affected by stochastic disturbances with bounded unknown distributions. By constructing a Wasserstein ambiguity set to characterize distributional uncertainty, the control problem is formulated as a two-stage optimization problem, which aims to minimize control costs while penalizing constraint violations under the worst-case distribution within the set. This framework is derived from two-stage distributionally robust optimization, and offers the advantage of adaptive constraint tightening without requiring prior computation. To solve the problem, a cutting-plane algorithm is developed, which iteratively solves mixed-integer linear programming and linear programming subproblems. The proposed method is applied to a battery thermal management system for electric vehicles. Simulation results demonstrate that the approach achieves robust temperature regulation and efficient energy management under uncertain thermal power disturbances.

Keywords: Control devices, sensors and actuators
Abstract: Brushless Direct Current (BLDC) motors are widely used in applications requiring high efficiency and fast dynamic response, yet their performance is often limited by torque ripples, nonlinear characteristics, and disturbances from varying loads. This study evaluates the capability of a PID controller in regulating BLDC motor speed and examines its behavior under different load conditions. A BLDC motor model was developed using system identification, followed by the implementation of PID controller. Experimental validation was conducted using a microcontroller-based drive system equipped with voltage, current, and speed sensors. The PID controller successfully achieved reference speeds of 200, 300, and 400 rpm; however, all experiments exhibited underdamped responses characterized by overshoot, oscillations, and steady-state ripples. Input voltage and power measurements showed increasing energy demands at higher speeds and under heavier loads. While PID control provided acceptable performance for baseline conditions, its sensitivity to disturbances highlights the need for more advanced strategies. These results provide practical guidance for selecting effective BLDC motor control strategies based on PID in regard to the torque dynamics due to load variation.

Keywords: Control devices, sensors and actuators, Adaptive systems, Autonomous vehicles
Abstract: Safe navigation of electric wheelchairs requires reliable surrounding perception, particularly in crowded and confined environments. However, perception systems based on a single LiDAR are limited by their field of view and mounting positions, resulting in blind spots and incomplete coverage. To address this issue, we propose a multi-LiDAR system that enhances environmental perception through multi-sensor fusion and grid-based point-wise labeling. The system adopts an offline calibration pipeline with pose graph optimization to ensure precise sensor alignment. Experimental results demonstrate that our approach achieves near-complete surrounding coverage, successfully classifying points into passable, occupied, and unknown categories. Additionally, a soft synchronization strategy ensures stable, real-time performance even with asynchronous sensor inputs. This provides a robust and efficient perception solution for autonomous wheelchair safety.

Keywords: Control devices, sensors and actuators, Control theories
Abstract: Large-scale multivariable processes with strong interactions pose significant challenges for centralized control due to high computational and communication requirements. Distributed Model Predictive Control (DMPC) addresses these challenges by decomposing the plant into smaller subsystems regulated by local controllers. This paper presents a comparative study of Centralized Model Predictive Control (CMPC), cooperative DMPC, and non-cooperative DMPC applied to an octuple-tank benchmark system. A graph-theoretic decomposition framework is used to generate modular distributed control architectures and to examine the influence of subsystem partitioning on control performance. Both cooperative and non-cooperative DMPC schemes are implemented with iterative coordination strategies. Closed-loop performance is evaluated under regulatory and servo operating conditions using performance indices and computational time metrics. Simulation results show that cooperative DMPC achieves performance close to CMPC while significantly reducing computational complexity, whereas non-cooperative DMPC provides greater computational savings at the expense of moderate performance degradation. The study highlights the importance of modularity-based decomposition and coordination strategies for scalable and efficient distributed predictive control of large-scale interconnected systems.

Keywords: Control devices, sensors and actuators, Deep learning and machine learning, Autonomous vehicles
Abstract: Sensor outages, caused by signal loss, interference, or hardware faults, present a major challenge for reliable state estimation which in turn result in degraded tracking performance in autonomous and control systems. During such outages, classical estimators operate in prediction-only mode, which can result in drift and significant degradation of estimation accuracy. In this paper, for outdoor environments, we consider an inertial measurement unit (IMU) – global positioning system (GPS) integrated navigation scenario, where temporary sensor loss, such as GPS outages which occur when natural or artificial factors obstruct GPS signal, that may severely affect the system performance. To address this limitation, we propose an event-driven hybrid long short-term memory (LSTM) and extended Kalman filter (EKF) framework that activates an offline trained LSTM network only during sensor outages. The LSTM network predicts surrogate sensor measurements from historical patterns, which are then fused with the EKF to sustain accurate and stable estimates. Once normal sensor operations resume, the system seamlessly switches back to original sensor measurements for state estimation. Simulation results demonstrate improved robustness, reduced estimation error, and resilience under diverse outage scenarios, highlighting the suitability of the proposed approach for safety-critical applications.

Keywords: Control devices, sensors and actuators, Health systems, Motion and vibration control
Abstract: Mechanical vibration can induce upper-limb motion through tonic vibration reflex (TVR), providing a lightweight alternative to conventional externally actuated rehabilitation systems. Previous studies have demonstrated that vibration-induced responses can be extended to coordinated multi-degree-of-freedom (multi-DOF) upper-limb motion, and that heterogeneous vibration frequencies can modulate inter-joint coordination. However, the control of such vibration-induced motion has not been sufficiently investigated. In this study, the effective actuator is regarded as a coupled vibration-muscle neuromuscular unit rather than a standalone motor. Based on previous experimental findings, a simulation framework is developed for shoulder-elbow motion control, in which joint-specific vibration frequencies are treated as control inputs and upper-limb kinematic responses as system outputs. Closed-loop simulations are conducted for dual-joint velocity tracking, coordination regulation, and motion-mode switching. The results demonstrate the feasibility of frequency-based control for multi-DOF upper-limb motion and provide a foundation for future human experiments and intention-aligned rehabilitation systems.

Keywords: Control devices, sensors and actuators, Industrial applications
Abstract: The hardware architecture for a flat-skin vision-based tactile sensor (VBTS) is presented in this paper. It is designed to be integrated with robotic manipulators that work in contact-intensive environments. The suggested system uses visual markers embedded in a flexible elastomeric skin to produce high-resolution optical images of surface deformation, in contrast to traditional tactile sensors that depend on discrete force or pressure sensing elements. The design combines important hardware factors into a single, repeatable architecture, including material selection, sensing element selection, camera placement, illumination strategy, and signal acquisition. Strongness, scalability, and compatibility with current robotic control systems are all taken into consideration. Experimental findings from several sensor prototypes show dependable imaging performance in a variety of contact situations, increased mechanical compliance, and better marker tracking. The architecture also makes it easier to integrate with downstream control and perception algorithms. To encourage reproducibility and make it easier to adapt the VBTS to various robotic platforms and manipulation tasks, comprehensive design specifications and hardware justifications are supplied.

Keywords: Control devices, sensors and actuators, Industrial applications, Measurement and instrumentation
Abstract: Template matching is a reliable technique for object detection in industrial applications where labeled data is limited. This paper compares the robustness of four similarity metrics namely error-based correlation (EBC), normalized sum-of-squared differences (NSSD), normalized cross-correlation (NCC), and zero-mean NCC (ZNCC), against synthetically generated real-world disturbances. Using a Monte-Carlo simulation with 1000 randomized trials, we evaluated performance under Gaussian noise, non-linear brightness changes, and directional lighting. We also analyzed the efficiency of a coarse-to-fine search strategy. Our results show that while ZNCC remains the most robust against illumination variations, the EBC metric serves as a superior ``strict validator" for precision tasks, offering better noise resilience while enforcing strict intensity compliance. Additionally, we demonstrate that a pyramid scale of 0.2 reduces computational time by approximately 25 times without compromising detection accuracy. We conclude that ZNCC is optimal for dynamic environments, while EBC is best suited for strict quality control.

Keywords: Intelligent control, System identification and modelling, Control devices, sensors and actuators
Abstract: High-Temperature Short-Time pasteurization requires precise temperature control to ensure food safety and preserve product quality. While Model Predictive Control offers advanced multivariable management, its performance degrades significantly under industrial sensor noise and transport delays, forcing controllers into reactive behaviors that degrade mechanical components and waste energy. This paper proposes a Kalman-augmented Model Predictive Control architecture to regulate a thermal pasteurization miniplant. By utilizing a multiple-input single-output (MISO) model, a discrete-time Kalman Filter is integrated to reconstruct unmeasurable states and provide optimal state estimation against highly pulsatile flow disturbances. Simulation results demonstrate that the proposed architecture greatly reduces high-frequency control signal chattering in the heating element compared to a standard output-feedback predictive controller. During setpoint tracking, this noise mitigation translates to a significant 3.37-fold reduction in actuator switching counts. While total energy reduction is marginal at 0.06 percent, the improved control smoothness directly enhances Solid State Relay (SSR) longevity. Furthermore, under complex fluctuating flow disturbances, the Kalman-augmented system reduces actuator switching activity by 51.18% and achieves a 74.36% improvement in steady-state tracking precision compared to the standard configuration. The findings confirm that combining predictive c

Keywords: Nonlinear control and applications, Intelligent control, Adaptive systems
Abstract: This paper presents a comparative study of control strategies for robotic manipulators under actuator input saturation and model uncertainties. Four control schemes are evaluated on a 2-DOF robotic manipulator: classical Proportional-Derivative (PD) control, Computed Torque Control (CTC), Slotine-Li's adaptive control, and a proposed model-free adaptive control (MFAC) based on an enhanced Zeroing Neural Network (ZNN). Simulation results reveal that classical method is primarily bottlenecked and degraded by model uncertainties rather than input saturation, model-based method is affected by both model uncertainties and input saturation, whereas model-free approach faces the exact opposite challenge than classical method. Because the proposed enhanced-ZNN--based-MFAC requires no prior dynamic modeling, it easily overcomes model uncertainties to deliver superior tracking. However, its transient learning phase inherently generates massive initial control efforts, inevitably triggering severe actuator saturation. To ensure hardware viability, this paper proposes a direct Lyapunov approach that explicitly handles this non-smooth saturation nonlinearity as a bounded disturbance. We mathematically prove and empirically demonstrate that the proposed enhanced-ZNN-based-MFAC can safely absorb its own self-induced saturation spikes while achieving faster convergence and higher precision compared to classical PD, CTC, and Slotine-Li's adaptive controllers.

Keywords: Adaptive systems, Nonlinear control and applications, Intelligent control
Abstract: CMAC generates its output by discretely partitioning the input space and referring to the weight values corresponding to the cell to which the input belongs. As a result, the output tends to become discontinuous near the boundaries of the partitioned input space. In contrast, Fuzzy Cerebellar Model Articulation Controller (Fuzzy CMAC) evaluates the degree to which the input belongs to each input region in a continuous manner, enabling smooth linear approximation. This paper proposes an adaptive output feedback control system design strategy, which employs an adaptive feedforward compensator based on the Fuzzy CMAC and an adaptive feedforward input based on CMAC. The effectiveness of the proposed approach is shown via numerical simulations.

Keywords: Autonomous vehicles, Artificial intelligence, Deep learning and machine learning
Abstract: This study investigates the impact of depot placement on operational efficiency in demand-responsive mobility services using the Smart Satellite City Simulator. To isolate the effect of depot layout, the online dispatching strategy is fixed to a greedy rule that assigns each request to the nearest available vehicle. We compare the simulator’s original depot placement (Baseline) with clustering-based placement derived from k-means (K). Simulation results under 1,000-user and 5,000-user scenarios show that clustering-based depot placement reduces total travel distance by up to 7.7% and decreases required fleet size by over 10%, with negligible impact on user travel-time satisfaction. These results demonstrate that optimized depot placement significantly improves system efficiency without degrading service quality.

Keywords: Autonomous vehicles, Artificial intelligence, Robotics and swarm intelligence
Abstract: In a dynamic environment, an electric wheelchair must consider pedestrian motion to avoid collisions. Although future position and velocity can be predicted, how to incorporate such predictions into path planning remains unclear. This study constructs three methods within an A*-based framework: (i) a Hard constraint approach treating predicted regions as non-traversable, (ii) a Risk-aware approach incorporating prediction as a continuous cost, and (iii) a time-dependent Hybrid approach switching between these strategies depending on prediction horizon. Simulation results show that the Hard approach improves safety but increases stopping, while the Risk-aware approach enables smoother motion but reduces the safety margins. The Hybrid approach balances these characteristics, suggesting that constraint design should reflect prediction uncertainty.

Keywords: Control devices, sensors and actuators, Measurement and instrumentation
Abstract: This paper presents a systematic investigation of underwater ultrasonic wireless power transmission using wafertype piezoelectric transducers with various thicknesses. Unlike previous studies primarily focused on general transmission feasibility and fixed transducer configurations. This work specifically analyzes the influence of transducer thickness on acoustic and electrical transmission characteristics in underwater environments. A 3D COMSOL Multiphysics® was developed to evaluate the effects of frequency, transducer thickness, transmission distance, and beam directivity on electric field distribution, acoustic pressure, and sound pressure level. Thinner transducers exhibit sharper beam focusing and higher localized acoustic intensity, whereas thicker transducers provide more stable and spatially uniform transmission characteristics. Higher frequencies produce narrower directional beams and enhanced acoustic concentration but also increase side-lobe sensitivity. Polar directivity analysis the significant influence of transducer thickness and transmission distance on beam propagation behavior. Experimental validation was performed using a PZT-5A transducer pair (transmitterreceiver) in underwater environment with a 50-ohm load, where an input of 19 V achieved a practical efficiency of around 3% at 20 kHz. Since the current experiments were limited to selected 20 kHz cases, the multi-frequency simulation results at 40, 60, and 80 kHz.

Keywords: Control theories, Cyber-physical systems and security, Communication
Abstract: This paper addresses the distributed observer design problem on sensor networks operating under randomly intermittent communication. We derive cooperative gain conditions that improve estimation error convergence for both bidirectional and unidirectional inter-node communication scenarios. Based on the conventional Kalman consensus filter, we propose a novel approach in which a Lyapunov function is constructed as a weighted sum of subsystem Lyapunov functions, and node-dependent cooperation gains are assigned according to these weights. Furthermore, by formulating the gain conditions within a linear matrix inequality (LMI) framework, the conservative constraints imposed in previous studies are relaxed, and practical design guidelines adapted to communication probabilities are provided. Numerical simulations demonstrate that the proposed method significantly improves convergence speed compared with conventional approaches in both bidirectional and unidirectional communication cases.

Keywords: Control theories, Deep learning and machine learning, Intelligent control
Abstract: In large-scale mobile welding scenarios, mobile manipulators must achieve high-precision tracking over long straight and large-radius curved seams. However, deployment remains challenging due to strong kinematic coupling between the mobile base and manipulator, control precision disparities, and visual perception degraded by noise, bias, and intermittent feature loss. This paper proposes a Robust Closed-loop Seam Tracking (RCST) method for practical deployment. A slow-perception and fast-control temporal architecture with observation buffering maintains closed-loop continuity under intermittent detection failures. The base and manipulator are modeled as collaborative agents: the base performs coarse heading and velocity tracking, while the manipulator provides lateral compensation. A unidirectional action-sharing mechanism mitigates base-induced disturbances and improves TCP stability. Extensive experiments show that RCST outperforms PID, MPC, single-agent RL, and standard MARL baselines. The method maintains stable closed-loop performance under severe perceptual disturbances, achieving lower tracking errors, smoother control, and real-time feasibility.

Keywords: Control theories, Motion and vibration control
Abstract: We consider a flexible structure governed by a wave equation. It has been shown that under specific boundary conditions, actuation and sensing, the transfer function of the system can be expressed as a ratio of two quasi-polynomials. We examine the finite dimensional approximations of this model and study under which conditions a robustly stabilizing controller exists. In particular, we show that for an undamped system, finite dimensional controllers obtained from a specific approximate model cannot stabilize the original infinite-dimensional model. For the damped system we propose a method for finite dimensional robustly stabilizing controller design.

Keywords: Control theories, Nonlinear control and applications
Abstract: A prescribed performance control (PPC) problem under given constraints for the magnitude and rate of the control signal is considered for uncertain nonlinear relative-degree-one systems affected by external perturbation. In order to solve such task, we propose a robustified version of the well-known Nested Saturation Feedback called Nested Saturation Robust Feedback (NSRF), which: i) simultaneously ensures the predefined transient quality and boundedness by given values of not only magnitude but also rate of the control signal, ii) does not use reciprocal barrier functions (i.e., functions, which grows to infinity whenever its arguments approach some finite limits) in feedback to avoid discontinuities, iii) provides opportunity to chose all control law parameters via solution of a set of linear inequalities by conventional software tools. The theoretical results are validated via an illustrative example.

Keywords: Autonomous vehicles, Intelligent control, Motion and vibration control
Abstract: This study mitigates low-frequency vibrations during electric vehicle regenerative braking in accordance with the ISO 2631-1 standard. An enhanced MPC (eMPC) was designed integrating a frequency weighting filter along with a PMSM-based second-order longitudinal vehicle dynamics model. By structurally optimizing motion sickness-inducing factors through an index-integrated strategy, the eMPC demonstrated a 99.8% reduction in Jerk RMS and an 86.1% reduction in Motion Sickness Dose Value (MSDV) compared to conventional PID control. Future research will validate this control structure using a nonlinear third-order extended model.

Keywords: Cyber-physical systems and security, System identification and modelling, Control devices, sensors and actuators
Abstract: DC microgrids play a crucial role in integrating renewable energy sources (RES), but are more susceptible to incipient faults resulting in serious accidents. Considering the incipient fault symptoms and numerous disturbances, the incipient fault detection in DC microgrids is challenging. Further, the cyber-physical features in DC microgrids establish correlations among subsystems, which can be used to enhance fault detection capability. In this paper, a distributed detection scheme for the incipient fault in DC microgrids based on residual correlation analysis is proposed. To be specific, based on the Lyapunov stability theory and H∞ performance, a distributed observer is designed to generate residuals with strong robustness against disturbances. Then, the canonical correlation analysis (CCA) is employed to offline identify the correlations between local and interconnected residuals. Further, the online distributed fault detector is constructed using these residuals, which reduces the uncertainty in the detection signal and thus enhances fault detection sensitivity. Finally, a DC microgrid consisting of 6 distributed generation units (DGUs) is used to demonstrate the superior detection performance of the proposed method over existing conventional methods using observer-based direct residual assessment.

Keywords: Brain-computer interfaces
Abstract: Alzheimer’s disease (AD) has traditionally been characterized as a disorder of disrupted brain connectivity. However, emerging evidence suggests that alterations in temporal dynamics may play a more fundamental role. In this study, we investigated whether AD is associated with reduced temporal variability of neurovascular complexity networks derived from resting-state functional near-infrared spectroscopy (fNIRS). Data were collected from 83 participants, including healthy controls (HC, n = 27), mild cognitive impairment (MCI, n = 37), and AD patients (n = 19). Spectral entropy (SE) was computed using a sliding-window approach and used to construct dynamic complexity coupling networks. Graph-theoretical metrics, including clustering coefficient, nodal strength, and global efficiency, were extracted across time, and variability descriptors (standard deviation, coefficient of variation, and range) were quantified. Compared to HC and MCI, AD exhibited significantly reduced variability across all network metrics, accompanied by increased temporal persistence, indicating reduced flexibility and increased rigidity of network dynamics. These findings suggest that AD is characterized not only by reduced network integration but more critically by a contraction of dynamic variability, reflecting impaired neurovascular adaptability. Dynamic variability measures may serve as sensitive biomarkers for early detection and monitoring of Alzheimer’s disease.

Keywords: Intelligent control, Nonlinear control and applications
Abstract: Attitude control of reusable launch vehicles (RLVs) during the terminal area energy management (TAEM) phase is challenging due to high non-linearity, multi-channel coupling, and significant uncertainties. This paper proposes a robust end-to-end control strategy based on an improved twin delayed deep deterministic policy gradient (TD3) algorithm. By integrating appointed-time sliding mode control (SMC) principles into the deep reinforcement learning (DRL) framework, a sliding-manifold-augmented state space is designed to enhance structural stability, complemented by a hybrid reward function to accelerate convergence and eliminate steady-state errors. Simulation results demonstrate that the proposed controller achieves superior tracking precision and robustness under 40% parameter perturbations and external disturbances. Furthermore, the learned policy exhibits excellent generalization capability when subjected to expanded, untrained mission profiles.

Keywords: Nonlinear control and applications, Intelligent control, Autonomous vehicles
Abstract: Autonomous Underwater Vehicles (AUVs) operating in dynamically disturbed marine environments require robust control strategies to maintain stable navigation performance. In particular, Internal Solitary Wave (ISW) disturbances can induce significant velocity variations and trajectory deviations. This paper proposes a Disturbance Observer–based Nonlinear Model Predictive Control (DO-NMPC) framework to enhance AUV trajectory tracking under such conditions. The disturbance observer estimates unknown external forces online by comparing predicted and measured velocities, and the estimated disturbances are incorporated directly into the NMPC optimization process. The proposed method is validated through simulations in a Gazebo-based environment with varying current disturbances. Results demonstrate that DO-NMPC improves tracking accuracy, reduces steady-state error, and generates smoother thruster commands compared to conventional NMPC. These findings indicate that disturbance-aware predictive control significantly enhances robustness and navigation stability for AUV operations in uncertain underwater environments.

Keywords: Nonlinear control and applications, Adaptive systems, Control theories
Abstract: This paper presents an adaptive control for a 6 DoF quadrotor unmanned aerial vehicle (UAV) carrying a fluid container mounted at the bottom of the vehicle. The motion of the UAV induces fluid sloshing, which in turn generates instability in the overall system. This phenomenon is captured as a bounded but unknown variation in inertia parameters. Identifying the sloshing as the 2-DoF pendulum motion and modeling the moment of inertia through it, we propose an adaptive control strategy to mitigate these parametric uncertainties. Using a coordinate-free setting and persistence of excitation, the control design has two stages: 1) inner-outer loop control for the underactuated UAV which is singularity-free, and 2) the adaptive control, to estimate the unknown mass, all elements in the moment of inertia matrix, and external disturbances in the rotational dynamics. The resultant closed-loop system is asymptotically stable, and the simulation results obtained show the efficacy of the proposed controller.

Keywords: Nonlinear control and applications, Adaptive systems, Control theories
Abstract: This paper develops a constrained nonlinear control framework that integrates phase-switch sequential regulation, zone barrier Lyapunov function (zBLF) based boundary protection, and adaptive disturbance compensation for disturbed MIMO systems. The proposed design aims to simultaneously realize three objectives: structured channel-wise transient shaping, deactivation of unnecessary planar regulation inside a prescribed soft region, and enhanced repulsive protection when the system state approaches a hard constraint boundary. To this end, a phase-switch mechanism is introduced to organize the regulation process according to a prescribed yaw-sway-surge priority, while a circular zBLF-based term is embedded to preserve constraint satisfaction and strengthen boundary-sensitive recovery under adverse disturbances. An adaptive disturbance-compensation structure is further incorporated to improve robustness against unknown matched perturbations. Rigorous analysis shows that the developed controller guarantees the intended sequential regulation behavior, maintains positivity of the hard-boundary margin, and suppresses repeated large-amplitude excursions outside the soft admissible region. Finally, the proposed theory is validated through simulations on a disturbed 3-DOF marine vessel model, where the results demonstrate improved sequential transient performance, reduced repeated soft-boundary crossings, and effective hard-boundary protection compared with representative baseline desi

Keywords: Nonlinear control and applications, Autonomous vehicles, Control theories
Abstract: In this paper, we propose computationally inexpensive obstacle avoidance constraints based on the Separating Axis Theorem (SAT) that account for geometric shapes of obstacles and the robot. Model Predictive Control is typically computationally expensive, particularly for obstacle avoidance constraints which require complex computations. To address this problem, we introduce a new approximation that restricts the candidate separating axes. We then reconstruct the algorithm to utilize the complementary biases of this axis restriction and the Log-Sum-Exp function, enabling them to compensate for each other and effectively reduce computational load. Implemented in a hierarchical MPC framework, the proposed method maintains collision avoidance and prevents the robot from stalling. Experimental results with an electric wheelchair demonstrate real-time control while suppressing stalling, collisions, and excessive avoidance.

Keywords: Nonlinear control and applications, Autonomous vehicles, Measurement and instrumentation
Abstract: This paper presents a solution for runway localization and feature extraction for visual estimation of deviations of the aircraft with respect to an unknown runway. In this setting, a combination of visual features and known camera velocity from the IRS (Inertial Reference System) is used as input to a reduced-order observer. Strategically, we choose runway threshold front corners as two visual features to avoid the loss of observability during the landing maneuver. A scheme to select the relevant visual features from multiple coordinates of a mask contour, obtained from a YOLOv8 runway segmentation algorithm, is proposed. The efficiency of the feature selection algorithm is verified by integrating it with a non-linear observer to estimate aircraft deviations during the landing phase. It is evaluated on a realistic scenario generated from synthetic images for a conventional trajectory of a civil aircraft landing. A quantitative evaluation of the proposed feature extraction algorithm against classical corner detection methods such as Harris, Shi–Tomasi, and FAST shows comparable performance in terms of localization accuracy and detection success rate, while achieving superior computational efficiency.

Keywords: Industrial applications, Deep learning and machine learning
Abstract: Aiming at the limited cross-condition generalization of few-shot bearing fault diagnosis methods, this paper proposes the Semi-Supervised Meta-Mahalanobis Prototypical Network (SS-MetaMPN), which integrates unlabeled data to reduce reliance on scarce labeled samples, thereby enabling the extraction of richer features even in few-shot scenarios. Within the meta-learning framework, the time domain vibration signal is mapped to the frequency domain by Fast Fourier Transform (FFT) to obtain the amplitude spectrum, and a convolutional neural network is used to extract the multi-scale discriminative embeddings. The class prototypes are then computed by taking the arithmetic mean over the support samples. Next, a learnable diagonal covariance matrix and a global temperature parameter are introduced to enable adaptive feature reweighting and flexible decision boundary adjustment. Furthermore, a distance-ratio based threshold is adopted to select high-confidence pseudo labeled samples, and a weighted combination of labeled and pseudo-labeled losses is optimized jointly, allowing the network parameters, covariance matrices, and temperature parameter to be updated in a unified manner. Experimental results on two bearing datasets demonstrate that the proposed method significantly outperforms traditional Euclidean-metric-based meta-learning baselines in semi-supervised scenarios under multiple working conditions.

Keywords: Deep learning and machine learning, Artificial intelligence, Computational intelligence
Abstract: Font generation aims to synthesize new fonts from a limited number of reference characters, thereby reducing the labor cost of manual font design. This task remains challenging, especially for Chinese characters, because font styles exhibit high diversity and complex structures. The GB2312 standard contains 6,763 Chinese characters, including 3,755 commonly used characters, 3,088 less frequent characters, and 284 common radicals. This large character space makes structural modeling difficult. Although font generation can be viewed as an image generation task, fonts contain radicals, spatial layouts, stroke thickness variations, and sharp stroke details. These properties make font synthesis more difficult than ordinary natural image generation. As a result, existing methods often produce missing strokes, blurred details, shape distortions, and style drift. To address these problems, we propose FG-Font, a diffusion-based font generation framework built on a U-Net architecture. Without changing the denoising backbone, FG-Font improves content modeling, style alignment, and spatial consistency. Specifically, we design a Structure-aware Content Encoder (SCE) for structural representation, a FlowRSI module for optical-flow-based residual style alignment, and a Style Contrastive Module (SCM) for more discriminative style control. Extensive experiments demonstrate the effectiveness of the proposed method.

Keywords: Autonomous vehicles, Nonlinear control and applications, Control theories
Abstract: This paper presents a Gaussian process regression (GPR) assisted model predictive control framework for safe landing of an unmanned aerial vehicle (UAV) on moving platform. The UAV has known dynamics, while the motion of the landing platform is uncertain and time varying. A GPR is employed solely for predicting the future trajectory of the platform using observed position and velocity data, without modeling its full dynamics. The Gaussian process (GP) provides short horizon future position estimates together with prediction uncertainty, and these predictions are incorporated into the model predictive control optimization to enable anticipatory and robust landing control. Simulation studies show that the proposed approach achieves smooth, accurate, and reliable landing performance even under irregular and unpredictable platform motion.

Keywords: Autonomous vehicles, Nonlinear control and applications, Robotics and swarm intelligence
Abstract: This paper proposes a nonlinear guidance strategy capable of intercepting a constant-velocity, non-maneuvering target while strictly satisfying the prescribed bounds on the control input (commanded acceleration). Unlike conventional strategies that estimate time-to-go using linearization or small-angle approximations, the proposed strategy employs true proportional-navigation guidance (TPNG) as a baseline, which utilizes an exact time-to-go formulation and is applicable over a wide range of target motions. In contrast to most existing strategies, which do not incorporate control input bounds into the guidance design, the proposed approach explicitly accounts for these limits by modeling the interceptor's acceleration as a dynamic variable. Based on the sliding mode control technique, an effective guidance law that achieves time-constrained interception while accounting for bounded input is then derived. The performance of the proposed strategy is evaluated for various engagement scenarios.

Keywords: Autonomous vehicles, Nonlinear control and applications, System identification and modelling
Abstract: This paper proposes a blind-spot area detection approach for an Automated Guided Vehi- cle (AGV) using a corner-mounted multi-ultrasonic array and a bird’s-eye (top-down) monitoring con- cept implemented in MATLAB/Simulink. The sys- tem is evaluated through a scenario-based open- loop simulation in which a Scenario Reader pro- vides AGV pose, lane boundaries, and dynamic ac- tors, while dedicated Detection Generator blocks emulate radar, camera, LiDAR, and an ultrasonic short-range safety subsystem. The main contribu- tion is a 6×3 (azimuth×elevation) beam matrix at each body corner (FL/FR/RL/RR) together with a geometry-based visibility model, discretized range classes (Near / Medium / Far), and rising-edge event logs. Simulation results for forward-corridor and right-turn maneuvers show that multi-ultrasonic sen- sor fusion improves near-field obstacle awareness in typical blind-spot regions, particularly during side-clearance and turning conditions. The proposed framework provides a reproducible Simulink baseline for further algorithm development and hardware-in- the-loop validation.

Keywords: Autonomous vehicles, Robotics and swarm intelligence
Abstract: This paper presents a new heuristic strategy for a sampling-based autonomous path-planning algorithm for single autonomous mobile robots (AMRs) that uses a flexible step size to handle workspace complexities. A model of auxiliary framework inspired by the geometric structure of Bohr’s atomic model is introduced. The model consists of a set of concentric orbits centered at start or target points. The generated path is modelled as a chain that is constructed by a series of waypoints that are arranged in a bidirectional fashion, i.e., from both the orbit centers. The waypoints are generated and placed on the orbits rather than at a fixed position, as is typical. The von Mises distribution is used to orient generated waypoints to reduce placement randomness. The performance of the proposed method is evaluated based on processing time and precision. Simulations showed that the proposed method can perform path generation with short, precise processing time, and that the method's success rate in complex workspaces can be increased by adjusting the concentration rate of the von Mises randomization parameter.

Keywords: Autonomous vehicles, Robotics and swarm intelligence, Mechatronics
Abstract: Reliable position estimation is essential for Autonomous Mobile Robots (AMRs) operating in indoor environments. Monocular camera navigation offers a lightweight sensing solution for such systems. Classical visual odometry based on optical flow can estimate relative robot motion efficiently. However, this approach often suffers from cumulative drift and the absence of global environmental references. This study proposes a lightweight monocular visual navigation framework. The framework integrates pseudotrajectory estimation derived from the Lucas–Kanade optical flow method. Feature tracking is performed using Shi–Tomasi corner selection to improve tracking stability. Environmental characteristics are represented using texture features extracted through the Gray-Level Co-occurrence Matrix (GLCM). The extracted texture features are mapped into an RGB representation to construct a colored trajectory. The resulting trajectory encodes both geometric motion and spatial texture characteristics of the environment. Experiments were conducted using an Autonomous Mobile Robot in a warehouse like environment. The results show that colored trajectory patterns consistently correspond to different environmental sectors. The proposed approach supports monocular topological localization with low computational complexity suitable for edge computing platforms.

Keywords: Autonomous vehicles, Nonlinear control and applications, Robotics and swarm intelligence
Abstract: This paper addresses inter-sample collision avoidance in multi-agent systems using Model Predictive Control (MPC) and Control Barrier Functions (CBFs). Conventional MPC-CBF methods guarantee safety only at sampling instants. To address this issue, we propose an enhanced MPC framework using a robust control barrier function (RCBF) with a robustness term that compensates for inter-sample effects. The proposed method is applied to a formation control problem in which multiple agents avoid static and dynamic obstacles while maintaining a desired formation. Simulation results using wheeled mobile robots demonstrate that the proposed approach successfully guarantees collision avoidance throughout inter-sample intervals.

Keywords: Robotics and swarm intelligence, Mechatronics, Autonomous vehicles
Abstract: This paper presents a distributed MPC-based leader–follower framework for a heterogeneous robotic pair composed of a differential-drive ground robot and a quadrotor UAV. The leader performs cyclic way-point navigation, while the follower tracks a desired formation point behind the leader using only the most recently received leader information. To represent networked operation, the communication layer includes delay, jitter, packet loss, and measurement noise. Separate MPC controllers are designed according to the distinct dynamics and constraints of the two platforms. Simulation results show successful leader way-point progression and bounded follower tracking behavior in the nominal case. Under communication impairments, the follower exhibits larger transient tracking error and relative-distance variability, but the closed-loop response remains bounded within the tested conditions.

Keywords: Nonlinear control and applications, Autonomous vehicles, Control theories
Abstract: By introducing two polar coordinates transformations, the marine vessel's original two-input-three-output second-order tracking model can be reduced to a two-input-two-output feedback form. However, the resulting system does not confirm to the strict-feedback structure, leading to potential singularity when designing the stabilizing function for the virtual input in the recursive controller design. Moreover, the polar coordinate transformation itself inherently introduces singularities. To address these singularity issues, this paper employs a control barrier function (CBF) based approach and formulates the trajectory tracking problem as a quadratic program (QP) solved via a QP optimizer. Numerical simulations are carried out to demonstrate the effectiveness of the proposed method.

Keywords: Deep learning and machine learning, Artificial intelligence
Abstract: In complex maritime environments, small object lead to missed detections and background misclassification. To address this issue, this paper proposes YOLO-SS, an improved YOLOv11-based object detection and recognition algorithm. Firstly, a spaceto- depth (SPD) module is introduced into the backbone network to replace traditional convolutional downsampling, thereby preserving spatial information of small objects. Secondly, a lightweight Slim-Neck structure based on GSConv is adopted to reduce computational redundancy and enhance multi-scale feature fusion. Finally, a SimAM is incorporated to enhance the model's feature learning capability without increasing the amount of parameters. Experimental results demonstrate that YOLO-SS achieves a mAP of 97.4% on the self-constructed maritime dataset. Additionally, YOLO-SS achieves a mAP of 36.4% on the VisDrone dataset with an improvement of 3.1% over the baseline.

Keywords: Deep learning and machine learning, Artificial intelligence, Industrial applications
Abstract: Efficient retrosynthetic planning requires not only accuracy but also the diversity of proposed routes. However, current confidence-driven search algorithms often suffer from local redundancy, yielding highly similar pathways that limit practical utility. To address this, we propose a representation learning framework coupled with a dual-stage diversity optimization strategy. First, we introduce Hierarchical Contrastive Learning (HCL) to map reaction difference fingerprints into a continuous latent space. Guided by an expert-curated reaction hierarchy, HCL captures graded semantic similarities between chemical transformations. Second, we leverage this learned similarity metric to intervene in the planning pipeline via two mechanisms: an In-search Similarity-aware Reranking (SR) to dynamically guide node expansion, and a Post-search Path Filtering (PF) strategy to extract a concise, non-redundant subset of successful routes. Evaluations on standard benchmarks demonstrate that our method achieves a more favorable trade-off between top-k accuracy and route-level diversity compared to existing baselines.

Keywords: Deep learning and machine learning, Computational intelligence, Industrial applications
Abstract: This paper addresses critical limitations in current industrial fault diagnosis methods, specifically their insufficient incorporation of domain knowledge and the modal incompatibility between large language models (LLMs) and industrial time series data. Alignment-based Time series and Text Large Language Model (ATTLLM) for fault diagnosis is proposed. It is a novel framework that integrates domain knowledge with temporal process data through three key innovations. First, domain-knowledge-driven prompt templates enable LLMs to effectively interpret time series patterns. Second, an alignment layer facilitates cross-modal fusion between time series representations and fault attribute descriptions. Third, LoRA-based fine-tuning recasts fault classification as a text generation task, leveraging the generative capabilities of LLMs. Experimental validation on the TEP and CSTR demonstrates that ATTLLM substantially outperforms existing state-of-the-art approaches, confirming its efficacy in delivering accurate and interpretable fault diagnosis for complex industrial systems.

Keywords: Deep learning and machine learning, Health systems, Measurement and instrumentation
Abstract: Left Ventricular Hypertrophy (LVH) is a significant independent risk factor for major adverse cardiovascular events. However, current non-invasive diagnostic modalities often fail to achieve an optimal balance between sensitivity, specificity, and accessibility. Methods: This study proposes a novel diagnostic framework utilizing Spin-Exchange Relaxation-Free (SERF) Magnetocardiography (MCG) combined with machine learning algorithms. We conducted a retrospective analysis of 312 subjects (198 LVH patients and 114 healthy controls). A comprehensive set of 55 multi-domain features — encompassing time-domain metrics, frequency-domain spectral entropy, and radiomic texture features from magnetic field maps — was selected using the Boruta algorithm. Six machine learning classifiers, including Support Vector Machine (SVM) and Random Forest, were evaluated for diagnostic performance. Results: The SVM model demonstrated superior efficacy, achieving an accuracy of 92.6%, a sensitivity of 96.6%, and an Area Under the Curve (AUC) of 0.981.SHAP-based feature interpretation indicated that MCG-specific markers, such as magnetic field compactness and T-wave spectral entropy, successfully capture the myocardial fiber disarray and electrophysiological heterogeneity associated with LVH. Conclusion: These findings validate that high-resolution MCG, augmented by multi-domain feature fusion, significantly outperforms traditional methods in detecting LVH, offering a promising tool for early, non-invasive subclinical screening.

Keywords: Deep learning and machine learning, Health systems, Measurement and instrumentation
Abstract: Reliance on single-sensor modalities often limits the robustness and sensitivity of diagnostic systems due to the lack of complementary information. To address this, this paper proposes a multi-modal sensor fusion framework for the non-invasive detection of Valvular Heart Disease (VHD). We integrate heterogeneous data sources by combining the high-precision anatomical parameters from Echocardiography with the high-dimensional functional features derived from Spin-Exchange Relaxation- Free (SERF) Magnetocardiography (MCG). A unified multimodal feature space is constructed to fuse structural specificity with functional electrophysiological sensitivity at the feature level. The proposed approach is validated on a dataset of 699 subjects using machine learning classifiers. Experimental results demonstrate that the fusion strategy significantly outperforms single-modality methods, with the XGBoost-based fusion model achieving an accuracy of 93.6% and an AUC of 0.975. This study confirms that multi-sensor fusion effectively enhances the decision boundary and provides a robust, high-precision objective basis for intelligent medical diagnosis systems.

Keywords: Deep learning and machine learning, System identification and modelling
Abstract: The lack of prior knowledge regarding target dynamics poses a significant challenge for accurate maneuvering target tracking. This paper presents a multi-output Gaussian process (MOGP)-based tracking framework designed to learn unknown dynamics online. Unlike traditional independent Gaussian Process models, the proposed MOGP State-Space Model (MOGP-SSM) explicitly models the correlations among output dimensions, thereby enhancing fitting accuracy. To ensure real-time feasibility, the framework incorporates a sliding window mechanism and sparse inducing points, significantly reducing computational complexity. The target state estimation is formulated as a maximum a posteriori (MAP) problem. Addressing the challenge of intractable gradients in the complex posterior distribution, a derivative-free particle swarm optimization (PSO) algorithm is employed to obtain the optimal state estimate. Simulation experiments verify that the proposed algorithm provides superior tracking accuracy and robustness compared to state-of-the-art methods, especially when the target motion model is unknown.

Keywords: Control theories, Deep learning and machine learning, Intelligent control
Abstract: Torque ripple in brushless DC (BLDC) motors degrades speed regulation and generates acoustic noise in closed-loop drives. Adaptive Harmonic Cancellation (AHC) based on the Least Mean Squares (LMS) algorithm can suppress periodic ripple without motor model knowledge, but its performance is critically governed by the step-size μ, which cannot simultaneously optimize convergence and steady-state ripple under varying operating conditions. This paper proposes a hierarchical PI,+,AHC,+,TD3 architecture in which a Twin Delayed Deep Deterministic Policy Gradient (TD3) agent jointly optimizes μ and the Exponential Moving Average (EMA) time constant 𝜏_EMA from real-time observations. Validated in MATLAB/Simulink across three load conditions and five speed setpoints (50--250 RPM), the AHC compensator achieves 55-65% RMS ripple reduction over PI-only control, and TD3-based optimization yields a further 24.8-30.7% improvement over the fixed step-size baseline, with a 27.0% average at 200 RPM.

Keywords: Control theories, Deep learning and machine learning, Intelligent control
Abstract: This paper proposes an energy-efficient path planning method for a single drone to measure 2D data. It uses Gaussian Process Regression to model observation uncertainty, such as camera noise. Reinforcement learning then optimizes the flight trajectory to minimize the overall estimation uncertainty.

Keywords: Control theories, Robotics and swarm intelligence
Abstract: This paper proposes a Control Barrier Function (CBF) that utilizes the Zero Moment Point (ZMP) stability criterion to enhance the walking stability of quadruped robots. By defining the ZMP condition as a safe set, we develop a method to mathematically ensure balance during locomotion. Simulation results demonstrate that the proposed CBF-based controller successfully prevents falls under external disturbances, outperforming nominal controllers by maintaining the ZMP within the support polygon.

Keywords: Robotics and swarm intelligence, Computational intelligence, Artificial intelligence
Abstract: Online three-dimensional bin packing (online 3D-BPP) can produce feasible local rearrangement behavior, but its outputs are poor execution interfaces for downstream robot planning because they typically preserve one replay order rather than the dependencies that are actually required. This issue becomes more pronounced when buffers and temporary storage are used, since local replanning may include repeated handling, support-induced precedence, and removal-to-placement clearance constraints. We therefore present an execution-oriented compilation layer that converts each local replanning result into an event-level spatio-temporal dependency graph and then into a primitive-level partial-order pick-and-place graph (PP-DAG), which is exported as a TAMP-facing symbolic-geometric bundle. The resulting interface preserves mandatory temporal, support, and clearance structure while avoiding accidental serialization, and it can be grounded in both single-robot and multi-robot execution settings. Fixed-case evaluations show that PP-DAG improves downstream TAMP behavior over flat replay in both a dependency-rich rearrangement case and a dependency-free control case. These results suggest that dependency-aware compilation is a practical interface between online 3D-BPP and downstream task and motion planning.

Keywords: Robotics and swarm intelligence, Deep learning and machine learning, Nonlinear control and applications
Abstract: Accurate control of soft robots requires balancing compliance and precision. This paper proposes a stiffness modulation strategy for a hybrid-actuated continuum robot that combines tendon-driven and pneumatic antagonistic actuation. To bypass the complex dynamics of continuum robots, we utilize a deep learning approach to derive a data-driven model based on the robot’s tip states and actuation inputs, which are initially generated from a proven analytical model. A data-driven model is integrated into a Deep Neural Network-based Model Predictive Control (DNN-MPC) framework, enabling accurate trajectory tracking and stiffness regulation with reduced computational overhead. The proposed method effectively handles physical constraints, offering a wide dynamic range of stiffness and improved motion precision. Numerical simulations demonstrate that the controller modulates stiffness independently, seamlessly switching between low-force compliant and high-force stiff modes without deviating from positional targets. Furthermore, adjusting the MPC's weighting matrix allows a direct, tunable trade-off between kinematic precision and force regulation based on application priorities. In parallel, a data-driven approach proved reliable, significantly outperforming standard Jacobian-based Nonlinear MPC in computational efficiency and control speed.

Keywords: Robotics and swarm intelligence, Industrial applications, Computational intelligence
Abstract: Collaborative 3D printing with multiple robots can substantially enhance manufacturing efficiency for large-scale, geometrically complex components. However, it also poses significant challenges for path planning, including constraints on task executability, task allocation, and execution sequencing. Existing path planning methods for Collaborative 3D printing typically decompose the problem into separate stages and often oversimplify practical workspace constraints, making it difficult to accurately characterize collaborative printing scenarios with overlapping robot workspaces. To overcome these limitations, this paper proposed a colored traveling salesman problem-based approach to address the Multi-Robot collaborative 3D Printing Routing Problem (MR3DPRP). Specifically, MR3DPRP is formulated as an asymmetric colored traveling salesman problem, which provides an integrated representation of task allocation and route sequencing. Based on this formulation, a K-nearest-neighbor Variable Neighborhood Search (KVNS) is proposed. KVNS incorporates a multi-step insertion neighborhood structure and an adaptive perturbation strategy to improve both search efficiency and solution quality. Experimental results on 20 benchmark instances generated from five 3D models demonstrate that KVNS can effectively solve the MR3DPRP and outperforms the compared state-of-the-art algorithms.

Keywords: Robotics and swarm intelligence, Intelligent control, Deep learning and machine learning
Abstract: We propose a method for visual servocontrol of rigid bodies in the eye-to-hand setting, where the controlled body's movement is observed by a stationary RGB-D camera. The method is based on establishing and maintaining correspondences between keypoints on the body's surface across multiple images of the body in various configurations, and leverages recent advances in keypoint tracking and matching methods based on deep learning. The proposed method is pose-based, using the 3D positions of the keypoints extracted by the depth camera to estimate the relative pose of the body with respect to a reference pose. Furthermore, the method identifies the true configuration space of the controlled body by performing principal component analysis on the computed relative poses over a training sequence and decouples the control loop along each identified configuration variable.

Keywords: Robotics and swarm intelligence, Nonlinear control and applications, Mechatronics
Abstract: Underwater Remotely Operated Vehicle (U-ROVs) is a robotic platform that can be used for various underwater applications. It can also be equipped with state-of-the-art imaging sensor (camera) for data visualization. Producing a blur-free imagery data is essential for underwater visual tasks, due to the unpredictable underwater environmental conditions that can induce disturbances on the U-ROV platform. In this paper, we present a stereo Image-based Visual Servoing (IBVS) controller for an in-house-developed U-ROV platform with an Intel RealSense D455 stereo camera. By augmenting the third row of the interaction matrix with depth features derived from stereo triangulation, the controller resolves the sway--yaw coupling ambiguity inherent in traditional planar IBVS formulations. Then we employ a PI-augmented control law to reduce steady-state error, and project the resulting camera velocity command onto the four controllable degrees of freedom of the vehicle---surge, sway, heave, and yaw---through a camera-to-body frame transformation. Experimental validation in an underwater pool environment demonstrates that the controller successfully drives image feature errors toward zero and stabilises the U-ROV at the desired pose relative to an AprilTag marker under external disturbances.

Keywords: Mechatronics, Robotics and swarm intelligence, Nonlinear control and applications
Abstract: The proposed 6 degrees of freedom (DoF) tendon driven continuum robot consists of two sections, offering an expanded workspace and enhanced dexterity for operation in complex confined environments. However, accurate tip-pose control remains challenging because analytical inverse kinematics is difficult to derive from a mechanical-model-based forward kinematic. To address this issue, this paper presents a control framework combining a mechanical model with Jacobian-based inverse kinematics. The Jacobian is numerically estimated using the forward finite difference method, while the inverse solution is computed using the damped least-squares approach. Simulation results demonstrate accurate tracking of the desired tip position.

Keywords: Control system education, Adaptive systems, Robotics and swarm intelligence
Abstract: Reinforcement Learning and MPC-Based Multi-Robot Safe Shape Formation for Collision Avoidance in Human-Aware Indoor Environment Multi-Robot Systems (MRS) are becoming increasingly important in ensuring safe movement in environments with the coexistence of robots and humans. In this paper, we demonstrate shape formation among three differential drive mobile robots in an indoor industrial setting environment, while ensuring safety through collision avoidance and operational security. We propose a nonlinear learning-based control strategy in which shape formation occurs in a manner that provides safety around a faulty robot. To further enhance safety, a novel metric is introduced for heterogeneous MRS operating in a human-aware environment. It is aimed at preventing collisions both among robots and between robots and humans. Additionally, a comparative analysis of these control strategies, including Non-linear Model Predictive Control (NMPC) and Reinforcement Learning (RL), is conducted to identify the most effective approach for controlling multi-robot systems. From experimental results, position accuracies have been found between 98.73% and 99.82%, with the error values staying small at (1.0846, 1.9406) and (6.0843, -1.0632), which are near their target objectives. The NMPC Controller is superior to the RL-based DDPG controller regarding stability and accuracy. However, for orientation control, the RL technique has been able to perform competitively up to 99.86%.

Keywords: Control theories, Intelligent control, Adaptive systems
Abstract: Networked control systems (NCSs) often execute repetitive tasks over a finite time horizon, exhibiting inherent two-dimensional characteristics in both time and iteration domains. However, conventional control approaches mainly focus on uncertainties along the time axis and fail to fully exploit system repetitiveness, which limits tracking performance. To address this issue, this paper develops iterative learning model predictive control (ILMPC) methods for NCSs. By integrating the advantages of iterative learning control and model predictive control, the proposed approach enables batch-to-batch learning while optimizing transient responses within each iteration, thereby improving learning efficiency and tracking performance. Furthermore, to address the high computational and communication burden of ILMPC in bandwidth-constrained and uncertain networked environments, resource-efficient strategies are developed to reduce resource consumption while maintaining satisfactory control performance.

Keywords: Artificial intelligence, Intelligent control, Robotics and swarm intelligence
Abstract: Autonomous exploration with unmanned aerial vehicles (UAVs) in unknown environments remains challenging due to incomplete maps, limited onboard sensing, and long-horizon decision-making. Reinforcement learning (RL) has been successfully applied to local UAV navigation and obstacle avoidance, but purely learning-based approaches often do not scale well to large unknown spaces. This paper proposes a hybrid exploration framework that decouples global frontier planning from local RL-based navigation. Frontier-based goal generation operates on an online occupancy map built from onboard depth sensing. A reachability-aware filter restricts candidate goals to the connected free-space component that contains the current UAV pose, and an information-gain-based scoring function ranks candidates by jointly considering exploration benefit and navigation cost. The selected goals are executed by a goal-conditioned RL policy equipped with an external safety shield based on optimal reciprocal collision avoidance (ORCA). This combination of reachability-aware frontier filtering, information-gain-based scoring, and safe RL control improves exploration coverage and substantially reduces navigation failures caused by unreachable or unsafe targets.

Keywords: Adaptive systems, Intelligent control, Nonlinear control and applications
Abstract: Train tracking control is a core technology in automatic train operation (ATO). While traditional control methods have achieved high-precision tracking performance, existing schemes cannot dynamically accommodate temporary speed restrictions (TSRs) that arise from adverse weather or signaling failures. This paper proposes a novel control scheme that employs online constraint planning to ensure speed performance under such conditions. First, a method for the online synthesis of train speed constraints is introduced, generating a speed function that simultaneously satisfies TSRs and other safety or performance requirements. Next, a barrier function-inspired approach is employed to enforce the controller to maintain train speed within the upper and lower bounds of these synthetic constraints. Finally, simulation results validate the effectiveness of the proposed control scheme.

Keywords: Artificial intelligence, Robotics and swarm intelligence, Intelligent control
Abstract: Efficient robot autonomous exploration in unknown and cluttered environments is a challenging problem due to partial observability and the trade-off between information acquisition and cost for motion under real-time constraints. To address this problem, Global Entropy-based Attentive Target selection (GEATs) is proposed by selecting consecutive targets for path planning based on reinforcement learning, with predicting the entropy-distribution of unknown regions on the map. The entropy map is derived by a lightweight network, UtilityNet, and from which feasible target candidates are extracted via multi-scale Non-Maximum Suppression (NMS). An Actor Network in Proximal Policy Optimization (PPO) integrated with multi-head-attention, selects target from the candidates. Simulation results in randomized multi-room environments show that the proposed GEATs improves exploration efficiency and training stability compared with learning-based baselines, and ablation studies verify the benefit of the multi-head-attention design.

Keywords: Control theories, Adaptive systems
Abstract: This paper proposes an adaptive tracking control method for a class of linear systems with input delays. Although existing delay compensation strategies can reduce the influence of delay on system stability, they can not give a critical delay. If input delay is greater than the critical delay, the system may become unstable. To solve this issue, an adaptive tracking controller with delay compensation performance is proposed, by which the critical delay is always greater than the input delay. The stability of the closed loop system is proved in the Lyapunov sense, and the convergence of tracking error is also analyzed. Simulation results show the effectiveness of the proposed scheme.

Keywords: System identification and modelling, Control theories, Nonlinear control and applications
Abstract: This paper investigates the identification problem for finite impulse response (FIR) systems with binary-valued observations under event-triggered communication mechanism and data packet dropout. The challenge lies in the inability to distinguish between untriggered events and packet dropout when no information is received, which prevents us from obtaining the statistical properties of the binary-valued sequence. A compensation-oriented difference-driven identification (CODD) algorithm is proposed to estimate the parameter by recovering the mean of the original binary-valued sequence, where different values for the observation estimates are assigned when receiving 0, 1 or nothing. The almost sure convergence and the asymptotic normality of the CODD algorithm are established when data packet dropout probability is less than frac{1}{2}. By calculating the communication rate, it is proven that the difference-driven mechanism can save 50% of the communication cost compared to original binary-valued systems. Furthermore, when data packet dropout probability is high, an m-channel compensation-oriented identification (m-CODD) algorithm is constructed by utilizing retransmission of the each observation for m times, which is designed based on the packet dropout probability. The properties of m-CODD algorithm including convergence, asymptotic normality and communication rate are established. Numerical simulations are illustrated to show the theoretical results.

Keywords: Control theories, Nonlinear control and applications, Adaptive systems
Abstract: This paper presents the high-order sliding mode control (HOSMC) design for a class of nonlinear systems with time-varying parameters. It is noteworthy that the sliding dynamics are influenced by time-varying parameters, with only their variation range being available. The novel congelation-like practical finite-time adaptive high-order sliding mode control (C-PFTAHOSMC) law is proposed using the tools of congelation of variables and adding a power integrator. Gain overestimation in the literature is mitigated by designing the adaptive switching gain. Thus, the chattering is effectively mitigated. Moreover, the sliding variable can be stabilized to a small neighborhood of origin in a finite time. Finally, the feasibility of the proposed algorithm is validated through an illustrative example.

Keywords: Control theories, Intelligent control
Abstract: This paper addresses the fault estimation and fault-tolerant bipartite output consensus problem for heterogeneous multi-agent systems with competitive interactions under actuator faults and aperiodic denial-of-service attacks. A distributed observer with adjustable parameters is designed for simultaneous estimation of system states and faults while satisfying H∞ perfor￾mance and pole placement constraints. Using the estimated fault information, an output regulation-based distributed controller is developed to achieve bipartite output consensus. Closed-loop stability under communication interruptions is ensured through a piecewise Lyapunov analysis that explicitly links attack duration to convergence rates. The results of a numerical example confirm the effectiveness of the theoretical results.

Keywords: Control theories, Control devices, sensors and actuators
Abstract: The paper investigates state perturbations on Boolean networks (BNs) by employing the semi-tensor product (STP) of matrices and the discrete derivative method. To characterize the effect of state perturbations, the definition of diffusion effects is introduced. Instead of directly analyzing the state transition matrix—which often entails high computational cost—several criteria for diffusion effects are established using the incidence matrix and the node-connection diagram, significantly reducing computational complexity. Furthermore, a discrete derivative approach is adopted to construct an n-order Boolean matrix, where n denotes the number of nodes of in the BNs, enabling the analysis of diffusion effects under a perturbation scale of r=1.

Keywords: Control theories, Nonlinear control and applications, Communication
Abstract: For AC islanded microgrids, facing complex uncertain factors such as fluctuations in renewable energy sources and changes in loads, achieving precise and efficient control has become a challenge. In this work, we put forward two novel secondary control approaches with distributed fixed-time characteristics for addressing uncertain conditions such as renewable fluctuations and load changes, achieving fast and accurate restoration of frequency and voltage along with fair power sharing. By employing Lyapunov stability theory, system’s stability under this control strategy is rigorously proven. The advantage of this strategy is that an explicit bound on the system's convergence time can be accurately obtained merely by adjusting controller parameters, regardless of its starting conditions. This control strategy also substantially reduces the convergence time compared to prior research. Simulation experiments have thoroughly validated the proposed strategy, proving its effectiveness and feasibility for real-world use.

Keywords: Cyber-physical systems and security, Control theories, Communication
Abstract: This paper addresses the challenges of protecting the initial states of multi-agent systems against eavesdropping attacks, while achieving resilient consensus under asynchronous denial-of-service (DoS) attacks. Firstly, we introduce a flexible DoS attack model, where each adversary independently targets a single edge in the communication topology. This model extends the conventional synchronous DoS attack scenario to a more general and asynchronous setting. Subsequently, we extend the existing state-decomposition-based privacy-preserving approach to multi-agent systems subject to such asynchronous attacks, and propose a unified communication algorithm, which is designed to handle the complex scenario of both eavesdropping and DoS attacks. Further, we design a novel hybrid controller and establish relaxed sufficient conditions that effectively guarantee both privacy preservation and consensus convergence. Finally, a numerical example is presented to validate the effectiveness of the proposed method.

Keywords: Control theories, Nonlinear control and applications
Abstract: Most existing results on Boolean networks assume full knowledge of the logical update functions. However, in practical scenarios, these functions are not always available or easy to identify. To solve this issue, in this paper, a new data-driven framework is proposed for controller design of Boolean networks, without requiring prior knowledge of the network structure. A key advantage is its capability to handle various state, input, or performance constraints that commonly arise in practical control tasks while ensuring stabilization. Unlike linear systems, generating sufficient data for Boolean networks presents unique challenges. To address this, an implementable method is developed to construct suitable input sequences that guarantee data richness for controller design. The advantage and applicability of our framework are demonstrated through two representative scenarios, namely stabilization under temporal tasks and stabilization with optimal control, for illustration.

Keywords: System identification and modelling, Control theories, Cyber-physical systems and security
Abstract: In practical information transmission, observation signals received by estimators often deviate from true system outputs due to channel impairments and data loss, leading to distorted observations that degrade estimation performance. This paper presents a novel theoretical framework for joint state and mode estimation in switched Boolean control networks (SBCNs) under such distorted observations. Leveraging the algebraic state-space representation, we first derive an equivalent algebraic form of SBCNs subject to observation distortion, wherein the distortion operator is characterized as a Boolean negation function. A real-time distortion detection and correction mechanism is then developed, which identifies distorted observations by evaluating the divergence between forward predicted and updated state distributions, and subsequently rectifies the observations using historical information to improve estimator robustness. Moreover, we propose a dynamic programming-based estimation algorithm that computes globally optimal trajectories of both system states and switching modes over a finite horizon, addressing the inherent limitation of traditional Bayesian filters that achieve only local optimality. Finally, the effectiveness and advantages of the proposed framework are validated through its application to the p53-MDM2 regulatory network.

Keywords: Control theories, Nonlinear control and applications, Motion and vibration control
Abstract: Tube-based robust model predictive control (RMPC) is an effective approach for controlling constrained systems subject to uncertainties. Since future states cannot be accurately predicted due to disturbances, they are constrained to reside within well-constructed tubes. Depending on the characterization of the tubes, three widely-used RMPC algorithms—namely, varying-tube, rigid-tube, and homothetic-tube RMPC—are investigated. This paper presents the implementation of these three methods on a linear time-invariant system and provides a comparative analysis, with a particular focus on the underlying invariant sets and their respective domains of attraction. The theoretical findings are validated through numerical simulations.

Keywords: Robotics and swarm intelligence, Intelligent control
Abstract: As a variant of Model Predictive Control (MPC), Model Predictive Path Integral (MPPI) is a sampling-based method that enables effective optimization of nonlinear systems. However, when standard MPPI is trapped in a local minimum, it tries to find an alternative way by exploring unnecessary space or even hitting the obstacles without guidance. To address this limitation, Voronoi--MPPI is proposed to incorporate global topological guidance derived from a Voronoi diagram into the MPPI framework, and it mitigates the short-sighted behavior of standard MPPI, prevents convergence to local minima, and improves exploration efficiency in cluttered and challenging environments. The method is validated in three representative environments, demonstrating lower cost and higher success rates compared to the traditional MPPI baseline.

Keywords: Nonlinear control and applications, Intelligent control
Abstract: This paper investigates an optimal cooperative differential game guidance problem for multi-aircraft swarms intercepting maneuvering targets under input constraints and directed topologies. Based on the leader-following cooperative guidance architecture, a high-order nonlinear leader-following guidance model is established, and an intelligent cooperative optimal guidance strategy is developed in terms of integrating cooperative theory and differential game guidance. Further, to handle unknown information such as target maneuvers during guidance processes, an adaptive extended state observer is constructed for an accurate estimation. In addition, a critic neural network is built via the adaptive dynamic programming (ADP) algorithm so as to provide a self-learning online solution to solve the studied problem. By virtue of Lyapunov stability theory and the invariance principle, the stability of the closed-loop system and the convergence of the critic network weights are rigorously proven. Finally, comparative simulation results validate the effectiveness of the proposed method.

Keywords: Autonomous vehicles, Nonlinear control and applications, Control theories
Abstract: This paper investigates the optimal safety control problem of nonlinear control systems by proposing novel high-order control barrier functions (HOCBFs). Different from zeroing HOCBFs, two novel HOCBFs are derived and the safety controllers are designed in an explicit way. Next, we implement vector Lyapunov function approach to propose a novel high-order control Lyapunov function (HOCLF) for the stabilization control problem. The relations between the proposed and existing HOCBFs are discussed. Afterwards, the compatibility of the proposed HOCLF and HOCBF is addressed to guarantee the stabilization and safety control objectives simultaneously, and thus the optimal controller is established. Finally, a numerical example from the navigation problem of quadrotors is presented to illustrate the efficacy of the derived results.

Keywords: Intelligent control, Control system education, Artificial intelligence
Abstract: We consider the problem of approximating model predictive control (MPC) laws using neural networks (NNs). A learning-based explicit MPC approach is proposed with a dual-mode scheme which leverages terminal ingredients of MPC. Given a data set, such a dual-mode scheme produces a supervised control law that bears inherent robustness against approximation errors. To further ensure strict constraint satisfaction, a safety governor (SG) is constructed by formulating an optimization problem where the control input approaches the supervised control law subject to additional constraints generated from the augmented maximal constraint feasible set. With this safety governor, recursive feasibility can be formally guaranteed. We finally demonstrate the proposed approach on two numerical examples in terms of both control performance and online computation time.

Keywords: Energy Systems, System identification and modelling
Abstract: Against the background of high renewable energy penetration and frequent typhoon disasters threatening power grid security, this paper proposes a two-stage planning method for battery-hydrogen hybrid energy storage with multi-time scale characteristics. Firstly, the characteristics of multi-time scale energy storage are analyzed, and the models of battery energy storage and hydrogen energy storage are established. Secondly, based on the analysis of extreme weather random scenarios, a bi-level stochastic programming model for multi-energy storage aimed at enhancing the resilience of power systems is constructed. Finally, case verification is carried out based on the modified IEEE RTS-79 node system. The results show that the proposed method can give full play to the complementary advantages of multi-time scale energy storage, which achieves dual improvement of the economy and resilience level of the power system.

Keywords: Energy Systems
Abstract: This article proposes a data-driven framework to verify the distributed conditions that guarantee the system-wide stability for interconnected power systems. To guarantee systemwide stability, the dynamics of each bus are required to satisfy an output differential passivity (ODP) condition with a sufficient index. These ODP indices uniformly quantify the impacts on the system-wide stability of individual bus dynamics and the coupling strength from the power network. To obtain these indices without explicit physical models, we derive a data-driven linear matrix inequality (LMI) criterion based exclusively on measured inputstate trajectories. Furthermore, extracting the optimal ODP index is formulated as a convex semi-definite programming (SDP) problem. Simulations verify the effectiveness of the proposed method under both single-device offline evaluation and systemwide online certification scenarios.

Keywords: Deep learning and machine learning, Energy Systems, Control theories
Abstract: Electricity--hydrogen integrated energy systems offer an effective pathway for renewable energy accommodation and flexible multi-energy operation. However, renewable generation and multi-energy load uncertainties pose significant challenges to economic and reliable dispatch. Conventional stochastic programming often suffers from a high computational burden when accurate recourse evaluation is required under large-scale scenarios. To address this issue, this paper proposes an uncertainty-aware stochastic dispatch framework based on neural recourse approximation. A two-stage stochastic dispatch model is formulated to coordinate electricity, hydrogen, and thermal energy flows. The first stage determines day-ahead procurement, storage, conversion, and reserve decisions, while the second stage adjusts energy transactions, device operation, and load shedding after uncertainty realization. A deep neural network is then trained to approximate the expected recourse cost from day-ahead decisions and uncertainty scenarios, avoiding repeated solutions of second-stage subproblems. The trained network is equivalently embedded into the day-ahead model, transforming the stochastic program into a deterministic mixed-integer program solvable by standard solvers. By decoupling scenario representation from dispatch decision modeling, the proposed method reduces embedding complexity while retaining adaptability to finite scenario sets.

Keywords: Industrial applications, Intelligent control, Nonlinear control and applications
Abstract: This paper proposes a remote fault diagnosis system based on multimodal data fusion to address the problems of high communication delay, limited monitoring capability, and difficulty in fully grasping the true operating status of the existing high-power AC-AC variable frequency drive control system. The system utilizes a secure and encrypted remote data transmission mechanism, combined with a secondary development data preprocessing process, to achieve real-time collection and analysis of multidimensional operational data such as current, voltage, and speed of motors on engineering sites. In the research of fault diagnosis methods, this paper innovatively proposes a working condition recognition algorithm based on dual sliding windows and kurtosis mutation, as well as a multimodal evaluation algorithm that integrates time series similarity and image structure similarity, to achieve state monitoring of motor no-load and loaded working conditions and identification of system anomalies. In the actual deployment and application of a steel plant, the system runs stably without false alarms, significantly improving the efficiency of on-site inspection and fault response speed, reducing operation and maintenance costs, and providing a feasible technical solution for intelligent fault diagnosis of high-power AC/AC variable frequency transmission systems

Keywords: Control theories, Intelligent control, Robotics and swarm intelligence
Abstract: This paper addresses collaborative localization in multi-agent networks under malicious measurements. A pre-correction framework based on distance rigidity is proposed, integrating hierarchical geometric constraint verification to identify and exclude malicious data. A coverage-constrained replacement strategy reconstructs the topology while preserving rigidity. The resulting distributed algorithm achieves autonomous error detection and topology recovery, maintaining localization accuracy in adversarial environments. Simulations validate the effectiveness of the approach.

Keywords: Intelligent control, Nonlinear control and applications, Industrial applications
Abstract: This paper proposes an intelligent shaft speed control strategy for heavy-duty gas turbines (HDGTs) to enhance grid frequency regulation capability. First, a dynamic model is established via the volumetric method. Subsequently, an error-based active disturbance rejection control (EADRC) is designed, integrating a finite-time extended state observer (FTESO) for rapid disturbance estimation and a barrier Lyapunov function (BLF) for state constraints. The uniform ultimate boundedness of the closed-loop system is proven through Lyapunov analysis. Furthermore, an improved sinh cosh optimizer (ISCHO) is developed for efficient controller parameter tuning. Extensive simulation tests are conducted to demonstrate the effectiveness and superiority of the proposed strategy.

Keywords: Industrial applications, Big data
Abstract: Accurate excavator working condition recognition plays a critical role in improving operational efficiency, reducing energy consumption, and ensuring safety in complex construction environments. However, this task is challenged by inherently uneven inter-condition differences ranging from coarse distinctions to subtle variations induced by working tasks and objects. To address these challenges, this paper proposes a coarse-to-fine Omni-Scale Hierarchical Classification (OSHC) model. Omni-scale one-dimensional convolutions are employed to extract multi-scale temporal features without manual kernel scale selection, enabling compact yet expressive representation learning. On this basis, a coarse-to-fine hierarchical classification strategy is adopted to progressively discriminate working conditions by first separating categories with large inter-class differences and then refining distinctions among similar conditions. Experimental results on real-world excavator datasets demonstrate that OSHC achieves improved accuracy and robustness compared with convolutional classification approaches, validating its effectiveness for practical excavator working condition recognition.

Keywords: Industrial applications, Intelligent control, Deep learning and machine learning
Abstract: The thickening--dewatering process is a critical solid--liquid separation stage in mineral processing and a major source of energy consumption. Due to strong inter-unit coupling, strict safety constraints, and variable task horizons, collaborative scheduling under tiered electricity pricing becomes a challenging long-horizon decision-making problem. To address the limitations of model-based optimization and the deployment difficulty of online reinforcement learning, this study proposes a Hierarchical Temporal-Planning Decision Transformer (HTPDT) framework. The problem is reformulated as an offline hierarchical sequence-modeling task based on historical industrial trajectories. A high-level Planner predicts temporal segmentation boundaries according to system states and remaining tasks, while a low-level Decision Transformer-based Executor generates control actions within each segment. This decomposition converts the original long-horizon problem into manageable short-horizon subproblems. A reward function is designed to jointly capture task completion, safety, and electricity cost, and a rolling execution mechanism is introduced for online scheduling. Experimental results show that the proposed method significantly outperforms baseline approaches and achieves performance close to manual operation.

Keywords: Intelligent control, Artificial intelligence, Industrial applications
Abstract: Rural building clusters are often located in remote areas and remain highly dependent on conventional power supply from weak local utility grids, which makes their decarbonisation challenging. Local microgrids can increase self-sufficiency, while coordinated scheduling across clusters can further improve the utilisation of distributed renewables and storage. This paper investigates a multi-agent reinforcement learning approach for energy management of rural building clusters, where each cluster is equipped with photovoltaic generation and a battery energy storage system. A Multi-Agent Proximal Policy Optimisation framework is employed, in which each microgrid is controlled by an agent. Two operating scenarios are considered in a case study of rural building clusters in the UK, depending on whether inter-microgrid energy sharing is enabled. Comparative results show that energy sharing improves the total reward by 10.46% on the training dataset compared with independent operation, and reduces total energy curtailment by 24.36% on test data. These results suggest that effective scheduling with energy sharing among microgrids can reduce operating costs and improve renewable energy utilisation, which provides a promising pathway for sustainable rural development.

Keywords: Cyber-physical systems and security, Deep learning and machine learning
Abstract: Hardware performance counter (HPC) has been used for malware detection because of its lightweight overhead. Unfortunately, such detection models including their software implementation and hardware implementation suffer from performance decline in environments where resource competition among programs exists in the operating system. In this paper, we propose a framework called CarePlus, which enables two kinds of implementations of hardware performance counter-based malware detection (HMD) resilient to resource competition. The core idea is to detect malware by using invariants extracted from HPC-level behaviors in resource competition environments. To achieve this, a benchmark-based resource pressure generator is designed to generate controlled resource competition environments for observing typical HPC-level behavior of a program. Then, a behavior representation network is trained to map HPC-level behaviors in any competition environment into low-dimensional representations, which allows software-implemented HMD models built on them to detect malware regardless of resource competition. Finally, an adapter is introduced to project behavior representation in the competition environment into data that conforms to the training distribution of the hardware-implemented HMD model.the experimental results show that CarePlus is helpful to improve the performance of software-implemented HMD classifier and hardware-implemented HMD classifier in the competition environment.

Keywords: Control theories, Cyber-physical systems and security, System identification and modelling
Abstract: This paper investigates data-driven model reduction of networked dynamical systems via clustering. We consider the setting where the exact state-space model is unknown and only system data are available, while the system is assumed to possess an underlying network structure. Such systems typically exhibit semistability, which prevents the direct use of standard controllability and observability Gramians. To address this challenge, we propose the concept of data informativity to construct semistable Gramians directly from data, which are then used to define a dissimilarity measure between nodes, capturing their dynamical roles within the network. Based on this measure, nodes are clustered to obtain a reduced-order network representation that preserves essential structural and dynamical properties. The effectiveness of the proposed approach is demonstrated using data from an urban transportation network.

Keywords: Control theories, Nonlinear control and applications, Control devices, sensors and actuators
Abstract: This paper addresses the boundary stabilization problem for a coupled linear Korteweg–de Vries (KdV) system. The model consists of two coupled dispersive equations, with the input applied at the right Neumann boundary and the output given by the left Neumann boundary. First, a full-state boundary feedback controller is designed using a backstepping transformation, ensuring exponential stability of the closed-loop system. Second, an observer is constructed using the boundary measurements, leading to an output feedback control law. Finally, the stability of the combined plant–observer system is established through a Lyapunov analysis.

Keywords: Control theories, Autonomous vehicles, Nonlinear control and applications
Abstract: This paper investigates the stability and control of an Underwater Vehicle-Manipulator System (UVMS) comprising an autonomous underwater vehicle equipped with a two-link flexible-joint manipulator. The system exhibits strongly coupled, nonlinear, and underactuated dynamics resulting from vehicle–manipulator interactions and hydrodynamic effects, rendering controller design inherently challenging. A quasi-Lagrange formulation is employed to derive a comprehensive nonlinear dynamic model of the UVMS. The closed-loop behavior under Proportional–Integral–Derivative (PID) control is then analyzed, and asymptotic stability for trajectory tracking is rigorously established using Lyapunov stability theory. Using the theory of robotic manipulators, suitable Lyapunov candidate functions are constructed to prove boundedness of the system states and convergence of tracking errors under appropriate gain conditions. Numerical simulations are presented to verify the theoretical stability results and to demonstrate the effectiveness of the proposed control framework.

Keywords: Control theories, Nonlinear control and applications, Intelligent control
Abstract: A bicycle is a handy and useful transportation in our daily life. In these years, an autonomous bicycle has been under development. However, such a bicycle is a typical nonlinear system and its control scheme is more difficult than the standard linear systems. In this paper, a control scheme for an autonomous bicycle is considered. Because of its nonlinear system behavior, Takagi-Sugeno(T-S) fuzzy model is employed to describe an original nonlinear model of an autonomous bicycle. A controller for the bicycle is designed for a corresponding T-S fuzzy model which precisely represents an original bicycle model. In our control design, a guaranteed cost control is introduced not only for stability but also a better control performance. To demonstrate the advantages of our control design method proposition, an autonomous bicycle control is illustrated and compared with the previous results.