Keywords: Estimation and Identification, Adaptive Control
Abstract: This paper investigates the stationary target circumnavigation control problem for nonholonomic mobile robot. To this end, an image-based visual servoing framework is proposed, which relies solely on image measurements for feedback control, without any target depth or global position information. Meanwhile, the camera field-of-view (FOV) constraints and collision avoidance constraints are taken into account in the control design. Building on this framework, an adaptive control algorithm is developed to estimate the unknown parameters online, thereby guaranteeing that the robot asymptotically converges to the desired circumnavigation radius. Simulation results validate the effectiveness of the proposed method.

Keywords: Estimation and Identification, Fault Detection and Diagnostics, Modeling and Control of Complex Systems
Abstract: Complex products such as aero-engines are characterized by long manufacturing chains, strong cross-stage coupling, and multi-source heterogeneous data, which often lead to inconsistent product quality, degraded reliability, and high operation and maintenance costs. To address these challenges, this paper proposes a full lifecycle reliability evaluation and predictive maintenance method for complex products using large language models (LLMs). In contrast to existing methods that rely on isolated stage-wise analysis or single-source data, the proposed method integrates multi-source information from the design, production, assembly, and operation stages into a unified dataspace by exploiting the semantic understanding and text-processing capabilities of LLMs. Based on stage-specific failure characteristics, adaptive statistical models are employed to construct component-level reliability models, enabling dynamic reliability evaluation throughout the entire lifecycle. Furthermore, a safety–economic multi-objective optimization model is established, considering reliability and maintenance costs. The optimal predictive maintenance strategy is obtained using multi-objective particle swarm optimization. A case study based on a Chinese aero-engine manufacturer demonstrates that the proposed method effectively improves system reliability while reducing maintenance costs and failure frequency. This work enables lifecycle reliability management and maintenance decision-making for high-precision complex products.

Keywords: Estimation and Identification, Intelligent and AI Based Control, Linear Systems
Abstract: The Kalman filter is the optimal filter for accurate state estimation in linear Gaussian state-space models. Since the emergence of the Kalman filter, it has found extensive application in fields such as robotics, aerospace, and autonomous driving. Traditional derivations rely on orthogonal projection and Bayesian filtering theory. However, in this paper, we introduce a novel perspective by reinterpreting the Kalman filter through dynamic programming—a mathematical optimization technique that decomposes complex problems into overlapping subproblems solved recursively for optimal solutions. First, we construct a dynamic model of the estimation error for a linear system. By leveraging the model's iterative characteristics and assuming independence in the noise distribution, we have demonstrated that the dynamic model exhibits Markovian properties. Using mathematical induction and the established Markov model, we then prove that the Kalman filter possesses both optimal substructure and overlapping subproblem properties, laying the foundation for proving the Kalman filter from a dynamic programming viewpoint. Furthermore, we formulate the Bellman equation for the Kalman filter and apply policy iteration to alternately perform policy evaluation (solving for the optimal error covariance) and policy improvement (determining the optimal Kalman gain). Through this approach, we successfully reconstruct the Kalman filter from the perspective of dynamic programming. Our method not only provides a novel perspective for the proof of the Kalman filter but also bridges the gap in understanding the Kalman filter from the perspective of reinforcement learning.

Keywords: Estimation and Identification, Intelligent and AI Based Control, Real-time Systems
Abstract: This paper proposes a hybrid BiLSTM-EKF framework for real-time radome error slope estimation in precision-guided aircraft. By combining data-driven learning with model-based filtering, the method dynamically adapts noise parameters through temporal analysis of seeker measurements. This enables simultaneous achievement of real-time accuracy in low-noise conditions and robust performance in high-noise environments. Simulations demonstrate the approach maintains low mean square error across varying noise conditions with inference latency below 0.4 ms, while effectively suppressing parasitic feedback loops to enhance guidance stability. The proposed solution overcomes conventional EKF limitations and provides reliable performance under complex electromagnetic interference and noise uncertainty.

Keywords: Estimation and Identification, Linear Systems, Multi-agent Systems
Abstract: This paper investigates distributed observer design under dynamic event-triggered communication. For event-triggering mechanisms, ensuring a strictly positive minimum inter-event time (MIET) is crucial for practical implementation. For dynamic event-triggered distributed observers, existing works typically rely on time-regularization to guarantee such a positive MIET. In contrast to existing works, this paper demonstrates that, for the distributed observer problem, a dynamic event-triggering mechanism can inherently ensure a strictly positive MIET. Additionally, exponential convergence of the estimation error is guaranteed. Finally, the effectiveness of the proposed method is verified through numerical simulations.

Keywords: Estimation and Identification, Nonlinear Systems and Control, Process Control & Instrumentation
Abstract: Data-enabled moving horizon estimation (MHE) reconstructs system states by solving an online optimization problem based on a non-parametric model constructed from system data. However, its applicability to modern industrial processes, which typically have high nonlinearity and large scales, is limited. Specifically, the non-parametric representation is only applicable to linear time-invariant systems, and solving optimization problems is computationally intensive, especially for large-scale systems. To address these challenges, this paper proposes a computationally efficient data-enabled state estimation approach for nonlinear systems. By integrating Koopman operator theory with Willems’ fundamental lemma, the proposed framework enables nonlinear systems to be represented directly from data without explicit model identification. Moreover, a neural network is employed to generate an operator, based on which full-state information can be reconstructed without solving an optimization problem online. The effectiveness of the proposed method is demonstrated via an application to a membrane-based wastewater treatment process.

Keywords: Sensor Networks, Networked Control, Signal Processing
Abstract: In this paper, the problem of stochastic sensor scheduling has been studied. The dynamic of the estimation error is modeled as a Markovian jump system. Then the sufficient and necessary condition for the mean-square boundedness of the estimation error has been provided. Moreover, a unified optimization framework is proposed to jointly design the sensor selection probability and the transmission power. Since the optimization problem is non-convex, the successive convex approximation algorithm has been designed to obtain the suboptimal solution. Simulation results are provided based on the three-tank system to demonstrate the feasibility and the efficiency of the theoretical results.

Keywords: Modeling and Control of Complex Systems, Networked Control
Abstract: This paper proposes a data-based event-triggered control scheme for Markov jump systems (MJSs) with stochastic sampling. An event-triggered controller is constructed via transmitted sampled states, under which the stability criteria depending on system parameters are derived. Considering that the actual system parameters may not be fully known or available to the controller, this work further focuses on the control strategy under completely unknown system parameters. First, by fully considering the stochastic nature of the MJSs, input/state data corresponding to each mode are collected offline. Subsequently, by introducing a mode-specific data-based representation, the original model-dependent sufficient conditions are transformed into the fully data-based forms. Based on this, the designed controller can be implemented using only the offline collected input/state data. Finally, the simulation results verify the effectiveness of the proposed data-based control strategy.

Keywords: Multi-agent Systems, Learning-based Control, Nonlinear Systems and Control
Abstract: Multi-UAVs cooperative path planning in complex environments necessitates simultaneous optimization of trajectory efficiency and satisfaction of kinematic, safety, and communication constraints, significantly elevating problem complexity. To address this challenge, a multi-UAVs cooperative path planning method based on deep Q-learning is proposed. A multi-constraint environmental model is established, integrating a two-dimensional grid representation with explicit communication connectivity metrics. A multi-objective reward function is designed to jointly optimize path length, communication interruption rate, and maximum communication interruption time, guiding agents toward policies that balance navigational efficiency with network maintenance. Simulation experiments demonstrate that the proposed method enables 3-UAVs coverage of 3-4 targets and synchronized arrival of 2-UAVs, maintaining communication interruption rate below 15%. The results validate the feasibility and effectiveness of the method under multi-constraint conditions, highlighting its potential for real-world multi-UAVs cooperative missions.

Keywords: Learning Systems, Robotics
Abstract: Existing deep reinforcement learning (DRL) methods for the traveling salesman problem (TSP) are primarily developed under the Euclidean cost model. However, many practical routing tasks involve curvature-constrained vehicles, leading to the Dubins TSP (DTSP), where travel costs are curvature-dependent and the visiting order is tightly coupled with per-visit headings. This coupled structure poses a challenge to existing DRL approaches. This paper formulates DTSP as a Markov decision process and proposes DRL-DTSP, a DRL approach with encoder--decoder framework that autoregressively generates a coupled action pair to jointly select the next node and its heading. To better capture curvature-dependent costs, we develop a curvature-constraint-aware decoder with physics-informed node-selection biases and a dedicated heading-selection module driven by tailored geometric features. Instance augmentation and a lightweight heading refinement are further applied to improve solution quality. Numerical results demonstrate that DRL-DTSP achieves competitive or superior solution quality against representative baselines while remaining computationally efficient, with increasing advantages under tighter curvature constraints.

Keywords: Fuzzy and Neural Systems, Robotics
Abstract: 3D Gaussian Splatting (3DGS) has recently revolutionised the field of real-time new-view synthesis; however, its optimisation performance remains closely dependent on the quality of the initial sparse representation, which is typically derived from traditional Structure from Motion (SfM) algorithms like COLMAP. In complex environments featuring texture-free regions, repetitive patterns, or significant changes in lighting, traditional SfM often produces fragmented or overly noisy initialisations. These unreliable initialisations confine the 3D Gaussian distribution to erroneous spatial priors, leading to catastrophic ghosting artefacts and structural blurring. To address this, we propose Deep-GeoGS, a robust and efficient deep geometric initialisation framework designed to provide high-quality spatial priors for 3DGS. By explicitly modelling the geometric consistency between high-dimensional depth descriptors and epipolar constraints, we effectively eliminate spatial outliers that would otherwise degrade the quality of the radiance field. Our method demonstrates excellent performance across multiple datasets.

Keywords: Estimation and Identification, Robotics, Flexible Manufacturing Systems
Abstract: In industrial logistics environments, chaotically stacked flexible packages present substantial challenges for robotic sorting due to severe overlap, ambiguous object boundaries, and large morphological variations. These issues are further amplified in few-shot scenarios, where limited training data often leads to overfitting and poor generalization in conventional vision models. To address these challenges, this paper proposes a few-shot pose estimation framework for robotic sorting of flexible packages with complex overlapping and boundary ambiguity. First, to alleviate data scarcity, a generative data augmentation strategy is developed by combining explicit physical spatial transformations with implicit feature disentanglement using a β-VAE. This hybrid augmentation mechanism significantly improves the diversity and robustness of the training dataset under severe occlusion conditions. Second, a multi-task perception network based on Directional Mask R-CNN is introduced to simultaneously perform instance segmentation and orientation estimation, enabling accurate perception of overlapping flexible packages. The 3D spatial centroid (X,Y,Z) of each target is then efficiently recovered by aligning the predicted masks with depth data, enabling reliable pose estimation for robotic manipulation. Finally, a system level validation platform is implemented using a FRANKA robotic arm equipped with a vacuum suction gripper to execute autonomous sorting tasks. Experiments in real chaotic stacking scenarios demonstrate that the proposed method achieves a visual recognition accuracy of 99.7% and a physical grasping success rate of 86.5%, validating its effectiveness for few-shot robotic sorting of flexible packages.

Keywords: Robotics, Estimation and Identification, Signal Processing
Abstract: Teleoperated robot-learning datasets of- ten log actions, proprioception, and video under a shared sample index, but that indexing does not guar- antee physical synchrony. In low-cost leader-follower systems, actuator delay, middleware scheduling, and camera buffering can create persistent temporal off- sets that degrade dataset quality. We present LeCal, a LeRobot-native post hoc calibration workflow and released CLI for estimating command-to-state delay, reporting confidence diagnostics, estimating camera- to-proprioception offset when local video is available, and exporting alignment artifacts. The method remains usable without observation.leader state by treating the recorded action stream as the control-side surro- gate. On a released SO-101 calibration dataset with 72 analyzed episodes, LeCal finds a concentrated action- to-state delay distribution with mean 4.194 frames (139.815 ms at about 30 Hz). Applying the estimated lag reduces mean action-state residual from 4.007 to 1.070, a 64.559 % improvement. LeCal clearly outperforms synchronous indexing, a fixed 1-step shift, and GCC- PHAT, while remaining nearly tied with plain normal- ized cross-correlation on this joint-isolation dataset. A full benchmark over 388 episodes from eight public LeRobot corpora shows only a small overall edge over plain NCC, supporting LeCal primarily as a practical calibration workflow rather than a dramatic new cor- relator

Keywords: Motion Control, Fuzzy and Neural Systems, Robotics
Abstract: This paper presents a two-stage vision-guided autonomous docking method for UUV docking in underwater environments. A green optical beacon is used for long-range guidance, while an AprilTag marker is adopted for close-range relative pose estimation and fine attitude adjustment. Combined with a self-attention mechanism and finite-state machine-based control, the proposed method improves docking robustness and precision under varying water-clarity conditions. The method provides a practical solution for reliable underwater autonomous docking.

Keywords: Optimal Control, Nonlinear Systems and Control, Motion Control
Abstract: This paper investigates the trajectory tracking control problem for Autonomous Underwater Vehicles (AUVs) subject to unknown external disturbances. A prescribed-time adaptive dynamic programming (PTADP) scheme is proposed to achieve optimal tracking performance within a user-defined time. Within this scheme, a single critic neural network (NN) with time-varying activation functions is constructed to approximate the solution to the Hamilton-Jacobi-Bellman (HJB) equation. Rigorous theoretical analysis based on the Lyapunov method confirms the Uniform Ultimate Boundedness (UUB) of both the approximation errors and tracking errors. Simulation results validate the significant convergence speed of the proposed PTADP scheme.

Keywords: Learning Systems, Learning-based Control, Optimal Control
Abstract: This paper investigates the robust control problem for constrained systems subject to unknown multi-harmonic disturbances. We propose a composite control approach that integrates a Hopf-oscillator-based learning module with Tube-based Model Predictive Control (Tube-MPC). The core innovation lies in addressing the inherent conflict between learning-induced transients and system safety: while the adaptive oscillators learning disturbance features online, the Tube-MPC provides a robust positively invariant (RPI) safety buffer that effectively absorbs oscillatory and non-monotonic learning residuals. By incorporating the learned disturbance into a feedforward compensation term, the proposed method reduces the effective disturbance impact acting on the plant, thereby enhancing steady-state accuracy and reducing control effort. Theoretical analysis and simulation results confirm that the approach guarantees recursive feasibility and hard-constraint satisfaction throughout the adaptive learning process.

Keywords: Multi-agent Systems, Adaptive Control, Nonlinear Systems and Control
Abstract: Achieving consensus control is essential for high-precision satellite formation and the rapid deployment of laser communications within very low earth orbit (VLEO), constellations. However, formation flying missions pose strict pointing constraints on VLEO satellites. Furthermore, the satellite bodies are subjected to multi-source disturbances, such as flexible vibrations and environmental disturbances, degrading the attitude control accuracy, which in turn affects the relative attitude acquisition. To address the anti-disturbance high-precision consensus problem, this paper proposes a cooperative prescribed performance control scheme by integrating a distributed nonlinear disturbance observer (DNDO) and a distributed relative attitude observer (DRAO). Firstly, considering the actual satellite communication topology, a DRAO is designed to enable each satellite to acquire its desired pointing direction. Secondly, a DNDO is developed based on the disturbance dynamics of each satellite to achieve precise estimation of disturbances. Finally, by combining the DNDO and DRAO, a prescribed performance control scheme is designed based on barrier Lyapunov function under pointing angle error constraints. Simulation results based on a multi-satellite system consisting of five satellites demonstrate the effectiveness and superior performance of the proposed method.

Keywords: Control Applications, Nonlinear Systems and Control, Learning-based Control
Abstract: To address the high-precision control problem of the coarse loop pitch axis of an Electro-Optical (EO) pod under the combined effects of carrier maneuvering, vibration, mechanical friction, mass imbalance torque, and parameter uncertainties, a control scheme based on a Radial Basis Function (RBF) neural network is proposed. First, the dynamic model of the coarse loop pitch axis is established to clarify the characteristics of composite multi-source disturbances. Second, an adaptive law and a sliding mode controller based on a Radial Basis Function neural network were designed to accurately estimate the dynamic characteristics of the electro-optical pod and achieve precise control of the system. The effectiveness of the proposed method is verified through simulations.

Keywords: Nonlinear Systems and Control
Abstract: Secure attitude maneuvers of spacecraft in Very Low Earth Orbit (VLEO) are affected by composite disturbances, including center of mass variations and atmospheric drag. For systems subject to mandatory constraints, such disturbances can compromise attitude tracking precision and potentially lead to safety violations. To address this challenge, a maneuvering control scheme based on disturbance observation and barrier function is proposed. First, a coupled spacecraft attitude dynamics model is developed to reveal the influence of composite disturbances. Second, a composite controller incorporating a nonlinear disturbance observer and a state-dependent barrier function is designed to perform real-time compensation of these disturbances while ensuring constraints compliance. The effectiveness and robustness of the proposed approach are validated through numerical simulations.

Keywords: Man-machine Interactions, Signal Processing, Estimation and Identification
Abstract: In highly automated autonomous driving systems, passenger state perception is essential for safe operation, reliable decision-making, and effective human--machine collaboration. However, existing studies mainly focus on coarse-grained binary stress classification and lack the ability to distinguish stress induced by different sensory disturbances. To address this limitation, this paper proposes an EEG (Electroencephalogram)-based stress type recognition method using deep learning. A realistic autonomous driving environment with visual and auditory stressors is constructed, and EEG data are collected from multiple subjects. Sequential features are extracted using sliding windows and frequency-domain analysis. An LSTM-MHSA(LSTM with Multi-Head Self-Attention) network is designed to capture temporal dependencies and focus on critical information. Cross-subject experiments achieve an accuracy of 82.66%, outperforming traditional machine learning methods and the standard LSTM model. The results demonstrate the effectiveness of the proposed method and support intelligent perception and safety-oriented human--machine collaboration in autonomous driving systems.

Keywords: Real-time Systems, Sensor/Data Fusion, Factory Modeling and Automation
Abstract: Reliable localization is essential for autonomous operation in unmanned industrial environments, where vision-based methods often degrade due to poor illumination, occlusion, and privacy constraints. WiFi Channel State Information (CSI) enables fingerprinting based localization by capturing fine-grained signal characteristics. However, existing approaches either rely on limited handcrafted features or employ deep models such as CNNs, LSTMs, and Transformers, which suffer from restricted global modeling capability or high computational complexity. To address these challenges, this paper proposes a novel Mamba-based CSI fingerprinting framework built upon a selective state space model. The proposed method efficiently captures long-range dependencies in high-dimensional CSI sequences with linear computational complexity, inherently bypassing the sequential and computational bottlenecks of traditional architectures. Experimental results in a real-world indoor environment demonstrate that the proposed approach achieves high localization accuracy with a root mean square error (RMSE) of 1.04 m and a mean absolute error (MAE) of 1.08 m. Furthermore, complexity analysis on a standard CPU platform reveals that the proposed framework achieves an exceptionally low inference latency of 1.51 ms, outperforming conventional and deep learning baselines in efficiency. This optimal trade-off between localization accuracy and real-time execution speed confirms the immense potential of the proposed method for resource-constrained devices in dynamic unmanned systems.

Keywords: Micro and Nano Systems, Modeling and Control of Complex Systems, Robotics
Abstract: Planar printed circuit board (PCB) spiral coils are widely used in programmable magnetic actuation systems due to their compact structure and ease of array integration. For magnetic microrobot control, accurate and computationally efficient modeling of magnetic fields and their spatial gradients is essential, as they directly determine magnetic force generation and motion controllability. This paper presents a control-oriented semi-analytical magnetic field model for planar PCB spiral coils based on a parametric line-integral formulation of the Biot–Savart law along the Archimedean spiral conductor path. The proposed model enables efficient computation of three-dimensional magnetic flux density and magnetic field gradients, making it suitable for iterative evaluation and future real-time control applications. The modeling accuracy is validated through comparison with three-dimensional finite-element simulations in Ansys Maxwell. Based on the computed field gradients, magnetic forces acting on a spherical microrobot are derived, establishing a direct mapping from coil current inputs to mechanical actuation forces. Furthermore, a programmable planar PCB spiral coil array and a magnetic microrobot actuation platform are implemented, and experiments including multi-microrobot motion and microfluidic manipulation demonstrate the application potential of this system in microfluidic transportation and micro-operation.

Keywords: Micro and Nano Systems, Robotics, Motion Control
Abstract: Magnetic microrobots are promising platforms for targeted drug delivery, with sperm-inspired types attracting particular attention due to their flexible, wave-like motion. Despite significant progress in this field, the simple fabrication of sperm-like flexible robots remains a major challenge. This paper proposes a novel, universal strategy based on a magnetic head made of iron oxide and a flexible tail from filamentous algae for the fabrication of a bio-hybrid flexible sperm-like microrobot, which demonstrates potential for targeted drug delivery within microenvironments in the future. Based on its ingenious head-tail connection structure, the microrobot can effectively move under a precessing magnetic field and achieve reciprocating motion without turning back. Furthermore, vascular channel simulation experiments validate its excellent motion performance in microenvironments. This fabrication method is simple, versatile, and exhibits autofluorescence, providing a new pathway for the large-scale fabrication of high-performance medical microrobots.

Keywords: Micro and Nano Systems, Control Applications, Robotics
Abstract: Liquid transport via electrowetting-on-dielectric (EWOD) represents an effective microfluidic manipulation method. However, owing to the complexity of droplet motion at the microscale, in-depth research on the characteristic analysis of the droplet motion process remains lacking. In this study, an electromechanical model of parallel plate electrodes was introduced to conduct a force analysis of the droplet motion process. Furthermore, simulations based on COMSOL were performed to investigate the effects of dielectric layer thickness and dielectric constant on electric potential distribution. The droplet deformation characteristics induced by contact angle variation were analyzed, and the dynamic response characteristics of the droplet manipulation were evaluated. Meanwhile, to achieve closed-loop control of droplets, a deep learning based visual feedback system was adopted for real-time recognition and localization of droplets. Finally, an integrated droplet manipulation system was constructed, achieving automated transport and fusion of droplets.

Keywords: Control Applications, Micro and Nano Systems, Adaptive Control
Abstract: To address the deterioration in control performance caused by changes in load moment of inertia during mirror replacement in galvo scanners of laser processing systems, this work proposes a three-loop adaptive PID control method based on online moment of inertia identification. First, a dynamic model of the galvo motor and a three-stage closed-loop cascade control architecture are established to analyze how variations in load moment of inertia affect key performance metrics, including the system’s natural frequency and damping ratio. Subsequently, an online identification scheme directly derived from the mechanical motion equations is developed. By combining low-frequency logarithmic sweep excitation with a recursive least-squares algorithm, the proposed method enables accurate identification of the galvo scanner’s moment of inertia. Finally, based on the identified inertia values, the velocity-loop PI gains and position-loop P proportional gain of the controller are adaptively updated. Simulation results demonstrate that the proposed approach rapidly adjusts control parameters after mirror replacement while preserving the standard PID architecture. The load-adaptive controller enables the galvo scanner to achieve a settling time to 1% of full-scale less than 1 ms across mirrors with different moment of inertia. This method effectively resolves the trade-off between the requirement for mirror replacement and high precision control in galvo scanning systems. Moreover, the identification algorithm is computationally efficient, facilitating hardware implementation.

Keywords: Sensor/Data Fusion
Abstract: With the rapid development of maritime surveillance technology, the volume of data generated by sea-surface target detection systems has grown exponentially. The effective governance of massive heterogeneous detection data has become a critical bottleneck for maritime situational awareness. To address the issues of inconsistent attribute descriptions, difficult multi-source data fusion, and severe track fragmentation in current maritime detection data, this paper proposes an automated data governance method for massive maritime detection data. This method constructs a unified maritime target knowledge base to achieve attribute standardization for multi-source heterogeneous data from satellites, radar, electro-optical sensors, ADS-B, and AIS. A multi-source attribute alignment algorithm based on time-decay weighting is proposed to resolve attribute discrimination conflicts for the same target. A track merging mechanism based on attribute-spatiotemporal-batch three-dimensional association is designed to effectively handle track fragmentation caused by detection interruptions. Finally, a quadratic least-squares fitting method is employed to achieve track splicing and smoothing, realizing fully automated processing from raw detection data to high-quality track data. Experimental results demonstrate that the proposed method achieves 97.6% attribute normalization accuracy, 95.39% track merging correctness, providing a reliable data foundation for maritime target tracking and situational analysis.

Keywords: Motion Control, Linear Systems, Control Applications
Abstract: In order to solve the problem of precision tracking of fast steering mirror system (FSM), an anti-disturbance control method based on super-twisting sliding mode was proposed. Based on the super-twisting algorithm (STA), the perturbation observer and controller are constructed respectively. In this approach, the disturbance observer estimates both external disturbances and internal parameter uncertainties in real-time and provides feedforward compensation accordingly. The position closed-loop control of the fast steering mirror adopts a second-order super-twisting sliding mode controller to achieve effective tracking. Numerical simulation results demonstrate that the proposed method can successfully estimate disturbances, effectively mitigate the chattering phenomenon inherent in traditional sliding mode control (SMC), and reduce the impact of external disturbances, thereby enhancing the precision and robustness of the FSM control system.

Keywords: Intelligent and AI Based Control, Adaptive Control
Abstract: Unmanned Underwater Vehicles (UUVs) are essential for executing critical marine missions; however, traditional navigation systems face severe bottlenecks in complex underwater environments owing to limited sensing ranges, communication constraints, and the absence of autonomous decision-making. Consequently, intelligent trajectory planning has emerged as the foundational technology for realizing UUV autonomous navigation. This paper systematically reviews the state-of-the-art in UUV intelligent trajectory planning, focusing primarily on autonomous obstacle avoidance and energy consumption optimization. We first describe our literature search methodology and present a structured algorithmic taxonomy. We then categorize mainstream algorithms ranging from heuristic searches to deep reinforcement learning frameworks and summarize their key innovations and performance metrics. While contemporary methods demonstrate significant breakthroughs in simulated scenarios, transitioning these theoretical models to robust engineering systems remains a substantial challenge. This review identifies critical technical bottlenecks, including environmental uncertainty, the trade-off between real-time computational performance and algorithmic complexity, and multi-objective optimization conflicts—and discusses the simulation-to-reality (sim-to-real) gap with concrete engineering guidance. Finally, we outline future research priorities spanning hardware advancement, software algorithm optimization, and reliability enhancement, including standardized benchmarking frameworks and hardware-in-the-loop (HIL) validation approaches, aiming to promote the engineering practicality of highly adaptable, autonomous underwater embodied agents.

Keywords: Robotics, Motion Control, Modeling and Control of Complex Systems
Abstract: Autonomous underwater gliders rely on net buoyancy modulation and internal mass redistribution to achieve energy-efficient, long-endurance locomotion. Unlike conventional propeller-driven vehicles, their flight mechanics are fundamentally governed by natural trim equilibria rather than direct continuous thrust. This paper formulates a comprehensive multi-body dynamic framework that explicitly couples rigid-body kinematics, anisotropic added-mass effects, and nonlinear hydrodynamics with internal moving-mass actuation. We propose a novel trim-oriented control methodology that leverages a buoyancy adjustment system (BAS) to govern macroscopic dive-to-climb tendencies, while utilizing an attitude adjustment system (AAS) for bounded, localized pitch-trim regulation. Furthermore, a depth-dependent scheduling mechanism is integrated to systematically throttle volumetric transitions. Comprehensive numerical simulations demonstrate that the proposed framework yields highly stable, repeatable glide cycles. By mitigating severe transient oscillations and avoiding actuator saturation near apogee and perigee inflections, this approach provides a robust, physically intuitive, and hardware-aware control paradigm for underactuated marine robots.

Keywords: Sensor/Data Fusion, Robotics
Abstract: Underwater Autonomous Unmanned Systems (UAUS) increasingly rely on Simultaneous Localization and Mapping (SLAM) to enable closed loop autonomy in missions such as infrastructure inspection, ocean mapping, and environmental monitoring. Compared with terrestrial and aerial platforms, underwater operation presents distinctive sensing constraints. Global positioning is generally unavailable, optical perception can degrade severely in turbid water, and acoustic sensing, despite its reliability under low visibility, often exhibits structured artifacts, heavy tailed noise, low update rates, and non-negligible latency. These characteristics shape both the operating envelopes of practical sensors and the reliability of the measurement constraints provided to the estimator. This review connects sensor operating envelopes with deployable SLAM design for UAUS. We first summarize common underwater sensing modalities and their dominant constraints and failure mechanisms. We then review representative SLAM framework through a modular pipeline, focusing on data synchronization and association, local pose estimation, and loop closure with global optimization. Within local estimation, we organize methods into optimization based, filtering based, and learning enhanced categories, and discuss how they are combined in practice. Finally, motivated by failures observed in turbid vision and noisy acoustic mapping, we outline key limitations and future directions toward uncertainty aware front-ends, robust global estimation, adaptive multi-sensor fusion, and evaluation protocols that better reflect field deployment requirements.

Keywords: Smart Buildings, Fault Detection and Diagnostics, Signal Processing
Abstract: Underwater piles require routine inspection, but optical sensing is often unreliable in turbid water, making forward-looking imaging sonar an attractive alternative. Public datasets and reproducible evaluation protocols for pile inspection in sonar imagery, however, remain scarce. We present dataset{}, a forward-looking sonar benchmark for joint pile detection and state classification. The current release contains 8,120 raw sonar clips and a human-labeled core set of 1,981 three-frame temporal composites (T3) with 2,019 annotated pile instances. The released state labels represent sonar-observed appearance states inferred from acoustic evidence rather than externally validated physical ground truth. T3 encodes short-term dynamics in a three-channel form that remains compatible with standard 2D detectors. To support reproducible benchmarking, we release annotations, manifests, fixed split files, evaluation scripts, and a traceable model-in-the-loop expansion pipeline. We further report baselines for Faster R-CNN and several YOLO variants under COCO-style metrics. Benchmark results indicate that pile-state detection in noisy sonar imagery remains challenging, especially for minority fractured cases.

Keywords: Learning Systems, Modeling and Control of Complex Systems, Estimation and Identification
Abstract: Ocean current fields contain complex spatial patterns and temporal evolution, which pose significant challenges for efficient environment perception. This paper proposes a novel spatiotemporal environment perception framework for ocean current fields based on reduced-order modeling. After preprocessing and valid-ocean masking, singular value decomposition is applied to historical current snapshots to extract dominant spatial modes and form a compact low-rank representation. On this basis, two downstream tasks are considered. For spatial perception, sparse observations are used to reconstruct the full current field through least-squares estimation of modal coefficients. For temporal prediction, dynamic mode decomposition is introduced in the reduced-order coefficient space to perform one-step forecasting. Experimental results demonstrate that the proposed framework achieves accurate reconstruction from partial observations and effective short-term prediction, indicating that it can provide a compact and interpretable representation for unified spatiotemporal perception of ocean current fields.

Keywords: Modeling and Control of Complex Systems, Robotics, Estimation and Identification
Abstract: Semi-submerged hydrofoil systems provide an effective solution for reducing hydrodynamic drag in high-speed unmanned surface vehicles (USVs) by partially lifting the hull out of the water. However, the hydrodynamic characteristics of such systems vary significantly with changes in immersion state and vehicle motion, making accurate real-time modeling challenging.

In this paper, a real-time modeling framework for a multi-strut semi-submerged hydrofoil system is proposed. The hydrodynamic forces are reconstructed using a vector decomposition approach based on the approximately linear relationship between hydrodynamic characteristics and the submerged length of slender hydrofoil struts, while hydrostatic forces are evaluated using a volumetric discretization method.

The model is validated through both computational fluid dynamics (CFD) simulations and towing tank experiments. Results demonstrate that the proposed approach achieves accurate prediction of lift, drag, and pitching moment while maintaining real-time computational performance, making it suitable for dynamic simulation and control of hydrofoil-based USVs.

Keywords: Control of Distributed Generation Systems, Fault Detection and Diagnostics, Control of Smart Power Delivery Systems
Abstract: This paper investigates the problem of resilient secondary control for smart grid systems subject to false data injection (FDI) attacks. An event-triggered fixed-time secondary control strategy is proposed to simultaneously guarantee fast frequency regulation and proportional active power sharing while significantly reducing communication burden. The proposed control framework integrates fixed-time convergence properties with an event-triggered communication mechanism, ensuring that system states converge to desired equilibria within a predefined time bound independent of initial conditions, even in the presence of bounded FDI attacks. Rigorous theoretical analysis is provided to establish fixed-time stability and to exclude Zeno behavior. Simulation results demonstrate the effectiveness of the proposed approach in enhancing resilience against cyber-attacks while achieving substantial communication savings compared with conventional time-triggered schemes.

Keywords: Multi-agent Systems, Learning-based Control, Modeling and Control of Complex Systems
Abstract: This paper addresses the fault-tolerant control problem for multi-agent systems with flexible manipulators governed by coupled ordinary differential equation-partial differential equation (ODE-PDE) dynamics. A distributed control framework is proposed to handle actuator faults, flow-induced disturbances and obstacle avoidance constraints through three integrated mechanisms. First, each agent employs a model-based baseline controller that compensates for unknown faults while suppressing elastic vibrations. Second, a neural disturbance perception encoder is designed to extract low-dimensional latent features from dynamics residuals, providing all agents with a consistent representation of the flow environment. Third, residual reinforcement learning policies trained with disturbance-aware rewards augment the baseline control through Riemannian motion policy (RMP) flow composition, ensuring asymptotic convergence to consensus under disturbance-free conditions and bounded tracking errors under persistent disturbances. Numerical simulations demonstrate the effectiveness of the proposed methods.

Keywords: Networked Control, Multi-agent Systems
Abstract: This paper studies distributed Nash equilibrium seeking (DNES) for multicoalition cyber-physical systems over weight-unbalanced directed communication networks. Within each coalition, all nodes cooperate to optimize the coalition's objective under a consensus constraint, while different coalitions compete with one another. First, by integrating multi-step planning with optimal control techniques, a specified-time DNES framework is developed for such multicoalition cyber-physical systems, guaranteeing convergence to the Nash equilibrium (NE) within a specified settling time. Furthermore, to eliminate biases in both the average gradient estimation and the system state that arise from the unbalanced directed communication topology inside each coalition, a specified-time balance compensator is designed for each coalition based on in-neighbor sampling information. Then, by utilizing properties of iteration matrices and constructing discrete Lyapunov functions, the specified-time convergence of the compensator and the overall system is rigorously established. Finally, a simulation result involving the power generation game is presented to demonstrate the effectiveness of the proposed algorithm.

Keywords: Multi-agent Systems, Motion Control, Control Applications
Abstract: This paper proposes a cascaded cooperative control strategy using a fixed-time extended state observer (FTESO) for air-ground heterogeneous systems under complex disturbances. Firstly, an air-ground dynamic model incorporating a bidirectional signal connection is established.Then, a FTESO is designed to estimate modeling errors and wind disturbances in real time, ensuring observation error convergence within a fixed time. A cooperative integral sliding mode controller manages outer-loop coordination to reduce deviations, while an inner-loop controller tracks attitude commands. Lyapunov analysis confirms global system stability. Finally, simulations demonstrate superior robustness and accuracy compared to traditional nonlinear observers.

Keywords: Fault Detection and Diagnostics, Signal Processing
Abstract: Wind turbine gearboxes operate under variable loading and harsh environmental conditions, making early fault detection essential for reducing turbine downtime and operation and maintenance (O&M) costs. This study proposes an interpretable fault indicator based on wavelet packet decomposition (WPD) combined with a normalized rolling-energy ratio. The analysis focuses on the 10 Hz to 40 Hz frequency band, which is physically motivated by the frequency region surrounding the rotational frequency of the high-speed shaft (i.e., 1800 rpm ≈ 30 Hz). Using vibration measurements from the NREL gearbox condition-monitoring dataset, one-minute vibration segments sampled at 40 kHz are processed through a decimated WPD framework to reconstruct narrow subbands within the target frequency region. A peak-response time window is then identified using an energy ratio between a detector subband and a low-frequency reference band, enabling the extraction of compact windowed features. Experimental analysis across multiple accelerometer locations shows that the selected subband centered near 34 Hz provides the strongest separation between healthy and damaged gearbox conditions. The proposed time-localized wavelet packet feature extraction framework provides an interpretable and computationally efficient approach for wind turbine gearbox fault detection.

Keywords: Fault Detection and Diagnostics, Signal Processing
Abstract: Wind energy plays a key role in modern power systems, but the rapid expansion of the sector creates challenges for the long-term reliability and operational safety of wind turbines. Normal behavior models (NBMs) are widely used for turbine-performance monitoring, yet most existing approaches rely on single-target formulations that increase deployment complexity and often produce ambiguous fault indications. To address these limitations, we propose a multi-target NBM that jointly monitors multiple supervisory control and data acquisition (SCADA) signals through a unified health indicator. The framework integrates autoregressive (AR) lag features within a stacking ensemble to capture temporal dependencies and nonlinear operating patterns, and couples the resulting residuals with a LOESS-smoothed mean-squared-deviation (MSD) decision statistic. The proposed framework is then validated on pitch-fault anomaly data using active power and gearbox-oil temperature signals. Compared with conventional single-target NBMs, the proposed approach suppresses spurious alerts, increases confidence in detected anomalies, reduces monitoring effort, preserves predictive accuracy, and improves the reliability of fault diagnosis.

Keywords: Multi-agent Systems, Learning Systems, Robotics
Abstract: This paper proposes a multi-agent cooperative pursuit framework that integrates target state estimation and cooperative pursuit control. For state estimation, a Collaborative Adaptive Kalman Filter (CAKF) is developed, which (i) integrates bearing-angle measurements from multiple pursuers to improve accuracy and robustness under noisy conditions and (ii) performs online adaptive updates of the noise covariance based on an innovation-driven mechanism. For the pursuit task, a Distributed Multi-Agent Deep Deterministic Policy Gradient (DMADDPG) framework is proposed to alleviate the problem of poor generalization ability of single-strategy controllers. A large number of experiments verify the effectiveness, robustness, and scalability of the proposed system.

Keywords: Multi-agent Systems, Nonlinear Systems and Control, Optimal Control
Abstract: This paper investigates the distributed formation control problem for heterogeneous nonlinear discrete-time multi-agent systems under safety and stability constraints. The core challenge lies in ensuring safety within heterogeneous multi-agent systems operating in discrete-time environments, while addressing nonlinear uncertainties that further complicate control design. To address these issues, a control framework based on the backstepping method is proposed, integrating a control Lyapunov function for stability and a higher-order control barrier function for safety. To accommodate varying agent orders while guaranteeing bounded formation error and collision-free operation, a discrete-time formulation of the control barrier function is further developed. Moreover, neural networks are employed to approximate unknown nonlinearities and uncertainties, thereby enhancing the overall robustness and adaptability of the control strategy. Finally, a numerical simulation validated the effectiveness of the proposed formation control.

Keywords: Automated Guided Vehicles, Intelligent and AI Based Control, Multi-agent Systems
Abstract: Heterogeneous vehicle platooning relies on Vehicle-to-Everything (V2X) communication, exposing the system to False Data Injection (FDI) attacks. In such adversarial environments, standard Deep Reinforcement Learning (DRL) methods calculate rewards directly from compromised observations. This dependency creates a reward trap where agents optimize corrupted objectives and execute unsafe behaviors. To address this issue, a Meta-Gradient based Reward Shaping (MGRS) framework is proposed for resilient platoon control. Formulated as a heterogeneous optimization problem, the framework introduces a parameterized reward teacher updated via meta-gradients derived from ground-truth physical states. This mechanism dynamically rectifies biased reward signals, enabling heterogeneous agents to learn safe policies despite sensor falsification. Simulation results demonstrate that the MGRS framework significantly enhances system resilience, maintaining string stability and avoiding collisions under FDI attacks where traditional static-reward baselines fail.

Keywords: Sensor/Data Fusion
Abstract: Pose tracking failure in elevator environments is currently a most challenging issue for Visual-Inertial SLAM systems, mainly due to the contradict measurements between the nearly static visual features and the dramatical acceleration changes captured by IMU measurements. To address this challenge, this paper investigates the key features of the inconsistent measurements, and proposes VINS-LIFT, a novel framework that leverages Visual-Language Models (VLMs) to enhance the robustness of VI-SLAM in environments with elevators. Specifically, our system fuses lightweight visual scene understanding (distilled from Qwen2.5-VL) with IMU motion tracking through a novel fusion mechanism, triggering motion priors and adaptive covariance modulation once an elevator scenario is detected. This synergistic framework enables reliable state estimation despite sudden lighting changes, motion disturbances, or visual degradation common in elevator environments. We validate our method on real-world datasets, demonstrating significant performance gains in trajectory estimation and robustness compared to conventional VI-SLAM baselines. This work highlights the potential of VLM-driven adaptations for enhancing SLAM robustness in challenging motion scenarios.

Keywords: Control Applications, Estimation and Identification, Modeling and Control of Complex Systems
Abstract: Coaxial multirotors offer enhanced thrust density without increasing the platform size, making them attractive for aerial manipulations and aggressive flight. However, the intrinsic aerodynamic interaction between upper and lower rotors introduces thrust loss and torque coupling, significantly increases the complexity of control allocation. This paper proposes a physics-informed optimization control allocation solution for coaxial multirotors. Firstly, a three-dimensional nonlinear polynomial mapping model between motor DShot commands and the generated thrust and torque is experimentally identified. The identified aerodynamic knowledge is then incorporated into a constrained quadratic-programming-based optimization formulation. The resulting framework enables real-time optimal distribution of motor commands while satisfying actuator limits. Comparative experiments demonstrate that the proposed method significantly improves thrust estimation accuracy and mitigates aerodynamic coupling effects.

Keywords: Nonlinear Systems and Control
Abstract: This paper investigates disturbance-observer based model predictive control for nonlinear discrete-time systems. It develops a novel disturbance observer tailored to discrete-time nonlinear dynamics. To mitigate the effect of disturbances, the control law incorporates an explicit disturbance rejection term constructed from the observer estimate. A robust positively invariant set for the state-error dynamics is derived via linear matrix inequalities. Based on this set,tightened state and input constraints are formulated to guarantee constraint satisfaction for the actual system. Closed-loop stability and recursive feasibility are established analytically. Numerical simulations on a cart–spring system substantiate the effectiveness of the proposed disturbance-observer-based nonlinear MPC scheme.

Keywords: Adaptive Control, Nonlinear Systems and Control, Control Applications
Abstract: This study investigates high-speed pursuit-evasion under measurement delays by integrating the unscented Kalman filter (UKF) with Stackelberg differential game theory. The engagement is modeled as a nonlinear differential game, where the UKF compensates for temporal lags and estimates the relative states of non-cooperative players. Building on these estimates, a Stackelberg equilibrium framework enables real-time adaptive maneuvering. Numerical simulations validate the robustness and efficacy of the proposed approach in ensuring successful penetration within complex adversarial environments.

Keywords: Multi-agent Systems, Networked Control, Control of Distributed Generation Systems
Abstract: This paper investigates distributed online minimax optimization over multi-agent networks. A distributed online mirror descent algorithm integrated with compressed communication is proposed to address the limitations of Euclidean-based methods and communication bottlenecks. To the best of our knowledge, this is among the first attempts to design a communication-efficient distributed online minimax algorithm. The algorithm utilizes Bregman divergence to adapt to the geometric structure of decision variables and employs an absolute error compressor to reduce communication overhead. Theoretical analysis establishes a sublinear dynamic regret bound dependent on path variation.

Keywords: Multi-agent Systems, Networked Control, Nonlinear Systems and Control
Abstract: This paper investigates the distributed Nash equilibrium seeking problem for high-order players subject to input delay, unmeasurable states, and unmatched nonlinearities. The distributed estimator is designed to estimate the decisions of non-neighboring players and the state observer is designed to estimate the unmeasurable states. After skillfully introducing the tuning gain and nonlinearities into the design process, the distributed Nash equilibrium seeking strategy is constructed by using a non-recursive design method. Based on the Lyapunov stability theory, it is shown that the decisions of all players can globally asymptotically converge to the Nash equilibrium when the input delay satisfies certain conditions. Finally, the validity of the proposed results is demonstrated by using simulation examples.

Keywords: Control Applications, Linear Systems, Real-time Systems
Abstract: Federated learning has become a popular paradigm for privacy protection and edge-based machine learning. However, defending against differential attacks and devising incentive strategies remain significant bottlenecks in this field. Despite recent works on privacy-aware incentive mechanism design for federated learning, few of them consider both data volume and noise level. In this paper, we propose a novel federated learning system called FEDBUD, which combines privacy and economic concerns together by considering the joint influence of data volume and noise level on incentive strategy determination. In this system, the cloud server controls monetary payments to edge nodes, while edge nodes control data volume and noise level that potentially impact the model performance of the cloud server. To determine the mutually optimal strategies for both sides, we model FEDBUD as a two-stage Stackelberg Game and derive the Nash Equilibrium using the mean-field estimator and virtual queue. Experimental results on real-world datasets demonstrate the outstanding performance of FEDBUD.

Keywords: Learning-based Control
Abstract: Approximating Model Predictive Control (MPC) with neural networks is a promising approach for real-time control on resource-constrained embedded systems. An important consideration in this process is the generation of high-quality training data that effectively covers the feasible state space. In this paper, we propose a data generation framework based on the A_n^* lattice, a well-known covering lattice that is proven optimal for dimensions up to 5 and remains the best known covering for most dimensions up to 21. We develop an efficient algorithm combining Chebyshev centering, random rotation, BFS flooding, and binary search to enumerate lattice points within arbitrary polytopes. We prove a deterministic upper bound on the pointwise approximation error that links the lattice covering radius to the Lipschitz constants of both the MPC policy and the neural network, and further analyze the closed-loop stability. Experiments demonstrate that our lattice-based approach achieves improved approximation accuracy compared to random uniform sampling, while maintaining comparable computational efficiency.

Keywords: Robotics, Motion Control
Abstract: Planetary exploration rovers often must operate for extended durations under the low and nearly constant electrical power provided by radioisotope thermoelectric generators (RTGs). While energy-aware planning has been studied for aerial and underwater robots under battery limits, few works for ground rovers explicitly model power flow or enforce instantaneous power constraints imposed by RTG power conditioning and onboard subsystems. Classical terrain-aware planners emphasize slope or traversability, and trajectory optimization methods typically focus on geometric smoothness and dynamic feasibility, neglecting energy feasibility. We present an energy-constrained trajectory planning framework for RTG-powered rovers that explicitly integrates physics-based models of translational, rotational, and resistive power with baseline subsystem loads. By incorporating both cumulative RTG energy supply and instantaneous bus power constraints into SE(2)-based polynomial trajectory optimization, the method ensures trajectories that are simultaneously smooth, dynamically feasible, and power-compliant. Simulation results on lunar-like terrain show that our planner generates trajectories with peak power within 0.55% of the prescribed limit, while existing methods exceed limits by over 17%. This demonstrates a principled and practical approach to energy-aware autonomy for long-duration planetary missions.

Keywords: Modeling and Control of Complex Systems, Nonlinear Systems and Control, Learning-based Control
Abstract: Traditional quadrotors can only generate thrust in one direction. This limits sustained inverted flight and aggressive maneuvers such as fast 180 degree half-flips. Bidirectional thrust can remove this limit, but it also makes control harder because the motors must pass through a thrust-reversal deadzone. This paper proposes a Differentiable Physics Learning (DPL) pipeline for autonomous half-flips with a bidirectional-thrust quadrotor. The policy is trained with differentiable rigid-body and propulsion dynamics, including the thrust-reversal process. We compare DPL with Model Predictive Control (MPC) in a modified VisFly simulator and in real flight tests. Results show that DPL achieves faster flips and smaller altitude error than MPC. The learned policy is also more robust to thrust-reversal deadzones and requires less online computation. These results show that differentiable physics can be an effective way to train agile controllers for quadrotors with discontinuous actuation.

Keywords: Multi-agent Systems
Abstract: This paper investigates the design of perimeter-defense guidance laws for a defending autonomous surface vehicle (ASV) under input and collision avoidance constraints. A safety-critical perimeter-defense guidance method based on nonlinear model predictive control is proposed. Using the concept of dual optimization design, the optimal attack strategy of the attacker is first estimated by the defending ASV, utilizing information on the dynamics and intent of the attacking ASV. Subsequently, the predicted strategy is embedded into the cost function of the defending ASV, and constraints related to collision avoidance with static obstacles, as well as upper bounds on surge and angular velocities, are incorporated into the optimization framework. By solving this optimization problem, real-time guidance commands are generated that satisfy all constraints, ensuring effective interception of the attacker while guaranteeing safe navigation and avoiding collisions. Finally, the effectiveness of the proposed safety-critical perimeter-defense guidance method is illustrated by the simulation results.

Keywords: Learning Systems, Sensor Networks, Intelligent and AI Based Control
Abstract: Compared with traditional otologic surgery, transcanal endoscopic minimally invasive ear surgery offers prominent advantages of less surgical trauma, shorter operation time, and lower risk of postoperative complications. However, the narrow space of the ear canal makes it extremely challenging to integrate conventional force sensors into surgical instruments, leading to the lack of effective real-time state perception during the round window membrane (RWM) puncture procedure. To address this critical issue, this paper proposes an endoscopic image-based deep learning network for accurate recognition of RWM puncture states, named TBCA-SlowFast. The network takes the SlowFast network as the baseline and innovatively introduces a Temporal Branch Coordinate Attention (TBCA) module to enhance the model's ability to capture fine-grained spatiotemporal features and suppress complex background interference in surgical scenes. We further construct a self-built simulated RWM puncture dataset for model training and validation, and conduct comprehensive comparative experiments to evaluate the performance of the proposed method. Experimental results show that the TBCA-SlowFast network achieves a recognition accuracy of 88.23% on the test set, which is 1.96% higher than the original SlowFast network. Meanwhile, the TBCA module only introduces a small increase in parameters (0.7M) and computational complexity (14.1M FLOPs), realizing a good trade-off between recognition accuracy and real-time performance. The proposed method can provide reliable intraoperative state feedback for transcanal minimally invasive ear surgery, effectively reducing the risk of over-insertion during RWM puncture and improving the safety and reliability of the surgical procedure.

Keywords: Control Applications, Process Control & Instrumentation
Abstract: Conventional PCR instruments are often constrained by low heating and cooling efficiency, insufficient protocol compatibility, and poor suitability for rapid on-site testing. To address these issues, this study develops an ultra-fast temperature-control system for nucleic acid amplification with high efficiency and broad protocol compatibility. The system integrates time-domain and space-domain temperature control schemes and consists of a dynamic temperature cycling module, a thermostatic module, and a microfluidic chip switching module. In the dynamic temperature cycling module, a Peltier thermoelectric cooler serves as the core component, and a sandwich-stacked structure is adopted to enhance heat transfer efficiency and achieve rapid temperature ramping. The thermostatic module employs a combined design of a ceramic heater and an insulating bakelite structure to provide a stable thermal environment for annealing and fluorescence detection. Efficient chip transfer among functional modules is realized through a precision ball screw linear stage and a force-controlled push rod mechanism. A hierarchical closed-loop control architecture is adopted, integrating Bang–Bang control, incremental PID, integral-separation PID, and position–force dual closed-loop control to enable coordinated multi-module operation. In addition, fatigue life was evaluated using the finite element method to ensure the mechanical reliability of the module under frequent reciprocating opening and closing during PCR amplification. Among structural steel, aluminum alloy, and PA66-GF30, aluminum alloy was selected as the module bracket material due to its superior fatigue performance. The proposed system improves amplification speed, temperature-control accuracy, and detection stability while maintaining compatibility with different amplification protocols, providing a compact solution for point-of-care testing and rapid pathogen screening.

Keywords: Robotics, Optimal Control, Adaptive Control
Abstract: Autonomous laparoscope control plays a critical role in maintaining a stable surgical field of view (FOV), improving surgeons’ operational efficiency, and enhancing intraoperative safety. To address the requirement for reliable real-time surgical instrument tracking in minimally invasive surgery (MIS), this paper proposes a visual tracking framework for surgical robots based on Diffusion Policy and optimal remote center of motion (RCM) constraints.A two-stage RCM selection strategy is first developed, where candidate RCMs are predefined through robot kinematics and surgical constraints, and the optimal RCM is determined through multi-criteria evaluation. An RCM-constrained control method based on spherical linear interpolation is then introduced, which ensures zero lateral velocity at the RCM point.By incorporating an intuitive virtual plane (IVP) constraint, our method reduces view misalignment and improves the eye–hand coordination.EfficientSAM is employed for fast and accurate segmentation of surgical targets from laparoscopic images. Finally, a Diffusion Policy–based trajectory tracking method is trained on a dedicated laparoscopic dataset to achieve robust instrument tracking. The experiment obtained relatively good data results through methods such as visualization and error comparison, effectively demonstrating the effectiveness and feasibility of the proposed framework.

Keywords: Intelligent and AI Based Control, Robotics, Learning Systems
Abstract: Estimating tool–tissue interaction force from visual observations is important for surgical simulation and robot-assisted intervention, yet direct force sensing is often unavailable in minimally invasive settings. Existing vision-based approaches typically rely on 2D appearance cues or end-to-end mappings that lack explicit geometric and biomechanical interpretability.

In this work, we propose a multi-stage framework that classifies contact force state from surgical image sequences through explicit geometric and physical representations. Our approach has three key components. First, we reconstruct dynamic tissue geometry from temporal surgical images using 3D Gaussian-based scene modeling to capture deformable structure and motion. Second, we bridge vision and biomechanics by converting the Gaussian representation into mesh structures that enable localized deformation analysis. Third, we learn a deformation-to-force classifier by introducing an efficient MLP structure, in which localized geometric states around the tool tip are mapped to discrete contact/non-contact labels that exhibit step-wise tran-sitions over time.

Validation on a SOFA-based liver simulation (≈48k samples, 162 runs) demonstrates high prediction accuracy and geometric plausibility on public endoscopic data, providing a physically interpretable framework for vision-based surgical force state estimation.

Keywords: Intelligent and AI Based Control, Fuzzy and Neural Systems, Learning Systems
Abstract: 知识图谱（TKG）外推通过建模历史进化模式来预测未来事实。然而，当前方法受制于无关快照产生的特征噪声和高频交互的预测偏差，掩盖了涌现事件。我们提出PACL（模式感知对比学习）以适应性地平衡历史规律性和局部演变。PACL将统计先验与双视角语义特征整合，量化每个查询的历史依赖性，生成一个模式感知系数，调制自适应门控对比损耗。通过强化模式符合实例的一致性，同时放宽涌现事实的约束，PACL防止局部信号被历史惯性平滑化。对四个基准数据集的评估表明，PACL在复杂动态场景下持续优于最先进模型。

Keywords: Robotics, Control Applications, Learning Systems
Abstract: Infrared-visible (IR-VIS) feature matching plays an essential role in cross-modality visual localization, navigation and perception. Along with the rapid development of deep learning techniques, a number of representative image matching methods have been proposed. However, crossmodal feature matching is still a challenging task due to the significant appearance difference. A significant gap for cross-modal feature matching research lies in the absence of standardized benchmarks and metrics for evaluations. In this paper, we introduce a comprehensive cross-modal feature matching benchmark, CM-Bench, which encompasses 30 feature matching algorithms across diverse cross-modal datasets. Specifically, state-of-the-art traditional and deep learning-based methods are first summarized and categorized into sparse, semidense, and dense methods. These methods are evaluated by different tasks including homography estimation, relative pose estimation, and feature-matching-based geo-localization. In addition, we introduce a classification-network-based adaptive preprocessing front-end that automatically selects suitable enhancement strategies before matching. We also present a novel infrared-satellite cross-modal dataset with manually annotated ground-truth correspondences for practical geo-localization evaluation. The dataset and resource will be available at: https://github.com/SLZ98/CM-Bench.

Keywords: Automated Guided Vehicles, Control Applications, Motion Control
Abstract: Unmanned aerial vehicle-based autonomous aerial refueling (UAV-based AAR) is essential for extending mission endurance for long-duration operations, yet rendezvous operations under stochastic wind fields remain a key technical bottleneck. Current methods adopt passive tanker strategies or precomputed trajectories, lacking adaptive online correction for wind-induced deviations. This study proposes a tanker-initiated hierarchical rendezvous planning and replanning framework for four-dimensional (4-D) synchronization (time, position, airspeed, heading) with a receiver UAV on a fixed cruise route. A recursive wind field model fusing steady wind, improved Dryden turbulence, and discrete gusts is embedded into a 3-DOF numerical simulation platform. Offline, a constrained 3-D Dubins planner generates spatiotemporally synchronized reference trajectories satisfying kinematic and fuel constraints; online, the hierarchical replanner switches between full Dubins recomputation and receding-horizon pure-pursuit (RH-PP) refinement to detect wind-induced deviations and suppress error divergence. Numerical experiments verify the proposed framework achieves robust 4-D rendezvous under realistic wind disturbances, providing a practical technical solution for UAV-based AAR.

Keywords: Robotics, Signal Processing, Sensor/Data Fusion
Abstract: USBL integrated positioning systems are affected by position errors induced by the lever arm and misalignment angles between the acoustic array and attitude sensors, making installation-parameter calibration essential for accurate underwater target positioning. This paper proposes an improved adaptive triple sub-population whale optimization algorithm (ATWOA) to estimate installation parameters by minimizing positioning residuals from the installation-error model. Experiments show that, compared with GA-BP, PSO-VMD, conventional WOA, and improved UKF, ATWOA achieves lower positioning errors and smaller dispersion of estimated installation angles, demonstrating its effectiveness and reliability for USV-mounted USBL integrated positioning systems.

Keywords: Sensor/Data Fusion, Signal Processing, Fault Detection and Diagnostics
Abstract: Terminal carrier landing requires reliable redundant guidance information fusion under asynchronous multi-rate measurements, communication delays, packet loss, and non-stationary noise with occasional outliers. This paper proposes an integrated framework that combines time alignment, integrated scheduling, and a robust variational Bayesian adaptive filter with anomaly score driven strong tracking. Three-point Lagrange interpolation with time backtracking is used for time registration, while the scheduling module unifies coordinate frames and manages repeated updates. Simulation comparisons with an IMM baseline demonstrate improved fusion reliability in terminal landing conditions.

Keywords: Estimation and Identification, Intelligent and AI Based Control
Abstract: Aiming at the spacecraft maneuver detection problem under short-arc observation, an intelligent hierarchical method based on bidirectional long short-term memory network with integrated self-attention (BiLSTM-SA) is proposed. This method decomposes the spacecraft maneuver detection problem into two layers: maneuver identification and impulse velocity estimation. During the offline training phase, under short-arc observation condition, the classification and regression networks based on BiLSTM-SA are trained hierarchically with first-order and second-order difference inputs. At the online application phase, based on the output of the classification network for maneuver identification, the regression network for impulse velocity estimation is further applied to output the spacecraft's maneuver time and estimated impulse velocity. Simulation results show that the proposed method achieves higher maneuver identification accuracy and more precise estimates of maneuver time and impulse velocity vector, with reduced decision time compared with the baselines.

Keywords: Robotics, Intelligent and AI Based Control, Learning Systems
Abstract: Magnetic microrobots hold immense potential for biomedical applications such as targeted drug delivery. However, achieving precise and safe navigation remains a challenge due to the complex working environments. In this paper, a safety-critical autonomous navigation framework that integrates Reinforcement Learning (RL) with Control Barrier Functions (CBFs) is proposed to enable robust obstacle avoidance for magnetic microrobots. The CBF constraint is derived to define the safe admissible control space for microrobots navigation. The constraint is embedded as a safety filter layer within a RL policy network. This architecture allows microrobots to learn efficient navigation strategies in complex environments while strictly enforcing collision-free behaviors, effectively addressing the black-box safety concerns of RL. Simulations in cluttered environments demonstrate that the proposed method achieves successful navigation with zero collisions. Furthermore, real-world experiments using a helical microrobot achieve safe autonomous navigation with zero collisions, verifying the framework's feasibility and robustness against physical uncertainties.

Keywords: Robotics, Sensor/Data Fusion, Intelligent and AI Based Control
Abstract: For target search tasks in unknown large-scale environments, existing pure vision-based methods are constrained by the narrow field of view of sensors, making it difficult to strike a balance between rapid spatial coverage and avoiding redundant searches. To address this, this paper proposes an efficient autonomous target search system for unmanned aerial vehicles based on the collaboration of an omnidirectional LiDAR and a restricted FOV camera. The system establishes a novel geometric-guided semantic dual-modal framework, fundamentally decoupling the spatial exploration and potential target search processes. First, an incremental dual-modal viewpoint extractor is designed to efficiently separate high-quality geometric exploration viewpoints and potential semantic preview viewpoints from the environmental point cloud. Based on this, a global search planner based on the Asymmetric Traveling Salesperson Problem is proposed, which innovatively introduces a unified cost matrix incorporating a fused yaw penalty and a hysteresis state machine to achieve optimal scheduling for the two types of heterogeneous viewpoints. Comparative experiments across multiple complex simulation environments demonstrate that the proposed framework not only secures an extremely high true target discovery rate with minimal computational overhead but also enhances macroscopic search flight efficiency, providing a novel paradigm that balances high discovery rates with high efficiency in practical search and rescue missions.

Keywords: Sensor/Data Fusion, Signal Processing
Abstract: Unmanned Aerial Vehicle (UAV) remote sensing technology, owing to its operational flexibility, low cost, and rapid deployment capability, has become an important tool for remote sensing monitoring. Among existing UAV payloads, visible-light cameras are widely used, whereas the application of hyperspectral imaging systems on UAV platforms remains relatively limited. To address this gap, a UAV-borne hyperspectral camera has been developed, and a method integrating block affine model–based registration with cloud-control-based stitching is proposed. For heterospectral registration, a block-based affine transformation strategy is adopted. Specifically, grid partitioning combined with local feature matching is used to estimate spatially varying affine transformations, effectively compensating for local geometric distortions and enabling high-precision registration of heterospectral images. For flight strip stitching, the concept of “cloud control” is introduced, in which existing geospatial datasets are utilized to perform feature matching with individual flight strips. Control points derived from this process are then used to refine the position and orientation system (POS) parameters of the strips, thereby enabling high-accuracy inter-strip stitching. To validate the effectiveness of the proposed method, experiments were conducted in the Yushu area of Jilin Province, China. The results demonstrate that the positioning accuracy achieved by the proposed approach is better than two ground sample distances (GSDs). These findings confirm that the developed hyperspectral camera and processing method effectively enhance UAV-based remote sensing capabilities and provide an important technical foundation for intelligent perception applications such as precision agriculture and environmental monitoring.

Keywords: Multi-agent Systems, Control Applications
Abstract: Coordinating multi-agent swarms for simultaneous target assignment and motion control typically relies on explicit inter-agent communication. However, in communication-denied scenarios, such as electromagnetic jamming or enforced radio silence, this reliance undermines system reliability. To address this, we propose a fully communication-free, decentralized framework that jointly performs target assignment and motion control for missions with strict target quotas. The approach relies solely on onboard sensing and operates in two stages: a flocking-based controller ensures cohesion and collision avoidance, followed by a local “counting region” mechanism that enables agents to infer task completion and autonomously select targets. Simulation results demonstrate that the proposed framework satisfies target quotas and achieves efficient, conflict-free convergence without negotiation or information exchange.

Keywords: Multi-agent Systems, Adaptive Control, Intelligent and AI Based Control
Abstract: Achieving robust formation navigation for Multi-Robot Systems (MRS) in complex and non-convex environments remains a fundamental challenge in modern robotics. Traditional Model Predictive Control (MPC) methods, which rely on fixed parameters and rigid topologies, are highly susceptible to local minima and navigation deadlocks when encountering dense obstacle traps or narrow apertures. To address these limitations, this paper proposes LLM-DTS, a hierarchical method that utilizes a Large Language Model (LLM) as a cognitive reasoner to perform environmental semantic analysis and dynamically reconfigure the underlying MPC topology and control parameters. The simulation results demonstrate that the proposed method increases navigation success rates across diverse scenarios and enables autonomous formation reshaping to traverse extreme spatial bottlenecks, enhancing the stability and environmental adaptability of the system.

Keywords: Networked Control, Nonlinear Systems and Control, Modeling and Control of Complex Systems
Abstract: The conventional assumption that all node states are fully measurable and exchangeable is generally unrealistic in practical systems. To overcome the issues of inaccurate convergence time in large-scale switched system stability theorems, unmeasurable node states, and potential chattering in system states, a hyperbolic sine function-based fixed-time aperiodic intermittent control (FTAIC) approach is developed to achieve output synchronization in multi-layer networks (MLNs). Firstly, a novel output-coupled MLN model is proposed, explicitly depicting the heterogeneity of intra-layer and inter-layer dynamical structures allowed by different layers and nodes. Secondly, the novel hyperbolic sine function-based FTAIC is proposed, aiming to mitigate the chattering effects associated with conventional FTAIC controllers that rely on sign functions. By applying the fixed-time (FT) switching stability theorem, a compact criterion for achieving fixed-time output synchronization (FOS) is derived, eliminating the restriction on the output matrix being positive-definite and diagonal. Finally, numerical examples are provided to validate the effectiveness of the proposed control design and the derived criterion.

Keywords: Multi-agent Systems, Networked Control
Abstract: With the widespread adoption of multi-agent systems in open communication environments, security threats have become increasingly severe. In mixed threat scenarios where Byzantine attacks and Denial of Service (DoS) attacks coexist, existing methods struggle to simultaneously address identification efficiency and control convergence speed while often relying on predefined reference points. This study proposes the Hybrid-Resilient Adaptive Consensus (HRAC) algorithm to secure multi-agent systems against mixed Byzantine and DoS attacks. The algorithm employs a dynamic reputation mechanism to identify malicious nodes in real-time, combined with event-triggered control to maintain communication efficiency during DoS attacks. Simulation results demonstrate that without requiring predefined reference points, the algorithm significantly improves convergence speed, outperforming traditional norm-based methods in both recovery efficiency and topology adaptability. This work provides an adaptive security framework for multi-agent coordination in complex attack environments.

Keywords: Nonlinear Systems and Control, Energy Efficiency, Optimal Control
Abstract: In reservoir hydropower inspection, mission feasibility is often jointly constrained by endurance-limited energy budgets and stringent safety requirements imposed by rugged terrain and restricted airspace. This paper considers multi-UAV inspection over terrain reconstructed from a digital elevation model (DEM), where onboard energy is limited and safety must be maintained both near no-fly zones (NFZs) and during close-proximity multi-UAV execution. A two-layer framework is presented. In the offline layer, a DEM-derived terrain graph is constructed, and an additive edge cost is formulated by combining edge-wise traversal energy with a clearance-based safety floor to discourage boundary-hugging solutions. Building on an energy-elevation-aware A∗ (EEA∗) search framework, an energy-safety-aware EEA∗ (ES-EEA∗) planner is developed, which retains elevation-aware node ordering to discourage costly climbs while balancing energy-safety trade-offs through the proposed scalarized edge cost. In the online layer, hard inter-UAV horizontal separation is enforced by a deconfliction mechanism in which impending conflicts are resolved through stepwise yield/hold actions with bounded local recovery replanning when required. Over N = 30 planning instances, ES-EEA∗ achieves a mean energy reduction of 3.16% relative to geometric A∗, with zero safety-floor violations in the tested set. In a two-UAV execution case, zero violation duration (Tvio = 0) is achieved with only a 0.75% energy overhead.

Keywords: Estimation and Identification, Intelligent and AI Based Control, Learning Systems
Abstract: Accurate remaining useful life (RUL) prediction is critical for the safety and reliability of lithium-ion batteries. However, industrial battery data often exhibit complex non-stationary degradation patterns and are contaminated by sensor anomalies and structural outliers, rendering conventional data-driven models unreliable. To address these challenges, this paper proposes a unified robust probabilistic framework: Matern-Kernel Deep Gaussian Processes with Student-t Likelihood (MK-DGP-t). First, we construct a multi-layer Deep Gaussian Process (DGP) employing the Matern kernel to effectively capture non-smooth temporal correlations and high-dimensional degradation features, overcoming the over-smoothing limitations of standard Radial Basis Function (RBF) kernels. Second, to mitigate the impact of outliers, we replace the Gaussian observation noise assumption with a heavy-tailed Student-t likelihood. This formulation naturally down-weights anomalous observations during the variational inference process, ensuring robust posterior estimation. Finally, we design a risk-aware decision mechanism that decomposes predictive uncertainty into epistemic and aleatoric components, enabling the identification of hazardous operating conditions. Validations on the NASA PCoE lithium-ion battery dataset (cell B0005) demonstrate that the proposed MK-DGP-t achieves superior RUL prediction accuracy and robustness compared to state-of-the-art shallow GPs and deterministic deep learning baselines, particularly in the presence of heavy-tailed noise.

Keywords: Robotics, Multi-agent Systems, Nonlinear Systems and Control
Abstract: This paper proposes an event-triggered nonlinear model predictive control (NMPC) framework integrated with a discrete-time control barrier function (DCBF) for the trajectory tracking and safety-critical control of four-wheel independent steering and independent driving (4WIS-4WID) formation systems. To achieve the optimal performance while ensuring that the critical system variables never violate predefined constraint boundaries, the DCBF constraints are enforced over the prediction horizon to guarantee set forward invariance and safety properties including obstacle avoidance. An event-triggered mechanism is incorporated into the NMPC optimization problem to determine the next triggering instant, thereby significantly reducing the computational burden and conserving network resources without compromising closed-loop safety or performance. Under the proposed scheme, the optimization problem is recursively feasible, and closed-loop stability is established. Finally, simulations confirm the effectiveness of the proposed strategy.

Keywords: Multi-agent Systems, Robotics, Networked Control
Abstract: In our previous work, a virtual tube is proposed to guide a robotic swarm to pass through a cluttered environment. In this work, we use the null-space-based (NSB) method to design a passing-through controller for the robots. Firstly, the virtual tube is modeled in a Frenet-Serret frame using appropriate mathematical descriptions. Then, a distributed swarm controller is proposed based on the NSB method. We make the subtask of avoiding collision between robots have a higher priority, and the subtask of passing through the virtual tube has a lower priority. As the subtask of keeping within the virtual tube is considered a hard constraint, it has the same priority as the task of avoiding collision between robots. Finally, the effectiveness of the proposed method is validated through a numerical simulation.

Keywords: Man-machine Interactions, Multi-agent Systems, Modeling and Control of Complex Systems
Abstract: This paper proposes a distributed strategy for one-to-many unmanned aerial vehicle (UAV) transport, in which a single human operator can manage multiple UAVs in a semi-autonomous mode. Cooperative transport control and virtual tube passage control are integrated within the virtual system, allowing UAVs to safely get through pre-planned virtual tubes. A single human controller enters commands through the keyboard, and the router broadcasts the velocity scaling factor and desired rotation angle to all UAVs. The swarm executes keyboard commands for velocity adjustment along the tube, collective rotation, and collective return, which realize more diverse collective motion modes. Simulation with ten UAVs and outdoor experiments with three UAVs are conducted in two types of tubes: with a straight generator curve and with an irregular generator curve. The results indicate that the proposed strategy improves the consistency and adaptability of UAV swarms when unexpected situations arise.

Keywords: Multi-agent Systems, Robotics, Motion Control
Abstract: 无人机（UAV）的快速部署 密集城市环境中的群体需要 导航基础设施确保计算能力 效率和操作安全。本文提出 虚拟管道生成自动化框架 基于三维拓扑重建的网络。首先，一个 基于网格的广义沃罗诺伊图（GVD）被用于 从离散高度切片中提取骨骼骨架， 最大化与建筑障碍物的净空。其次， 我们引入了 的层级规划逻辑 异构空间，将二维平面拓扑桥接成 通过最小化水平实现统一的三维流形 位移关联。最后，参数虚拟 管子通过计算边界拟合合成 沿三维骨干的包络线，提供拓扑上的 保证群体导航的安全体积。模拟 结果是障碍物密度各异的环境 证明在拟议方法保持 在路径长度上

Keywords: Multi-agent Systems, Robotics, Control Applications
Abstract: Intersections are inevitable bottlenecks in emerging low-altitude air traffic networks that adopt a ``sky highway'' paradigm, where traffic is organized within structured virtual tubes to enable scalable, high-density operations. Among various junction patterns, the Y-shaped intersection is a fundamental trunk--branch primitive that supports both merging (multiple inbound branches joining a trunk airway) and diverging (a trunk flow splitting into destination-dependent branches). This paper develops a distributed control strategy for cooperative UAVs traversing a Y-shaped intersection under virtual tube constraints and inter-agent collision avoidance. Specifically, a feasible segmentation and geometric construction method for the Y-shaped intersection is provided, together with a switching control strategy that integrates the established straight-tube and trapezoid-tube distributed controllers. Numerical simulations are presented to validate the effectiveness of the proposed method.

Keywords: Multi-agent Systems, Robotics, Motion Control
Abstract: Inspired by hierarchical planning in intelligent vehicles—where offline route selection precedes online lane selection—this paper presents a dual-layer safety tube framework for distributed UAV swarms. It comprises: (i) an offline-constructed global virtual tube providing collision-free corridors in voxel maps, and (ii) online-updated individual prediction tubes that envelope closed-loop trajectories under frozen references. Environment and inter-UAV collision avoidance are respectively enforced via tube containment and separation conditions. To ensure explicit realizability, we introduce a pre-stabilized controller with an Explicit Reference Governor (ERG) that regulates an auxiliary reference to enforce safety constraints. The ERG integrates a navigation field for tube progression while avoiding boundaries and neighbors, along with a dynamic safety margin that scales reference derivatives based on minimum distances to tube walls and neighboring tubes. Requiring no large-scale optimization and communicating only low-dimensional tube parameters, the framework achieves low computational and communication overhead. Simulations in realistic underground garage environments demonstrate safe, efficient swarm navigation through narrow ramps and consecutive sharp corners while maintaining prescribed safety distances.

Keywords: Multi-agent Systems, Control Applications, Real-time Systems
Abstract: Unmanned aerial vehicle (UAV) swarms have attracted increasing attention because of their superior efficiency and robustness. However, safe and agile formation flight in cluttered environments remains a critical challenge. Existing methods offer partial solutions, yet still suffer from limited adaptability, high computational or communication costs, and a predominant focus on two-dimensional scenarios with pre-defined virtual tube information. To address these limitations, this paper proposes a dynamic virtual tube-oriented control method for UAV swarms in three-dimensional environments. The proposed approach first guides the leader UAV along a predefined route using the carrot-chasing algorithm. Then, a reference generating line is established from the leader’s traveled trajectory, based on which a 3D virtual tube is dynamically constructed using preset tube parameters. Finally, by integrating the dynamic virtual tube with leader-follower consensus control, a controller law is designed for follower UAVs, enabling them to maintain formation while remaining strictly inside the generated tube. Compared with existing methods, the proposed strategy extends virtual tube control from 2D to 3D space, and reduces dependence on pre-defined tube information. Simulation results demonstrate that the proposed method successfully achieves stable UAV formation flight within dynamically generated virtual tubes.

Keywords: Automated Guided Vehicles, Flexible Manufacturing Systems, Robotics
Abstract: Deadlock is a major bottleneck in heterogeneous robot fleet scheduling for smart-factory intralogistics, where kinematic asymmetry, battery-induced timing drift, and class-dependent resource holding exacerbate circular waits. This paper proposes a closed-loop prediction–prevention–resolution framework within a receding-horizon architecture. The core is a Heterogeneity-Aware Deadlock-Resilient Model Predictive Control (HA DRMPC) scheme that embeds calibrated multi-horizon deadlock-risk forecasts into constrained optimization, enabling adaptive rerouting, reservation adjustment, and task resequencing under explicit risk budgets. A conjunctive deadlock confirmation logic—combining risk exceedance, resource-allocation graph(RAG) cycle detection, and temporal persistence—triggers an event-driven recovery MPC that guarantees monotonic deadlock dissolution with minimal schedule disruption. Experiments on a high-fidelity digital twin (750-node factory map, four commercial robot types) show that under high-density (50 robots) and high-disturbance (≤130 ms latency) conditions, the proposed framework reduces mean time to recovery by 40.0% and blocking rate by 53.7% relative to the industrial baseline RCS 2000, while maintaining a normalized recovery rate above 94%. These results demonstrate a scalable and disturbance robust solution that shifts deadlock handling from passive mitigation to proactive, risk regulated control for heterogeneous robot fleets.

Keywords: Modeling and Control of Complex Systems, Learning Systems, Nonlinear Systems and Control
Abstract: To address the degradation in prediction accuracy of electric vehicle energy consumption under low sampling rates, this paper proposes an energy consumption modeling approach that integrates vehicle dynamics mechanisms with machine learning. First, multi-source time-series data were collected through real-world on-road vehicle experiments, followed by a resampling analysis across different sampling rates. Subsequently, a physically interpretable feature framework was developed to provide a unified representation of propulsion energy consumption, regenerative braking energy recovery, and auxiliary system energy use. Based on this framework, linear regression, neural network, support vector machine, and Gaussian Process models were trained and validated. The results demonstrate that, over a sampling interval range of 1 s to 60 s, the proposed method maintains stable and high-accuracy predictive performance, achieving a MAPE of approximately 3%–4%, which significantly outperforms conventional statistical feature-based models. This study provides an effective pathway for energy consumption evaluation and energy-efficient control of electric buses under low-frequency data conditions.

Keywords: Robotics, Real-time Systems
Abstract: Dangerous driving behaviors pose significant threats to traffic safety, especially on highways. Timely identification and risk assessment of such behaviors enable authorities to issue early warnings, implement targeted interventions, and ultimately enhance overall road safety. However, existing approaches tend to emphasize the identification of dangerous driving behaviors, yet fall short in assessing the degree of risk these behaviors entail. This gap limits their applicability in proactive traffic safety management and targeted risk mitigation. To address this issue, we propose a prediction-based driving risk assessment framework assisted by UAVs. Specifically, aerial imagery is utilized to capture vehicle trajectories rapidly. A hierarchical Transformer model is then employed to perform uncertainty-aware trajectory prediction. Based on the predicted trajectories, we introduce two risk assessment metrics—potential collision probability and potential collision severity—to estimate the potential driving risk associated with different driving behaviors. Our method is evaluated on the AD4CHE naturalistic highway driving dataset through the extraction and analysis of characteristic driving behaviors. The system has the potential to realize vehicle behavior monitoring and risk identification for specific areas in highway scenarios.

Keywords: Robotics, Signal Processing, Sensor/Data Fusion
Abstract: Multi-view stereo (MVS) has achieved significant advances in dense depth estimation and reconstruction by constructing differentiable cost volumes and learning to regularize them. However, cross-view geometric consistency is still predominantly enforced through post-hoc depth filtering, where geometrically inconsistent estimates are removed after inference, often resulting in incomplete point cloud reconstruction. This paper proposes a geometric consistency learning approach that explicitly incorporates cross-view geometric constraints into the training objective, thereby reducing reliance on post-hoc filtering. Specifically, the proposed method formulates cross-view geometric consistency directly in 3D point space by back-projecting image pixel coordinates of the reference and adjacent source views using predicted and ground-truth depths, followed by point alignment under known camera poses, enabling explicit evaluation of geometric discrepancies across views. Based on the resulting point-wise geometric residuals, a differentiable geometry-aware weighting mechanism is introduced to suppress geometrically unreliable depth hypotheses during training, rather than discarding them after inference. Extensive experiments demonstrate that the proposed approach yields consistent improvements in both reconstruction accuracy and completeness, achieves state-of-the-art performance on standard MVS benchmarks when integrated with complementary components, and further demonstrates its effectiveness in real-world outdoor scenarios.

Keywords: Nonlinear Systems and Control, Automated Guided Vehicles, Motion Control
Abstract: Path tracking for autonomous vehicles remains challenging when reference trajectories exhibit curvature discontinuities, such as sharp corners or interrupted segments. Directly following these non-smooth inputs often results in tracking instability or aggressive steering maneuvers. To address this, we propose an Artificial-Reference MPC (AR-MPC) framework. The core of this approach is the introduction of an artificial reference state and its corresponding input as additional decision variables within the optimization problem. In this way, the artificial reference is governed by the vehicle’s kinematic constraints and is optimized online to serve as a smooth, reachable intermediate target. To further refine tracking performance, a path segmentation strategy is incorporated to categorize road geometries into typical scenarios based on their geometric features. This allows the controller to employ scenario-specific weighting matrices, ensuring appropriate control efforts for different maneuvers. Co-simulations conducted in MATLAB/Simulink and CarSim environment demonstrate that the proposed AR-MPC significantly reduces tracking errors and enhances motion smoothness. Real-world experiments on the QCar platform further validate the effectiveness of the proposed approach.

Keywords: Robotics, Control Applications, Learning Systems
Abstract: Underwater 3D reconstruction is critical for marine exploration and remains severely constrained by visual degradation caused by wavelength-dependent attenuation and volumetric scattering. Although geometry-based and neural rendering reconstruction paradigms have advanced substantially in terrestrial settings, their performance degrades considerably underwater due to the lack of unified multi-view datasets with physically consistent degradation modeling and reliable multi-view geometric supervision. In this paper, we introduce UW3D, a unified multi-view dataset and benchmark for underwater 3D reconstruction that integrates physics-consistent synthetic degradations with real-world underwater multi-view sequences within a standardized evaluation framework. The synthetic component is constructed using a revised underwater image formation model that explicitly characterizes attenuation and backscatter while preserving strict multi-view geometric consistency inherited from high-fidelity in-air multi-view data, providing metric depth maps, ground-truth point clouds, and surface normal maps derived via local plane fitting with edge-preserving smoothing for supervised geometric learning and controlled degradation-aware evaluation. The real-world component comprises multi-view sequences captured under varying turbidity and illumination conditions. To address initialization instability under severe optical degradation, we adopt a globally optimized pose–structure estimation strategy that jointly refines camera poses and sparse geometry using dense geometry-aware correspondences and globally consistent optimization, establishing reliable camera–scene configurations for downstream reconstruction. UW3D enables consistent evaluation across geometry-based and neural rendering-based reconstruction paradigms on both synthetic and real-world datasets. Extensive benchmarking results and ROV-based reconstruction experiments demonstrate practical applicability in industrial scenarios.

Keywords: Multi-agent Systems, Nonlinear Systems and Control, Control Applications
Abstract: This paper explores the affine formation maneuver control problem for multi-autonomous surface vehicle systems under complex practical environment. A stress matrix-based approach is employed to define the formation geometry, enabling flexible affine maneuvers. To effectively suppress the influence of disturbances in dynamic maritime environments, a supertwisting sliding mode control strategy is proposed. Firstly, the control strategy ensures that the followers can track the affine transformations determined by the leaders, while mitigating the chattering phenomenon in traditional sliding mode control. Secondly, a Lyapunov function is constructed to analyze the stability of the system, guaranteeing that the formation tracking errors are uniformly ultimately bounded. Finally, the effectiveness of the proposed strategy is verified through simulation studies. The simulation results demonstrate that under the proposed super-twisting sliding mode control strategy, the ASV formation maintains high tracking precision and successfully executes affine maneuvers for obstacle avoidance.

Keywords: Multi-agent Systems, Modeling and Control of Complex Systems, Intelligent and AI Based Control
Abstract: 消防安全受领导层、管理层、员工、设施和环境之间的非线性相互作用影响，传统静态评估方法无法充分捕捉这些互动。现有实践主要依赖定性检查和静态评分模型，仅提供快照评估，忽视了多种因素积累形成系统性火灾风险的微观机制。本文开发了一个火灾复杂自适应系统智能体模型（FCAS-ABM），其中五个子系统被表示为异构智能体，特点是随机演化、非线性反馈和否决阈值。基于该模型，构建了一个包含动态平滑、分级和趋势导向预警的综合系统安全指数。该模型基于专家知识和工程数据进行校准，并通过五年离散时间模拟实现，以考察基线动态和替代治理策略

Keywords: Sensor/Data Fusion, Multi-agent Systems, Learning Systems
Abstract: Resilient collaborative perception is pivotal for near-space vehicle (NSV) swarms executing missions such as wide-area earth observation. However, the system-wide perception resilience is threatened by non-uniform vulnerabilities stemming from platform/sensor heterogeneity and vast variations in observed targets (e.g., in scale and morphology). Systematic assessment and lightweight enhancement of such resilience remain under-explored. To this end, we propose a holistic resilience evaluation and enhancement framework for NSV swarm perception. Our framework quantifies collaborative perception fragility by instantiating heterogeneous virtual perception nodes with an advanced object detector and diagnosing their performance disparities on the DOTA-v1.0 benchmark, which features diverse aerial scenes and targets from vehicles to ships. Based on the profiling, we devise a lightweight intraswarm attentive enhancement mechanism. Without modifying the base detection architecture, this mechanism adaptively reinforces the perception of vulnerable target categories by simulating principles of attentional coordination and feature fusion within a cluster. Experiments on DOTA-v1.0 demonstrate that our mechanism effectively boosts the overall detection accuracy and, more importantly, homogenizes performance across target categories, thereby enhancing systemic resilience. This work provides a quantifiable assessment tool and a plug-and-play enhancement solution for designing collaborative perception systems in NSV swarms, with potential implications for broader multi-agent systems.

Keywords: Intelligent and AI Based Control, Signal Processing, Estimation and Identification
Abstract: With the proliferation of unmanned aerial vehicles (UAVs), non-cooperative target detection has become a critical security concern. Traditional radar and optical methods often fail to detect small-scale UAVs in complex environments. This paper proposes a novel RF fingerprint identification technique leveraging the YOLO26 architecture. By transforming raw I/Q signals into high-resolution time-frequency spectrograms, we treat signal identification as a specialized object detection task. The YOLO26 model is optimized with the loss function to improve sensitivity to narrow-band frequency hopping signals. Experimental results demonstrate that the proposed method achieves a Mean Average Precision (mAP50) of 98.1% on public datasets. Furthermore, the system delivers a real-time inference speed of 21.73 FPS on the Huawei Ascend 310B AI chip, which verifies its effectiveness for edge deployment in low-altitude security scenarios.

Keywords: Multi-agent Systems
Abstract: Coordination in multi-agent systems offers a powerful paradigm for distributed optimization in unknown environments, yet extending these principles to high-dimensional parameter space optimization remains challenging. This paper proposes a two-stage hybrid variant of the Speeding-Up or Slowing-Down (SUSD) algorithm for derivative-free optimization where gradients are ill-defined. The proposed method first employs a weighted centroid direction (WCD) for efficient exploration and fast initial descent, then switches to the original PCA-based direction for refined local convergence. A convergence analysis establishes that the WCD direction locally aligns with the negative gradient under mild conditions, while the PCA stage inherits the convergence properties of the original SUSD. Numerical experiments on data-driven LQR problems and high-dimensional Rosenbrock functions demonstrate that the hybrid method achieves faster initial progress and lower per-iteration compared to the original SUSD. The result is a theoretically grounded derivative-free optimizer bridging multi-agent coordination with high-dimensional optimization.

Keywords: Modeling and Control of Complex Systems, Multi-agent Systems, Optimal Control
Abstract: Unmanned aerial vehicles (UAVs) have become indispensable assets for close-ground surveillance and strike operations, playing a vital role in modern military applications. This paper addresses the problem of multi-UAV task assignment within command and control (C2) systems. While existing research has explored various assignment models, limited attention has been given to dynamic task allocation for multi-UAVs. Conventional approaches often fail to account for voyage constraints when assigning multiple tasks to a single UAV. To overcome these limitations, we enhance the classical weapon target assignment (WTA) model by incorporating linearized voyage constraints within a mixed-integer linear programming (MILP) framework. Specifically, we establish a fuel–time relationship and develop a linear route estimation method to compute the voyage requirements for UAVs. The fuel–time relationship is validated through flight experiments, and numerical simulations demonstrate that the proposed MILP formulation outperforms existing methods, confirming its effectiveness in dynamic mission scenarios.

Keywords: Optimal Control, Multi-agent Systems
Abstract: This work investigates a prescribed-time observer-based human-in-the-loop (HiTL) optimal output tracking control problem for multiagent systems. The HiTL method facilitates the intervention and guidance provided by the human operator, thereby avoiding imminent dangers. To address the problem that the leader's information is unavailable to each follower, a distributed observer with prescribed-time convergence is designed to enable accurate estimation within a user-defined time. Subsequently, by combining the follower dynamics with this observer, the HiTL output tracking problem is reformulated into an optimal linear quadratic tracking problem. Meanwhile, the nonhomogeneous algebraic Riccati equations (AREs) are established to solve this problem. Finally, numerical simulations verify the effectiveness of the proposed control scheme.

Keywords: Robotics, Multi-agent Systems
Abstract: In unknown and complex environments, relying on a single or homogeneous robotic system often leads to inefficiency or even task failure, particularly when dealing with tasks governed by multiple objectives and constraints in temporal logic. These challenges become more pronounced for task assigners as robots enhance their capabilities during task execution. To address these issues, this paper presents a double reactive planning framework tailored for heterogeneous robotic systems. The framework consists of two main modules: offline task pre-allocation and online reactive allocation and planning. Offline task pre-allocation assigns tasks to drones and quadruped robots based on robot capability and task decomposition specifications. Meanwhile, online reactive allocation and planning updates the knowledge base by sharing perceptual information, incorporates internal logical relationships among task atomic propositions, and adjusts task assignments in response to changes in the robots’ capabilities. This ensures task completion while adhering to temporal logic constraints. To validate the framework’s effectiveness, we conducted theoretical analyses and simulation experiments. The results show that our approach successfully coordinates the reassignment of both local and global tasks, replanning movements to overcome the challenges posed by unknown environments while maximizing the efficiency and flexibility of heterogeneous robotic teams.

Keywords: Robotics, Multi-agent Systems
Abstract: Existing search-based task planning methods for multi-agent systems face computational intractability when simultaneously exploring sub-tasks and workspaces. This paper presents an asymptotic optimal search (AOS) method for multi-agent systems with Linear Temporal Logic (LTL) task specifications. AOS constructs a planning tree by searching over sub-tasks while deriving system states through iterative computation of sub-task completion positions and times. This approach decouples planning from workspace exploration, significantly reducing computational complexity. The method first generates a locally optimal plan rapidly, then systematically expands unexplored nodes to ensure global optimality. Sub-trees generated from unexplored nodes are pruned based on current minimum cost, restricting search space without eliminating optimal solutions. The minimum cost is updated asymptotically during sub-tree expansion, further constraining tree depth. Theoretical analysis proves AOS maintains completeness and optimality while achieving per-node linear complexity with respect to agent count, enabling practical scalability for large multi-agent systems. Simulation and experimental results demonstrate that AOS achieves superior computational efficiency for large-scale multi-agent systems with LTL-based tasks, solving problems with thousands of agents within seconds.

Keywords: Robotics, Multi-agent Systems, Learning-based Control
Abstract: Autonomous multi-robot exploration in unknown environments requires both rapid coverage and reliable coordination. Traditional frontier-based methods are computationally efficient, but they often suffer from redundant exploration and limited adaptability in complex multi-robot settings. To address these limitations, we propose FA-MARL, a decentralized framework that integrates frontier assignment with multi-agent reinforcement learning (MARL) under the centralized training and decentralized execution (CTDE) paradigm. In the proposed method, each robot learns an adaptive frontier evaluation strategy to balance travel distance and information gain, while a multi-level reward guides the team toward efficient coverage and suppresses redundancy. To support practical deployment, FA-MARL is implemented in a modular pipeline that combines Gmapping for mapping, A* for global planning, and DWA for local control. Simulation and real-world experiments with TurtleBot2 robots demonstrate that FA-MARL outperforms heuristic baselines in both exploration efficiency and stability, while remaining effective under asynchronous execution and varying team sizes.

Keywords: Automated Guided Vehicles
Abstract: Navigating complex interactive scenarios on highways requires autonomous vehicles to simultaneously ensure strict collision avoidance and seamless integration into natural traffic flows. Since traditional planning models struggle to capture the implicit cooperative dynamics inherent in human driving, this paper proposes a data-driven trajectory planning framework based on Maximum Entropy Inverse Reinforcement Learning. The framework adopts a practical generate-evaluate-select paradigm that explicitly decouples hard safety constraints from soft behavioral preference evaluation. To accurately characterize multi-vehicle interactions, we formulate a comprehensive spatial-temporal feature representation, notably introducing interaction risk and social adaptability features to explicitly quantify spatial safety margins and temporal velocity coordination. Evaluated on the real-world NGSIM dataset, the proposed approach significantly outperforms representative rule-based and expert-tuned baselines by achieving the lowest trajectory deviation and the highest interactive safety margins. These results indicate that the explicit integration of social adaptability and risk perception enables the planner to effectively handle complex multi-vehicle interactions, successfully reproducing naturalistic, safe, and comfortable driving behaviors.

Keywords: Intelligent and AI Based Control, Adaptive Control, Motion Control
Abstract: Autonomous Underwater Vehicles (AUVs) can autonomously complete environmental perception, positioning, analysis, and decision-making in complex environments, and play an important role in marine scientific research and resource exploration. This paper considers an underactuated AUV, which reduces mechanical complexity, weight, and energy consumption. However, the underactuated structure introduces strong nonlinear coupling in motion, complicating path planning and trajectory tracking. To address this issue, this paper introduces a kinematic-aware motion planning method together with a reinforcement-learning-based adaptive control scheme. The planning module employs an improved Rapidly-exploring Random Tree (RRT) algorithm that explicitly incorporates the kinematic constraints of the AUV, thereby generating smooth and dynamically feasible trajectories with reduced path length and energy consumption. On the control side, an adaptive tracking controller is developed, where reinforcement learning dynamically tunes the parameters of a PID structure. This design leverages the interpretability and robustness of classical PID control while enhancing adaptability and stability under environmental uncertainties. Finally, extensive simulations and comparative studies are conducted to validate the effectiveness of the proposed framework in deep-sea environments.

Keywords: Intelligent and AI Based Control, Robotics, Man-machine Interactions
Abstract: Fine-grained gait phase prediction from surface electromyography can provide richer timing cues than a coarse stance/swing split, yet its reliability across locomotion modes remains insufficiently characterized at the sub-phase level. This paper studies cross mode generalization under an interval-based five-segment formulation using SIAT-LLMD. Level walking is annotated into five consecutive gait-cycle intervals, while stair ascent and stair descent are treated as mode-specific subsets within a compatible interval template. Multi-channel sEMG is segmented by sliding windows, encoded by compact time-domain features, and classified using a feed-forward artificial neural network. We report within-mode performance and direct transfer, where we train on a source mode and test on a target mode without calibration. Results show strong separability within each mode, whereas cross mode deployment degrades substantially, particularly between level walking and stair ambulation, with errors concentrated near swing-related intervals. In contrast, transfer between stair ambulation is comparatively stable, suggesting closer neuromuscular signatures within stair ambulation. Overall, the findings indicate that fine-grained phase boundaries are highly mode-dependent, motivating context-aware modeling and domain adaptation for multi-environment deployment.

Keywords: Modeling and Control of Complex Systems, Man-machine Interactions, Adaptive Control
Abstract: To address the issues of lower limb motor function decline and increased metabolic burden in patients with sarcopenia and myasthenia gravis, traditional rigid exoskeletons often cause significant physical conflict or misalignment at the human-robot interface due to their large mass and joint axis misalignment. In this study, a lightweight flexible assistive hip exoskeleton was developed, and a hierarchical control strategy based on human-robot interaction was proposed. The hierarchical strategy employed a dual-layer architecture: the perception layer utilized an adaptive Hopfield oscillator to achieve accurate gait phase estimation, while the control layer implemented a variable parameter admittance algorithm based on a variant of the Sigmoid function combined with a feedforward compensation mechanism. This approach addressed the inherent hysteresis and nonlinearity of flexible transmissions, enabling adaptive adjustment of stiffness and damping parameters. Experimental results demonstrated that the system exhibited excellent force-tracking performance, with a mean root mean square error (RMSE) of 9.79~N and an average peak delay of 0.058~s. Metabolic cost evaluations indicated that the proposed dynamic parameter adjustment strategy provided more significant assistance than fixed-parameter strategies, reducing the subjects' metabolic cost by up to 11.2%. The system showed high potential for clinical promotion and application in pathological gait rehabilitation.

Keywords: Robotics
Abstract: Cognitive state assessment in functionally impaired populations relies heavily on subjective clinical rating scales, which limits continuous monitoring and early warning. To address this limitation, this study proposes an

electroencephalography (EEG)-based framework for early prediction of cognitive states. Using the intensive care unit (ICU) as a representative application scenario, continuous multichannel EEG signals are modelled through a sliding￾window strategy, and multi-scale structured time–frequency–modality features are constructed by combining local mean decomposition and short-time Fourier transform. On this basis, a cooperative prediction architecture is developed by integrating window-level feature encoding with cross-window temporal modelling, enabling prospective inference of future cognitive states. Experiments conducted on real ICU patient data demonstrate that, in short-term prediction scenarios, the proposed method significantly outperforms multiple temporal

baseline models in terms of accuracy, class balance, and cross￾fold stability. Furthermore, confusion matrix analysis indicates that prediction errors are predominantly confined to adjacent cognitive levels, which is consistent with the continuous evolution of cognitive states. These results confirm the feasibility of short￾term cognitive state prediction using limited historical EEG signals and provide an effective technical pathway for prospective monitoring and early warning in ICU settings.

Keywords: Sensor/Data Fusion, Robotics, Man-machine Interactions
Abstract: ，中风常导致运动功能障碍和步态 受损，显著影响患者的行动能力 生活质量。功能性电刺激（FES） 已经是 广泛应用于神经康复，激活瘫痪者 肌肉 并促进运动恢复。然而，实现协调性 控制多重下肢关节依然具有挑战性 应得 到复杂的肌肉骨骼动力学。本研究提出了一个 集成肌肉骨骼模型的控制框架 其中 强化学习以产生肌肉刺激 信号 用于协调下肢运动。肌肉骨骼 模型模拟髋关节、膝盖和脚踝的动态变化 关节，而强化学习则用于学习 最优 关节控制的刺激策略。模拟结果 节目 即所提方法能够实现 在有限空间系统驱动下的三个节点。均方根 髋关节、膝关节和踝关节的误差

Keywords: Modeling and Control of Complex Systems, Optimal Control, Intelligent and AI Based Control
Abstract: This paper proposes a hybrid intelligent control strategy for electrostatic precipitators (ESPs) used in industrial gas cleaning systems. The approach integrates a Multiple-Input Multiple-Output (MIMO) dynamic model, decoupling techniques, PID controllers, and fuzzy logic control to enhance gas distribution efficiency and regulate electrical operating conditions. The ESP system exhibits strong cross-coupling among gas flow channels and nonlinear dependencies in electrical parameters, complicating conventional control design. To address this, a MIMO model with cross-transfer functions is developed to capture system dynamics accurately. Initially, PID controllers are designed for decoupled subsystems. The interaction effects are then compensated to mitigate cross-channel coupling. To improve robustness under uncertain and time-varying operating conditions, a fuzzy logic controller is incorporated to regulate electrical operating conditions adaptively. Simulation results demonstrate significant performance improvements, including the elimination of overshoot, a reduction in steady-state error, and an enhanced transient response. The proposed method achieves an 8.5% relative efficiency improvement over the baseline control strategy. Overall, the proposed hybrid strategy enhances the energy efficiency, robustness, and adaptability of ESP operation, making it suitable for real-time industrial implementation.

Keywords: Robotics, Sensor/Data Fusion, Adaptive Control
Abstract: In the rapidly advancing domain of soft robotics, continuous progress in material science, structural design, and theoretical approaches has accelerated the development of soft robotic systems, which are increasingly evolving toward flexible and modular architectures with broad industrial applications. Nevertheless, a key challenge in this field remains the accurate representation of shape-morphing behavior, as existing visualization and simulation tools are limited in their ability to capture the complex, continuous deformations inherent to soft robots. In addition, there is a shortage of intuitive and user-friendly platforms that support effective visualization and interactive control of these adaptive systems. To address these limitations, this study proposes a novel digital twin (DT) framework for reconfigurable soft robots within an augmented reality (AR) environment. The proposed system enables more precise and natural visualization of three-dimensional soft deformations while offering an intuitive simulation interface. A parameterized curve-based approach is employed to dynamically update the digital twin in AR, ensuring smooth transitions across different shape-morphing states. Three primary deformation modes—stretching, bending, and twisting—are identified and supported by advanced visualization techniques for accurate representation. Furthermore, sensor fusion is integrated to capture real-time structural changes of the soft robot and translate them into parameterized curves. The system operates entirely within an AR environment, allowing users to perform immersive analysis and simulate the reconfiguration of real-world soft robotic systems.

Keywords: Networked Control, Sensor Networks, Signal Processing
Abstract: This paper addresses the problem of dynamic radio resource management in heterogeneous 5G networks (HetNets). Radio Resource Management (RRM) involves a complex multidimensional optimization problem in the context of high-density small-cell infrastructure, interference, and backhaul constraints. The study proposes a joint model that integrates the processes of resource allocation (RA), user association (UA), and power allocation (PA). The proposed model is described based on queue dynamics, the SINR metric, and Shannon’s capacity formula, while network stability is ensured using the Lyapunov method. Numerical simulation results showed that the proposed dynamic control method reduces the average queue length by 30–40% and increases network throughput compared to static scheduling algorithms. Furthermore, accounting for backhaul constraints allows for a more accurate assessment of network performance. The results confirm the importance of joint optimization approaches for effective radio resource management in 5G networks.

Keywords: Man-machine Interactions, Motion Control, Robotics
Abstract: Muscle coactivation, quantified by the coactivation index (CI) , is a fundamental mechanism in human motor control and also provides important bioinspiration for robot motion control. Understanding the muscle coactivation characteristics of professional athletes' limb joint movements is critical for advanced robot motion control strategies. However, most studies have focused on the general population, with limited attention to elite athletes and insufficient consideration of multiple joints and sports, leaving cross-joint and cross-sport differences poorly understood. This study investigated muscle coactivation in rock climbing and taekwondo athletes. Seven elite rock climbers and nine elite taekwondo athletes performed elbow and knee extension and flexion tasks under isokinetic and isotonic conditions while surface electromyography was recorded to calculate CI. Multivariate analysis of variance revealed that elbow coactivation in taekwondo athletes was significantly influenced by exercise mode and movement type, with higher CI observed during flexion and under isotonic conditions. Knee coactivation was primarily affected by movement type within individual sports, whereas team differences emerged in pooled analyses, with opposite flexor–extensor CI patterns between sports. Pooled analysis of dominant-side joint revealed higher CI in the elbow than the knee and significant interactions between team, movement type, and joint. These findings highlight the complexity and sport-specificity of muscle coactivation and provide a theoretical basis for applications in robot motion control and motion skill training.

Keywords: Robotics, Man-machine Interactions
Abstract: This work introduces a multipoint fluidic haptic device designed to improve performance in teleoperation tasks. The system features 16 taxels actuated by electroosmotic pumps, providing precise and localized haptic feedback while maintaining compatibility with commercially available hand-tracking interfaces. The proposed teleoperation system integrates a closed-loop feedback system, where a tactile sensor mounted on the robotic arm captures remote object surfaces and conveys them to the operator through the haptic display. This work highlights the potential of fluid-based haptic systems for advancing teleoperation and remote manipulation applications, offering a foundation for further exploration of tactile feedback in human-robot interaction.

Keywords: Robotics, Sensor/Data Fusion, Signal Processing
Abstract: Lateral movement training is essential for strengthening hip abductor muscles and improving pelvic stability. While lateral walking on level ground has been extensively studied, lateral stair ascending—a critical activity of daily living—presents substantially more complex dynamic challenges due to gravity, joint loading, and postural instability. In this paper, we investigate gait phase recognition for lateral stair ascending using wearable inertial sensors and propose a hybrid deep learning framework, termed TCNSE- LSTM. We construct a high-density dataset using 14 Inertial Measurement Units (IMUs) mounted on key lowerlimb muscle groups of 12 healthy subjects. Each gait window is represented by a compact 672-dimensional engineered feature sequence extracted from 84 raw IMU channels. The proposed network integrates: (i) a TCN-style dilated convolutional backbone for multi-scale feature interaction along the engineered feature sequence; (ii) a 1-D Squeezeand-Excitation (SE) channel attention module for adaptive reweighting across learned channels; and (iii) a Long Short-Term Memory (LSTM) layer for modeling long-range dependencies in the refined feature sequence prior to classification.Experimental results show that the proposed method achieves an overall accuracy of 97.50%, significantly outperforming standard CNN, LSTM, and dilated-convolution baselines under the same data processing and training protocol.In particular, our model exhibits superior robustness in identifying the highly unstable Split-step Double Support(SPDS) phase, where the body’s center of mass spans two stair steps. These findings indicate that the proposed framework provides a reliable sensing front-end for precise control of lower-limb exoskeletons operating on unstructured stair terrains.

Keywords: Real-time Systems, Man-machine Interactions, Signal Processing
Abstract: Surface electromyography (sEMG) offers a non-invasive interface for prosthetic hand control, yet clinical adoption remains hindered by poor robustness to electrode shift, inter-subject variability, and muscle fatigue. This study proposes a real-time prosthetic control framework based on muscle synergy decomposition of forearm sEMG signals to address these limitations. Muscle activation levels were extracted from 16-channel sEMG recordings, and a fixed synergy matrix was obtained via non-negative matrix factorization (NMF) during an offline calibration phase. Real-time synergy activation coefficients were computed via pseudo-inverse and mapped to continuous control commands. A support vector machine (SVM) classifier was trained on time-domain features extracted from these coefficients. Experiments with six able-bodied subjects demonstrated that three muscle synergies consistently explained over 90% of data variance, with inter-subject cosine similarity exceeding 0.8. Offline classification achieved 93.03% mean accuracy across five gestures. In real-time object manipulation tasks, the system enabled successful power grasp, lateral pinch, and three-finger grasp with low latency and no critical control failures. These results validate that synergy-based decomposition provides a physiologically interpretable, computationally efficient, and robust solution for intuitive multi-functional prosthetic hand control.

Keywords: Robotics, Sensor/Data Fusion, Sensor Networks
Abstract: Abstract—The seamless control of dexterous prosthetic hands requires human-machine interfaces (HMIs) that are not only highly accurate but also robust against online decoding insta bilities. In this paper, we propose SU-CT , a unified framework that leverages the temporal responsiveness of sEMG and the morphological stability of A-mode ultrasound . In our architec ture, a shallow 1D-CNN tokenizer extracts local spatiotemporal features, which are then fused and modeled by a Multimodal Transformer to capture global sequence context. Evaluated on 8 subjects across 9 hand gestures, SU-CT achieves a state-of-the art offline accuracy of 98.85%. In an online Target Matching Task(TMT), the system demonstrates a high Task Success Rate of 96.5% with a motion completion time of low delay. Our work provide a stable forward-mapping solution for human-in-the loop prosthetic control.

Keywords: Robotics, Motion Control, Modeling and Control of Complex Systems
Abstract: Mobile tensegrity robots have attracted significant attention due to their lightweight, resilient structures and potential for deployment in challenging environments. However, the majority of existing designs suffer from drawbacks such as inefficient zigzag locomotion mode, excessive motor requirements, and complex actuation schemes. Recently, a novel mobile tensegrity robot featuring a regular prismatic envelope and a simple dual-pendulum actuation scheme was proposed, which achieved smooth and efficient straight-line rolling with minimal actuation by eliminating the zigzag motion. However, its turning capabilities require further optimization, particularly in terms of actuation strategies. In this paper, we focus on the improvement of the turning process of this robot by optimizing its actuation trajectories of the two pendulums. A simplified dynamic model describing the turning dynamics is first developed. Based on the model, the actuation trajectory is parameterized using Fourier series and optimized within a finite-dimensional parameter space. An optimal trajectory is obtained, which is compared with a constant-velocity baseline trajectory in MuJoCo simulation. The results show that the optimized trajectory improves the turning performance by 102.8% in the simplified model and by 190.7% in simulation, compared with the baseline, which demonstrate the feasibility of the proposed model-based actuation trajectory optimization framework.

Keywords: Robotics, Motion Control, Control Applications
Abstract: Enhancing underwater locomotion can improve divers' operational efficiency while lowering the physiological risks associated with prolonged physical exertion. Although wearable robotic systems have been widely explored for terrestrial applications, their deployment in underwater environments remains largely unexplored. This study presents a rigid underwater hip exoskeleton designed to assist flutter kick during scuba diving. The mechanical structure enables bidirectional assistance throughout the kicking cycle while maintaining natural leg mobility. To achieve synchronization between the human motion and robotic assistance, a gait-based adaptive oscillator (GBAO) is used to estimate the kicking phase in real time, enabling phase-based torque assistance. A proof-of-concept experiment was conducted with one participant performing flutter kick in a swimming pool. The results show that the proposed exoskeleton can operate reliably underwater and deliver assistance synchronized with the user’s kicking motion, providing experimental evidence for the practical feasibility of rigid underwater exoskeleton assistance in a real underwater setting.

Keywords: Robotics, Modeling and Control of Complex Systems, Man-machine Interactions
Abstract: This paper presents a multi-mode coordinated control method for a bilateral upper-limb exoskeleton rehabilitation robot. A kinematic model is established using an improved D–H convention, and gravity compensation is derived based on the principle of virtual work. A hierarchical control framework is developed with three operating modes. The transparent mode incorporates feedforward compensation to enable near zero-force teaching. The passive training mode uses PID control for trajectory tracking during early rehabilitation. The active training mode applies admittance control, allowing the system to respond to interaction forces between the user and the robot. In addition, a mirror bilateral strategy is introduced, in which the motion of the healthy limb is mapped to the affected side. An active participation mechanism is included, with fault-tolerant trajectory adjustment based on force feedback from the affected limb. Experimental results on shoulder joint movements demonstrate high tracking accuracy in passive mode, effective force-based interaction control in active mode with adjustable admittance gain, and reliable trajectory generation in mirror training. These results confirm the system’s ability to adapt to different patient conditions and rehabilitation stages.

Keywords: Robotics, Signal Processing
Abstract: This study conducted comparative experiments under two conditions: with and without exoskeleton wear. Two healthy subjects performed standardized movements across four right-arm holding tasks while simultaneously recording surface electromyography (sEMG) signals from six muscles. The exoskeleton's assistive effect was evaluated based on muscle activation levels, Superlet time-frequency characteristics, and co-contraction of BB-TB muscle pair. The results indicated that overall activation levels in major upper limb muscle groups decreased after exoskeleton wear. Besides, This study demonstrates that passive upper-limb exoskeletons have the potential to reduce the load on upper muscle groups in specific type of task. The findings of this study could provide guidance for the structural optimization and performance improvement of passive upper-limb exoskeletons. Future work will be conducted with more subjects and long-term wear experiments.

Keywords: Robotics, Man-machine Interactions, Motion Control
Abstract: The emergence of micro-motors has made it possible to design direct-drive dexterous hands with flexible in-hand manipulation capabilities and without complex mechanical transmission errors. However, existing design methods in this context still rely on heuristic trial-and-error, lacking systematic methodologies. We propose a two-stage optimization framework that formalizes the design of direct-drive dexterous hands as an automated, gradient-free optimization problem. Using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES), the framework autonomously resolves highly-coupled design parameters under the physical constraints of selected micro-motors. The first stage identifies an optimal hand configuration by maximizing the opposability envelope volume. After that, the second stage refines phalanx proportions by reproducing natural human hand motions within a physics simulator. By optimizing a composite loss function that strictly penalizes physical unfeasibility, the optimized design reduces self-collision depth by 50.49% and joint limit violations by 32.01% compared with the empirical baseline. This methodology connects theoretical kinematics with physical hardware constraints, rapidly generating optimal dexterous hand design from micro-motors.

Keywords: Robotics
Abstract: This paper presents a sensorless method for estimating object stiffness using only the built-in feedback of smart servo actuators. A custom parallel-jaw gripper driven by two Dynamixel XM540-W150-R motors is mounted on a Staubli TX40 manipulator and performs controlled current- ramp squeezing while motor current and encoder position are recorded. Applied torque is derived from current readings through actuator conversion constants and experimentally identified friction compensation, while object compression is computed from a calibrated encoder-to-distance kinematic model. A compliance correction is introduced to reduce bias caused by structural deflection of the 3D-printed PLA fingers under load. The method is validated on six household objects spanning soft-to-rigid behavior, each tested in five repeated trials. Estimated linear stiffness ranges from 0.079 Nm/mm for a kitchen sponge to approximately 2.97 Nm/mm for an apple, with rigid objects showing near-zero compression at the measurement floor. Results confirm that the proposed torque- compression profiling approach can discriminate objects of different stiffness categories without external force/torque or tactile sensors, supporting low-cost integration into robotic manipulation workflows.

Keywords: Sensor/Data Fusion, Robotics
Abstract: In intensive cattle finishing systems, residual feed assessment is still predominantly performed by human bunk callers who visually estimate leftover volume in feed bunks. This subjective process leads to inconsistent feeding decisions, feed waste, and potential health risks for livestock. This paper presents a controlled, side-by-side evaluation of two perception modalities for automated volumetric feed estimation: (i) a static overhead Intel RealSense D455 RGB-D depth camera and (ii) a traversal-based dual Hokuyo 2D LiDAR system that reconstructs 3D geometry by stitching sequential scans. Both systems were tested on the same physical feed bunk using identical crushed barley samples ranging from 5 to 14 L. A shared grid-based height integration pipeline was used for volume computation. The depth camera achieved a mean absolute error (MAE) of 0.57 L with near-zero systematic bias, while the dual-LiDAR system produced an MAE of 1.78 L in raw mode and 1.37 L after applying an empirical calibration factor (k = 1.1631). The results indicate that single- frame RGB-D acquisition provides higher accuracy and better consistency under controlled bunk conditions, primarily due to reduced sensitivity to motion artifacts and higher spatial point density. The findings offer practical guidance for sensor selection in perception-estimation loops for precision livestock feeding automation.

Keywords: Robotics, Signal Processing, Sensor/Data Fusion
Abstract: This paper describes an antiphase dual-source illumination approach that creates a hardware-defined Region of Interest (ROI) through temporal superposition: two laterally modulated light fields overlap to form a central zone with substantially reduced flicker. Electrical verification confirmed near-ideal antiphase modulation, and high-speed camera analysis showed that intensity peaks in one lateral zone align with troughs in the other, producing a markedly flatter temporal profile in the overlap region. Photometric measurements confirmed that stabilization does not come at the cost of dimming: ROI reached nearly double the mean illuminance of the lateral areas of the ROI. Together, these results validate the additive superposition model and show that antiphase lighting can produce a brighter, more temporally stable illumination region in a controlled static setup. These findings establish the photometric basis of the proposed method and motivate future applications in various robotic systems.

Keywords: Intelligent and AI Based Control, Learning Systems, Learning-based Control
Abstract: Adaptive optimizers in the AdamW family usually use a fixed default preconditioning strength. However, this setting may not be suitable for facial expression recognition and transfer tasks with different dataset structures. In this paper, we extend p-norm preconditioning experiments to facial expression recognition and transfer to a child pain expression dataset, and study optimizer bias from the perspective of dataset structure. We use p as a unified control variable to adjust the balance between macro low-frequency expression patterns and local detail features. Experiments with both basic and newer backbones on conventional facial expression datasets and a self-collected child pain dataset show that the base setting of AdamW may still overemphasize detail information in facial expression recognition. Increasing p appropriately can improve the focus on macro expression patterns and achieve better performance. For example, on RAF-DB, p=0.6 improves performance by 1.91% over the base setting. We further find that the child pain expression task, with only three classes, has a relatively more dispersed feature space and therefore prefers a different optimizer bias from the seven-class RAF-DB task. These results suggest that AdamW is not optimal for all facial expression datasets and transfer tasks, and that the preferred optimizer bias is closely related to dataset structure.

Keywords: Intelligent and AI Based Control, Man-machine Interactions, Learning-based Control
Abstract: In complex human-machine collaboration systems, the psychophysiological states of operators directly affect system performance and safety, yet traditional control architectures cannot effectively quantify these impacts. This paper proposes a human factors risk assessment and intervention decision support framework integrating Bayesian Neural Networks (BNN) and Bayesian Networks (BN). To validate the framework under highstakes, high-uncertainty conditions, we adopt psychological crisis as a proxy for extreme human-factor states, for which robust physiological markers are well documented. For early warning of psychological crises, the BNN extracts risk probabilities and quantifies predictive uncertainty from multimodal physiological signals (e.g., heart rate variability, EEG alpha power). The BNN outputs serve as evidence variables in a BN that explicitly models interactions among biological, psychological, and social factors, facilitating the transition from black-box prediction to interpretable causal reasoning. Validated on synthetic data driven by the literature, the BNN achieves an AUC of 0.9625 with wellcalibrated uncertainty. Meanwhile, the BN identifies alpha power as the key biological driver and social support as the strongest protective factor through sensitivity analysis and counterfactual simulation. This framework provides an interpretable tool for the accurate and timely identification of human factors and supports personalized interventions, offering a new paradigm for integrated human-adaptive control systems.

Keywords: Man-machine Interactions
Abstract: Lower-limb exoskeletons are representative wearable human–robot systems whose control performance critically depends on compliant interaction under strong physical coupling with the user. While a wide range of compliant control strategies has been proposed, existing review studies predominantly classify these methods based on algorithmic structures, which limits a systematic understanding of their underlying human–exoskeleton coupling regulation mechanisms. This paper introduces the concept of Coupling Regulation Depth (CRD) from a system-level perspective to characterize the hierarchical extent to which human-related information is incorporated into control regulation and decision-making. Based on this perspective, compliant control strategies for lower-limb exoskeletons are conceptually categorized into three levels: low CRD, dominated by kinematic reference tracking; medium CRD, characterized by explicit regulation of interaction dynamics; and high CRD, featuring cooperation based on human states and movement intentions. Representative methods at each level are comparatively analyzed in terms of regulation targets, human roles, and applicable scenarios. The proposed CRD framework provides a unified perspective for understanding and analyzing compliant control strategies and offers insights for the design of human–exoskeleton cooperative control architectures.

Keywords: Man-machine Interactions, Motion Control, Nonlinear Systems and Control
Abstract: With the rapid development of intelligent human–machine interaction systems such as wearable assistive devices, rehabilitation robots, and virtual training platforms, significant progress has been achieved in interaction control, motion assistance, and adaptive feedback design. Most existing human–machine interaction studies primarily focus on system-level control strategies and external biomechanical signals, including motion kinematics, joint torque, and interaction force regulation. However, the human-side physiological limitations, particularly fatigue-induced internal biomechanical alterations, remain insufficiently understood. Muscle fatigue can significantly influence human motor capability and interaction stability, potentially leading to degraded system performance and increased injury risk. Therefore, investigating fatigue-induced biomechanical responses within the human body is essential for improving the safety and reliability of human–machine collaborative systems.This study is based on the framework of muscle bone digital twin modeling to explore the effect of muscle fatigue on the mechanical response and potential injury risk of tibial cartilage during squat exercise. Recruit 20 healthy male participants to repeat deep squats under different loads until exhaustion, and use motion capture systems, 3D force platforms, and surface electromyography systems to synchronously collect lower limb kinematic and dynamic data. Establish an individualized human-machine coupled musculoskeletal model based on OpenSim, calculate the contact force of the tibiofemoral joint, and use the contact force as the external load condition of the finite element model to analyze the maximum principal stress and equivalent strain of the tibial cartilage in ANSYS. The results indicate that as muscle fatigue increases, the axial compressive force of the tibiofemoral joint gradually decreases and the peak time is delayed. Despite the overall decrease in axial load, there is a signif

Keywords: Robotics, Learning-based Control, Real-time Systems
Abstract: Abstract—Precise real-time obstacle avoidance remains a critical bottleneck for robotic manipulation in highly constrained environments. While Diffusion Policy offers a robust framework for complex behavior generation, standard Transformer backbones often suffer from representation attenuation in deep layers and stochastic instability during the denoising process. We address these limitations by proposing an enhanced Diffusion Policy framework featuring a specialized Transformer architecture. Specifically, we integrate custom Diffusion Transformer (DiT) blocks with Adaptive Layer Normalization (AdaLN-Zero) to stabilize conditional denoising, and introduce symmetrical long skip connections inspired by U-ViT to preserve fine-grained spatial geometric features. Experimental results on the PushT benchmark demonstrate that our method achieves a 94.5% success rate, significantly outperforming the baseline in both precision and training convergence. Furthermore, our model achieves a 15.4 ms inference latency on an NVIDIA RTX 4090 GPU, satisfying the 50 Hz requirement for high-frequency closedloop control. These architectural refinements provide a scalable solution for deploying agile robotic agents in cluttered, real-world scenarios.

Keywords: Optimal Control
Abstract: Practical robotic control tasks are frequently formulated as time-varying quadratic programming (TVQP) problems. Recurrent Neural Network (RNN) exhibit superior efficacy in addressing TVQP, attributed to their inherent parallel processing and dynamic tracking capabilities. However, conventional techniques encounter significant challenges when handling multilevel inequality constraints, such as concurrent joint position and velocity limits, as their reliance on non-smooth piecewise functions inevitably induces control chattering. This study proposes a unified and differentiable multilevel constraint handling framework. By ensuring global differentiability, the proposed method satisfies the rigorous demand for continuous derivative information in gradient-based neurodynamics, while maintaining compatibility with traditional numerical solvers. Simulation experiments on a Franka robot arm executing complex trajectories demonstrate that the proposed approach achieves high-precision tracking while ensuring strict adherence to multilevel constraints.

Keywords: Man-machine Interactions, Robotics, Adaptive Control
Abstract: ，平衡操作透明度与动态稳定性 这仍然是机器人辅助手术中的一个关键挑战。 本文提出了一种灵活的自适应导纳控制方法 基于交互力反馈的策略以满足 多尺度手术任务需求。作者 利用相互作用力作为 外科医生意图，是一种基于 建立S形功能以实现连续且 阻尼系数的稳定跃迁。整合 自适应巴特沃斯滤波器和非线性饱和 约束条件进一步确保部队信号的准确性和系统 安全感。骨科机器人的实验结果 证明该策略能够准确调整动态 根据实时能量特性进行响应。 与传统的固定高阻尼模式相比， 所提方法可抑制震颤现象 高效操作，并能显著降低 大工作区导通时的工作电阻 动ƃ

Keywords: Intelligent and AI Based Control, Learning-based Control, Man-machine Interactions
Abstract: ，机器人辅助全膝关节置换术（TKA）截骨术要求高度合规且安全的人机物理互动（pHRI）。重型手术器械和高频非固定振动阻碍了传统时间序列模型从6D力信号中稳健提取外科医生真实的手术意图。本文提出了基于TimesNet的骨科机器人意图识别与控制框架。实时数据预处理首先进行重力/偏心力矩补偿和自适应巴特沃斯滤波。随后，TimesNet将一维时间力信号转换为二维张量，有效将局部高频振动与跨多周期尺度的长期意图趋势解耦。识别的意图会输入平滑各向异性导管，实现合规跟踪。线下评估显示，TimesNet的表现优于LSTM和Transformer等基线模型。在模拟截骨实验中，该系统展现出强大的端到

Keywords: Robotics, Man-machine Interactions, Motion Control
Abstract: Spinal endoscopic surgery is a representative minimally invasive technique for the treatment of spinal disorders, but the narrow operative corridor and constrained workspace impose high requirements on the dexterity and accessibility of surgical instruments, especially for electrocautery hemostasis. To address the limitations of conventional electrocautery instruments in complex surgical regions, this paper proposes a variable-curvature continuum electrocautery surgical manipulator for spinal endoscopic surgery. The proposed manipulator adopts a modular design integrating an electrocautery tool, a flexible distal segment, and a proximal actuation mechanism, and achieves distal bending, curvature adjustment, axial translation, and overall rotation through tendon-driven actuation. A constant-curvature kinematic model is established to describe the tip pose, and a workspace analysis is conducted to evaluate the reachable region of the manipulator. In addition, an intuitive master-slave mapping method based on a master device is developed to improve teleoperation performance. A prototype of the proposed manipulator is fabricated, and in vivo animal experiments are conducted to validate its operational feasibility and hemostatic effectiveness. The experimental results demonstrate that the proposed manipulator can achieve precise targeting and safe electrocautery hemostasis in complex tissue environments, indicating its potential application in minimally invasive spinal endoscopic surgery.

Keywords: Networked Control, Learning-based Control, Optimal Control
Abstract: This paper investigates policy gradient methods for the linear quadratic regulator (LQR) problem with Bernoulli packet loss. Since packet losses alter the system dynamics and the expected cost function, existing results on policy gradients for deterministic LQR cannot be directly extended. Therefore, we first prove that, although the LQR with packet losses is a non-convex optimization problem, policy gradient methods can converge to the globally optimal policy, as gradient dominance still holds. We then characterize the ``almost” smoothness property of the packet-loss LQR cost function, which provides a foundation for step size selection and convergence-rate guarantees. Furthermore, we heuristically tailor a Gauss-Newton method to the packet-loss LQR and rigorously prove its exponentially fast convergence to the global optimum. By appropriately scaling the step size according to the loss rate, the convergence rate of the Gauss-Newton method under losses shares the same expression as in the loss-free case. Finally, numerical simulations validate the theoretical results.

Keywords: Multi-agent Systems, Optimal Control
Abstract: This paper studies a multiplayer reach-avoid differential game where a noncooperative target moves along a line segment. The pursuers cooperate to escort the target moving from the start point to the goal point of the line segment against the evaders, whose objective is to attack the target before the target reaches the goal point and avoid being captured by the pursuers. Due to the complexity of direct analysis, the entire game is decomposed to multiple subgames, each involving one pursuer and one evader. First, the properties of the players’ trajectories and control inputs are analyzed under optimal play for various payoff functions. Then, an evasion region is introduced for the pursuer-evader pair, from which a safe distance is defined to evaluate the safety of the target. Three pursuit winning conditions and strategies are presented to guarantee the safety arrival of the target, regardless of the evasion strategy. Finally, a maximum matching is obtained by combining all subgame outcomes, which guarantees a lower bound on the number of defeated evaders. Numerical simulations are presented to illustrate the theoretical results.

Keywords: Automated Guided Vehicles, Nonlinear Systems and Control, Robust and H infinity Control
Abstract: During capture tasks, the quadrotor inevitably suffers from impact disturbances and payload uncertainties. These composite disturbances pose a significant challenge to controllers. This paper proposes a dual-loop disturbance observer-based nonlinear model predictive control framework. The position loop estimates body-axis disturbance forces to decouple the payload disturbances from the attitude, while the inner loop employs a predefined-time sliding mode disturbance observer to ensure reliability. The Nonlinear Model Predictive Control (NMPC) integrates disturbance estimation to generate anti-disturbance actions. High-fidelity simulations validate the robustness of the proposed controller against both impact and payload disturbances.

Keywords: Multi-agent Systems, Intelligent and AI Based Control, Learning Systems
Abstract: Heterogeneous multi-agent adversarial tasks require balancing global collaboration with local constraints under partial observability. Centralized LLM-based frameworks suffer from tight coupling between global awareness and local execution, causing context overload, action failures and short-sighted policies. We propose HGLAD, a hierarchical agentic framework decoupling global situation assessment from local execution. It comprises: a low-frequency Strategic Cognition Layer for global assessment and long-term planning via zero-shot chain-of-thought (CoT); and a high-frequency Tactical Execution Layer that generates executable actions by integrating strategic guidance with local constraints using SayCan and ReAct. Experiments show HGLAD outperforms strong centralized baselines, ensuring policy robustness and action feasibility while improving adversarial performance.

Keywords: Fuzzy and Neural Systems, Intelligent and AI Based Control, Multi-agent Systems
Abstract: 近年来，大型语言模型（LLMs）越来越多地应用于时序知识图谱问答（TKGQA）。然而，在处理复杂的时间推理时，LLM常常表现不佳，并生成不准确或误导性的回答。造成这一问题的主要原因是现有方法通常会让大型语言模型暴露在从知识图谱中检索的庞大、未修剪的候选动作空间。如果不清楚细粒度实体语义角色（例如区分主体与对象或时间锚点），LLMs极易受到工具调用幻觉和推理循环的影响。为应对这一挑战，我们提出了一种新的角色感知推理框架，适用于TKGQA。我们的方法引入了实体角色提取模块，以动态识别自然语言查询中实体的语义角色。通过利用这些结构化角色

Keywords: Multi-agent Systems, Networked Control, Linear Systems
Abstract: This paper studies the Nash equilibrium (NE) seeking problem for multicoalition games with varying number of heterogeneous general linear players. In such games, the players cooperate within their own coalitions while competing against other coalitions. Each player can only access its own cost function but is required to optimize the cost function of the entire coalition. To address this issue, an average gradient estimate is designed over weight-balanced digraphs utilizing strategy information and the estimation results are then integrated into the control input to guide players to the NE. In response to players joining or leaving due to task completion and new task initiation, the prescribed-time protocol is introduced to drive players to the NE solution before such changes occur. Then the convergence properties of the proposed algorithm are analyzed based on the Lyapunov method. Finally, a numerical simulation of the multicoalition mobile sensor connectivity game is conducted to demonstrate the effectiveness of the algorithm.

Keywords: Signal Processing, Motion Control, Fuzzy and Neural Systems
Abstract: Specificity analysis of motor function after stroke is of great significance for guiding rehabilitation and developing intelligent rehabilitation systems. This study focused on an integrated EEGCerebral Oxygenation crossmodal analysis for poststroke motor function assessment towards clinical neurorehabilitation applications. We recruited eight healthy volunteers and four post-stroke hemiplegic patients, and designed a standardized elbow flexion-extension paradigm. For analysis, the PLI and Pearson correlation coefficient were used for functional connectivity analysis between EEG and cerebral oxygen (fNIRS), while transfer entropy (TE) was adopted for cross-modal coupling strength quantification. The results showed that the hemiplegic patients had specific EEG features, with elevated β/γ band connectivity during affected limb movement, δ band hyper-connectivity, and stronger connectivity in unaffected brain regions. For fNIRS signals, the connectivity in resting exceeded that in motor state, the prefrontal cortex had the strongest connectivity, and the unaffected brain regions also showed higher connectivity. There is a bidirectional information transmission between EEG and cerebral oxygen signals, with significantly higher TEΔHbO→EEG in motor state, on patients, and during affected limb movement, respectively. This study preliminarily revealed the neurophysiological mechanisms of post-stroke motor dysfunction and compensation, and verified the EEG-fNIRS coupling as an objective biomarker for function assessment, which may provide a support for developing intelligent rehabilitation robots and precision rehabilitation.

Keywords: Intelligent and AI Based Control, Fault Detection and Diagnostics
Abstract: Long-tailed data distribution and limited annotated samples present major challenges for wind-turbine surface defect detection, particularly for rare contamination-related defects. To address this issue, this paper proposes a diffusion-based data augmentation method tailored for long-tailed wind-turbine defect detection. The method generates additional minority-class defect samples while preserving the surface context and visual realism of the original images. To validate its effectiveness, the augmented dataset is evaluated using multiple mainstream object detectors under different training settings. Experimental results show that the proposed augmentation strategy consistently improves minority-class detection performance, increasing mAP@0.5 from 0.346–0.377 to 0.694–0.775. These results demonstrate that diffusion-based augmentation is an effective way to alleviate class imbalance and enhance long-tailed defect detection in wind-turbine inspection.

Keywords: Intelligent and AI Based Control, Estimation and Identification
Abstract: Accurate long-range 3D reconstruction in underwater environments remains challenging due to severe light absorption, scattering, and the high acquisition cost of dense delay scanning in conventional range-gated imaging systems. To improve reconstruction efficiency while reducing hardware and data collection costs, this paper proposes a deep learning framework that integrates coded gating strategy design with data-driven depth estimation. Instead of dense temporal scanning, three carefully designed coded range-gated images are employed to encode complementary depth information, significantly reducing acquisition complexity. A multi-encoder--decoder architecture with DINOv2 backbones is adopted to extract multi-scale features from the coded observations, followed by cross-level feature fusion and a DPT-based prediction head for dense depth reconstruction in a single forward pass. To address the scarcity and high cost of real underwater paired datasets, a depth-prior-guided low-cost data synthesis pipeline is developed to construct a large-scale training dataset without expensive field measurements. This strategy enables effective supervision while maintaining physical consistency with underwater imaging characteristics. Extensive experiments on both simulated and real-world data demonstrate that the proposed method achieves superior performance compared to traditional analytic reconstruction methods and baseline learning models in terms of Absolute Relative Error (AbsRel) and threshold accuracy δ<1.25. The proposed approach provides an efficient, scalable, and cost-effective solution for underwater 3D perception, with strong potential for applications such as seabed mapping and underwater inspection.

Keywords: Sensor/Data Fusion, Signal Processing
Abstract: Three-dimensional imaging technology has significant application value in endoscopic medicine. Accurate 3D depth information can assist both surgical procedures and routine diagnosis. This paper proposes a narrow-baseline binocular endoscopic system and a corresponding 3D reconstruction method. After completing binocular calibration and epipolar rectification, Foundation Stereo is introduced to achieve dense disparity estimation. Combined with chessboard calibration information for scale correction, 3D reconstruction results with real-world metric scale are obtained. The proposed method is validated on both real endoscopic images and the Middlebury 2014 benchmark dataset, and is compared with the traditional Semi-Global Block Matching (SGBM) method. Experimental results show that in real endoscopic scenarios, the proposed approach can generate structurally continuous 3D reconstructions with reasonable geometric shapes under complex imaging conditions. The results provide a feasible technical route and experimental basis for narrow-baseline binocular endoscopic 3D imaging.

Keywords: Control Applications
Abstract: Aiming at the aerial security threats posed by highly maneuverable unmanned aerial vehicles (UAVs), existing electro-optical tracking systems often struggle to achieve both rapid response and reliable performance. To address this issue, this paper proposes a multi-stage fast tracking method based on a fusion strategy for highly maneuverable aerial targets. First, the characteristics of the kernelized correlation filter (KCF) tracking algorithm are analyzed, along with the core problems encountered in conventional tracking centering processes, including target escape, motion-induced image blur, and tracking loss caused by limited servo rotation speed. Subsequently, a multi-stage tracking framework is constructed, comprising four sequential stages: initial target acquisition, deviation conversion and template pre-storage, fast servo rotation centering, and stable tracking guided by the pre-stored target feature template. Finally, experimental evaluations and analyses are conducted to verify the effectiveness of the proposed method. The results demonstrate that the multi-stage method improves tracking centering efficiency by a factor of 3.44, effectively enhancing both tracking speed and stability. Moreover, it substantially improves the rapidity, accuracy, and robustness of electro-optical systems in tracking highly maneuverable aerial targets. This work provides a novel solution for advancing electro-optical tracking technologies against highly maneuverable aerial targets.

Keywords: Control Applications
Abstract: Swarm operations of unmanned aerial vehicles (UAVs) have emerged as a disruptive technology reshaping future operational paradigms, where reconnaissance perception and autonomous target locking serve as core capabilities. To tackle the engineering challenges including scarce effective features of dim and small targets, easy loss caused by target maneuvering, significant system link latency, and the difficulty in balancing locking accuracy and real-time performance, this paper proposes an optimized autonomous locking strategy for dim and small targets in small UAV platforms. By optimizing the system management architecture, an efficient closed-loop kill chain of detection-localization-tracking-aiming-engagement is constructed. Based on the YOLOv8 detector, weak target features are preserved via image block-based inference. A frame-number encoding and cache traceability mechanism is designed to achieve precise information-image synchronization and reduce link latency. An intelligent autonomous switching logic between detection and tracking is established to support target reacquisition after loss and human-in-the-loop intervention. Field experiments demonstrate that the proposed method provides a reliable technical solution for autonomous reconnaissance and engagement of small UAVs.

Keywords: Sensor/Data Fusion, Man-machine Interactions, Real-time Systems
Abstract: Acquiring complete and high-fidelity 3D facial geometry is essential for applications such as digital avatar generation and biometric analysis. However, existing methods struggle to balance high-resolution data acquisition with ef- ficient processing, often suffering from heavy computational costs and cumulative drift during rapid head movements. To address these challenges, we present a complete high-speed 3D facial reconstruction system that tightly integrates a custom monocular structured light scanner with an efficient, globally consistent registration pipeline. First, our hardware system acquires 1.77-megapixel facial geometry at 35Hz. To handle this massive data stream, we implement a spatiotemporal keyframe selection strategy based on spherical grid sampling to effectively filter redundant frames. For global registration, we introduce an angular-guided pipeline. By utilizing a spherical projection mechanism, we bypass exhaustive pairwise feature matching, thereby significantly reducing the graph construction complexity. Furthermore, by replacing the heavy non-linear Pose Graph Optimization (PGO) backend with cascaded matrix multiplications along a Minimum Spanning Tree (MST) struc- ture, we completely bypass the time-consuming optimization phase. Experimental results demonstrate that our entire system achieves a 14.8× overall speedup compared to standard PGO baselines, while maintaining robust sub-millimeter geometric precision (Inlier RMSE of 0.339 mm) under large pose varia- tions.

Keywords: Robotics, Sensor/Data Fusion, Signal Processing
Abstract: Search and rescue (SAR) in dense forest poses significant challenges due to heavy canopy occlusion and poor visibility. Traditional methods like thermal imaging and LiDAR often fail to penetrate dense vegetation effectively. This paper proposes a novel system that combines Light Field Rendering (LFR)—a synthetic aperture imaging technique using drones to capture unstructured light fields for occlusion removal—with a super-lightweight You-Only-Look-Once (YOLO) model generated via YOLO-Light, a NeuroEvolution-based architecture optimization method. By integrating multi-perspective thermal images into occlusion-free integral images and applying the lightweight YOLO for real-time detection, our approach achieves high precision and recall in victim localization under heavy occlusion. Experimental results on datasets from forested terrains demonstrate a reduction in model parameters by more than 10 times while maintaining detection accuracy above 90%, enabling deployment on resource-constrained drones.

Keywords: Nonlinear Systems and Control, Learning-based Control, Linear Systems
Abstract: This paper investigates the safety-critical control problem for continuous-time nonlinear polynomial systems subject to perturbations, where the system matrices A and B are entirely unknown. To ensure safety without prior knowledge of the model, we propose a robust, data-driven framework based on control barrier certificates (CBCs). Unlike existing deterministic approaches, the proposed method explicitly incorporates unmodeled dynamics and perturbations into the safety constraints by leveraging sum-of-squares (SOS) programming. Using the collected input and state data, a quadratic CBC and a corresponding safe controller are synthesized simultaneously. The framework guarantees that system trajectories remain within the safe set even in the presence of bounded perturbations. Simulation results on a nonlinear polynomial system demonstrate the robustness and effectiveness of the proposed data-driven synthesis approach.

Keywords: Nonlinear Systems and Control, Robotics, Modeling and Control of Complex Systems
Abstract: This paper focuses on ensuring the safety of tethered human--robot guidance systems in obstacle-cluttered environments. In such systems, the robot and the human are physically coupled by a tether, which imposes unique geometric constraints and requires coordinated safety guarantees for both participants. To address this, we first establish a unified human--robot coupled kinematic model. Then, a safety-critical control framework is proposed by integrating Signal Temporal Logic (STL) with Control Barrier Functions (CBF). Specifically, the high-level task goals are described by STL to generate nominal motions, while the low-level CBF-based optimization ensures real-time obstacle avoidance and maintains the required tethered guiding mode. The performance of the proposed method is validated through Gazebo simulations and hardware experiments with a quadruped robot. Results show that our approach effectively guides the human while strictly satisfying all safety and tether-related constraints.

Keywords: Multi-agent Systems, Networked Control
Abstract: This paper investigates the distributed Nash equi- librium seeking problem of N players with discrete-time dy- namics under jointly strongly connected switching networks. First, by integrating gradient play technique with a novel discrete-time distributed observer design, we develop a discrete- time distributed Nash equilibrium seeking controller to solve the problem. Next, we establish two technical lemmas to analyze the uniform exponential stability of discrete-time linear switched systems. Then, we obtain the exponential stability of the interconnected closed-loop systems by constructing a Lyapunov function and providing a controller gain design method with the small gain arguments. Finally, the discrete-time distributed Nash equilibrium seeking is achieved over jointly strongly connected switching communication networks. The effectiveness of the developed approach is demonstrated by a numerical simulation example.

Keywords: Nonlinear Systems and Control, Robotics, Control Applications
Abstract: Teleoperation systems are essential in high-risk domains such as minimally invasive surgery, nuclear mainte- nance, and space exploration. However, effective haptic feed- back usually requires expensive force sensors and specialized hardware, resulting in high cost and system complexity. To address this issue, this paper presents a cost-effective sensorless teleoperation system based on a unified impedance control framework. The system adopts a master–slave architecture, where a custom high-stiffness Delta parallel manipulator serves as the master device and a 6-DOF serial manipulator serves as the slave device. The forward channel maps master-side motion to slave-side pose commands, while the feedback channel syn- thesizes virtual spring-damper interaction forces from master– slave pose discrepancies without distal force/torque sensors. Experiments demonstrate stable tracking performance and contact-related haptic cues during constrained interaction. The proposed system reduces hardware overhead and integration complexity, providing a practical solution for educational and general industrial teleoperation applications.

Keywords: Intelligent and AI Based Control, Learning-based Control, Robotics
Abstract: Reach-avoid (RA) games serve as fundamental mathematical models for safety-critical autonomous systems. However, in complex environments characterized by dense obstacles, balancing efficient adversarial interception with safe obstacle avoidance presents significant challenges. Traditional analytical methods suffer from the curse of dimensionality, while standard deep reinforcement learning (DRL) approaches often converge to suboptimal policies due to the conflicting dual objectives of safety and mission efficiency. To address these limitations, we propose an hierarchical reinforcement learning (HRL) framework which decouples strategic planning from motion control. The high-level manager incorporates a gated recurrent unit predictor to reconstruct the adversary's belief state under partial observability. By leveraging dynamic temporal abstraction, it generates macroscopic tactical commands, intelligently switching between target protection and active pursuit to maximize long-term game payoffs. Simultaneously, the low-level worker employs a neural module network (NMN) architecture to disentangle task-oriented flows from obstacle-avoidance flows, achieving efficient maneuvering strictly premised on safety compliance. Extensive simulations demonstrate the proposed method's superior robustness against diverse and complex attack strategies, including evasive, orbiting, and composite behaviors. Compared to baseline algorithms, our framework significantly outperforms in terms of capture rate and decision efficiency, validating the superiority of the hierarchical architecture in RA games with non-convex geometric constraints.

Keywords: Control Applications, Multi-agent Systems, Nonlinear Systems and Control
Abstract: This paper investigates the resilient formation control problem for vehicle platoons subject to the synergistic coupling of parameter uncertainties and composite cyberattacks. Unlike prevalent approaches predicated on model linearization, which may prove inadequate against structural nonlinearities and complex threat environments, this work establishes a unified decentralized framework tailored for rigorous nonlinear longitudinal dynamics. Specifically, the system is modeled to account for the specific characteristics of both additive and multiplicative attacks. A novel kinematic evolution based on a bidirectional nearest-neighbor topology is first developed, theoretically revealing a fundamental structural decoupling property where the macroscopic centroid motion of the leaderless platoon remains invariant to microscopic formation errors. Then, a resilient controller is designed according to the kinematic evolution. It incorporates a quadratic compensation term to actively mitigate the worst-case bounds of aggregated uncertainties and attacks without requiring precise estimation. Rigorous Lyapunov stability analysis demonstrates that the controlled closed-loop system is uniformly ultimately bounded. Finally, the efficacy and robustness of the proposed strategy are further validated through comprehensive numerical simulations.

Keywords: Man-machine Interactions
Abstract: 自闭症谱系障碍（ASD）的临床诊断 目前高度依赖标准化评分标准， 系统性临床观察。然而，这些 传统方法受主观回忆的限制 偏见和对专业医疗资源的依赖， 强调客观早期评估的紧迫需求 工具。为应对这些挑战，本研究提出了 新颖的桌面触摸互动系统。系统 设计色彩和形状的多阶段测试范式 认知。它通过系统提取内部交互数据 事件记录，以及获取外部行为数据 利用摄像机视频录制和算法 处理。一项涉及20名受试者的受控实验 （10 ASD，10 典型发育）证明了 ASD儿童表现出显著的异质性 与典型发展（TD）儿童在任务中有所不同 完成度、准确性、反应时间和注意力 指标。这些定%

Keywords: Estimation and Identification, Sensor/Data Fusion, Smart Structures
Abstract: The shape perception and parametric description are prerequisites for the optimization and control of flexible camber-morphing wing surfaces, while traditional Class function/Shape function Transformation (CST) methods are difficult to meet the demand for accurate description of large-scale trailing-edge camber variation. To address this issue, this paper conducts a study on shape perception and high-precision parametric description for wings with fixed leading edge and large trailing-edge camber deformation. A three‑segment variable‑camber morphing wing configuration is proposed, which adopts concave hexagonal cells combined with sliding skin. And, a segmented CST difference fitting method with C¹ continuity constraint is proposed to achieve low-dimensional and high-precision description of complex wing surfaces. Furthermore, a distributed sensing system composed of strain sensor arrays, laser sensor arrays, IMU, and linear displacement sensor is established to achieve shape perception under both static and dynamic wing conditions. Experimental results show that compared with the conventional CST method, the accuracy of the proposed parametric description method has been improved by one order of magnitude, and its representation of large trailing-edge deformation variable-camber wings satisfies the typical wind-tunnel tolerance. The shape sensing system effectively validates the feasibility of real-time application of the proposed method.

Keywords: Multi-agent Systems, Motion Control
Abstract: This paper investigates cooperative motion planning for communication-constrained formation turning of Dubins vehicles. The vehicles move at constant speed and are subject to sector-limited communication constraints. To ensure connectivity during large-angle maneuvers, a structured cooperative turning strategy is developed. The overall heading change is decomposed into repetitive minimal turning units, each consisting of equivalent approach, steering, retreat, and steering phases. By properly selecting adjustable parameters, the relative distance between neighboring vehicles first decreases and then returns to the desired formation spacing, while the relative bearing remains within the communication sector. Consequently, formation connectivity is preserved throughout the maneuver provided that the communication range exceeds the nominal spacing. Simulation results validate the effectiveness of the proposed cooperative motion planning strategy.

Keywords: Optimal Control, Nonlinear Systems and Control
Abstract: We implement the universal Birkhoff pseudospectral theory over Legendre-Gauss-Radau grid by developing new covector mapping theorem for generic nonlinear programming algorithms. Considering the barriers between the stationarity conditions and optimality, we utilize the tightened Birkhoff theories specifically for nonlinear programming implementations to augment and relax primal conditions, and the relating covector mapping theorem is constructed with the specially modified Lagrangian. Examples of singular optimal control problem are provided, demonstrating the effectiveness and potential generality of our new theories for implementations.

Keywords: Robotics, Control Applications
Abstract: This paper presents an iris-inspired tactile gripper system for aerial grasping. Unlike previous grippers designed for aerial robots, the proposed gripper integrates compliant tactile sensors that provide both passive adaptability and active sensing capability. First, we design the mechanical structure of the gripper based on the radially symmetric contraction mechanism of the biological iris. It uses a single motor to actuate an embedded cable for the synchronized opening and closing of six fingers. Then, three of the rigid fingers are replaced with TacTip optical tactile sensors, which enable contact perception by tracking internal marker displacements and support feedback-driven grasp control. Finally, a multi-stage grasping strategy is developed that uses a finite-state machine to achieve a stable grasp. Experimental results show that the proposed system significantly improves the grasp success rate compared with a baseline gripper without tactile sensing. Preliminary integration with a quadrotor UAV platform is also demonstrated, highlighting the potential of the system for aerial manipulation tasks.

Keywords: Estimation and Identification, Control Applications, Robotics
Abstract: Attitude stability plays a key role in flight control performance for unmanned aerial vehicles, especially for the multi-rotor configuration with large propellers that are intrinsically sensitive to wind disturbance and actuator delay. Sophisticated modeling strategies are hence required to achieve desired performance with sufficient bandwidth. In this paper, we propose a system identification method for a multi-rotor UAV under the hovering condition with direct model-based excitation. Experiment setup and implementation details are also given. Results show that the proposed identification method is quite friendly for implementation, and the identified model in transfer function form can capture dynamics frequencies up to 20 rad/s.

Keywords: Nonlinear Systems and Control, Control Applications, Robotics
Abstract: Hybrid ground/aerial vehicles possess cross-domain operation capabilities, enabling them to accomplish more complex tasks,thus holding significant application value in various fields. However, existing systems generally suffer from issues such as bulky structure, unsmooth modal switching, and poor ground adaptability. In this paper, a foldable aerial-ground multimodal robot system and its control method are proposed. The flight mode adopts a geometric controller to ensure stable flight, while the ground mode achieves omnidirectional movement through a motion controller. The system features an ingenuous structural design, low power consumption, and strong environmental adaptability. The control method exhibits low computational complexity and easily tunable parameters, meeting real-time control requirements. Finally, the effectiveness and performance of the proposed method are verified in the ROS-Gazebo simulation environment.

Keywords: Networked Control, Fault Detection and Diagnostics
Abstract: There is a crucial yet challenging issue in the study on performance improvement of the observer-based fault estimation approaches. That is, the unknown fault signal usually gives rise to very low usage rate of fault estimation error (FEE) information, making it difficult to further enhance the estimation accuracy. To address this, the technical note develops an indirect approach to acquire the FEE by using current and historical sensor measurements along with their estimations, and then proposes a novel fault observer with PI-type error-feedback structure. The new structure is able to make good use of the FEE, thereby significantly enhancing the estimation accuracy in comparison to the existing results. An illustrative example validates the presented method.

Keywords: Multi-agent Systems, Adaptive Control
Abstract: This paper is concerned with the fault-tolerant consensus problem of heterogeneous multi-agent systems with unknown parameters. The design of the cooperative controller is challenging due to the coupling among unknown parameters, actuator failures and network topology analysis. To conquer this challenge, the hierarchical control framework consisting of the virtual layer and the physical layer is put forward in this paper. The virtual layer achieves the virtual cooperative control objective by designing the adaptive virtual cooperative controller. The physical layer estimates unknown parameters via constructing a parameter identifier, and further develops a tracking fault tolerant controller to accomplish the consensus control task. The theoretical findings are illustrated by a simulation example.

Keywords: Estimation and Identification, Fault Detection and Diagnostics, Linear Systems
Abstract: This paper studies the vulnerability of remote state estimation systems under physical false data injection (FDI) attacks. The attacker seeks to keep the attack stealthy with respect to residual-based detectors while causing the bias in the physical process state to diverge to infinity. We establish the necessary and sufficient conditions for achieving this objective, i.e., the measurement matrix is not full rank or the system matrix A is unstable. For the first attack scenario, we show that the resulting security vulnerability can be addressed by employing a moving target defense (MTD) strategy. For the second scenario, however, the attacker can achieve the same objective by only slightly perturbing the initial state, which makes the attack much harder to detect and prevent. This observation suggests an important avenue for future work. Numerical simulations are finally presented to verify the developed theoretical results.

Keywords: Fault Detection and Diagnostics, Estimation and Identification, Nonlinear Systems and Control
Abstract: This work presents a systematic method for the design of residual generators that enable both fault detection and estimation in discrete-time nonlinear systems. The proposed residual generator is a linear functional observer built for an extended system that integrates fault dynamics modeled from a linear exo-system. Additionally, the proposed residual generator possesses disturbance-decoupling properties allowing it to isolate the effects of faults from those of unknown disturbances. Necessary and sufficient conditions for the existence of such residual generators for discrete nonlinear systems are derived. As long as these conditions are satisfied, explicit design formulas for the residual generators are obtained. The results are illustrated through a chemical reactor case study, which shows that the proposed scheme estimate faults accurately.

Keywords: Fault Detection and Diagnostics, Intelligent and AI Based Control, Fuzzy and Neural Systems
Abstract: This paper proposes an AI-driven pipeline leak detection framework based on UAV thermal imagery and a hybrid CNN–Transformer architecture. The proposed method integrates convolutional feature extraction with Transformer- based global attention to effectively capture both local thermal patterns and long-range spatial dependencies. To improve robustness under complex environmental con- ditions, a physics-aware thermal constraint is introduced, en- forcing consistency between predicted and actual temperature gradients. Experimental results on a dataset of 4200 UAV thermal images demonstrate that the proposed method outperforms con- ventional CNN and Transformer-based approaches, achieving up to 0.90 detection accuracy. Real-time evaluation shows that the system operates at 22 FPS, making it suitable for practical deployment. The proposed approach provides an efficient and scalable solution for intelligent pipeline monitoring and early leak detection.

Keywords: Fault Detection and Diagnostics, Process Control & Instrumentation, Signal Processing
Abstract: Anomaly detection aims to identify abnormal patterns in large-scale, dynamically evolving time-series data, however, conventional methods often struggle to effectively model high-dimensional feature representations and long-term dependencies. To address this challenge, this paper proposes a causality-enhanced normalizing flow (CENF) framework that introduces piecewise time-varying directed acyclic graphs (DAGs) to model latent causal structures. The proposed CENF framework integrates conditional normalizing flows with causal graph structures, enabling invertible probability density modeling while capturing time-varying causal dependencies. During training, intervention-consistency and counterfactual-contrastive regularization are incorporated to implement soft intervention mechanisms that guide the dynamic evolution of the causal structure while enforcing approximate invariance among non-descendant nodes. In addition, a fused Lasso constraint is introduced to sparsely detect operating-condition transitions, thereby enhancing the model’s adaptability to complex and dynamic environments. Experimental results on real industrial process datasets demonstrate that the proposed CENF method achieves superior anomaly detection accuracy under complex operating conditions.

Keywords: Nonlinear Systems and Control, Control Applications, Adaptive Control
Abstract: This paper proposes a new asymptotically stabilized control method for Fractional-order Memristor-Based Neural Networks (FMBNN) with active links to fractional orders. The traditional fractional-order stability criterion is firstly advanced into a form that establishes a scientific relationship with the fractional order. Then, a novel feedback control linking fractional order method is proposed to control the FMBNN based on this propellant criterion. And next, the asymptotic stabilization of the controlled FMBNN is discussed in detail, using the method of combining vector Lyapunov function with M-matrix. Finally, the effectiveness as well as the superiority of the proposed method is verified by comparing the traditional feedback control method with the novel feedback control linking fractional order method by numerical simulation of the controlled FMBNN with larger time delay from low-dimensional to high-dimensional.