Keywords: Fuzzy Systems, Cybernetics Automation and Control
Abstract: Electro-optical gimbals in industrial inspection face disturbances, jitter, and limited tracking accuracy. This study models a two-axis, four-frame gimbal and proposes a fuzzy sliding mode control (FSMC) strategy, using fuzzy logic to adjust the sliding-mode gain in real time for jitter suppression and performance improvement. Simulations show faster response and lower overshoot than conventional sliding mode control. Experiments confirm a maximum tracking error of 0.48 mrad under diverse disturbances. The FSMC-based line-of-sight control effectively enhances stability and tracking precision, enabling reliable continuous target locking in complex inspection scenarios.

Keywords: Medical Robots and Systems, Haptics, Tele-robotics
Abstract: This paper presents a probabilistic shared-control framework for teleoperated robotic needle puncture to mitigate tracking latency and the "pop-through" effect. The system integrates a Force Dimension Omega.6 master device with an EtherCAT-driven dual-axis slave mechanism and a Kunwei KWR36 6-DOF force sensor. A 7-state Markov transition model, utilizing fuzzy logic fusion of real-time probability metrics, dynamically adjusts four control weights to provide adaptive force protection and autonomous needle rotation. Experimental results in synthetic tissue phantoms demonstrate a 20% reduction in peak puncture force and significantly enhanced position tracking compared to pure bilateral teleoperation. The framework ensures statistically secure control handovers based on verified human intent.

Keywords: Force/Impedance Control, Motion Control, Dynamics
Abstract: The space manipulator system needs to be autonomous to perform on-orbit operations such as satellite servicing, refueling, and active debris removal, especially the removal of non-cooperative space targets (NCSTs). The impact phase is the most critical, as the end effector must maintain stable contact with the NCST to prevent separation without disturbing the free-floating base. Further, the harsh environment increases complexity during post-capture manipulation due to uncertainties in dynamics, ranging from internal factors such as joint friction and unknown inertial parameters to external disturbances, which challenge traditional model-based control methods. This paper proposes an integrated control that combines impedance control with an estimation method to address model uncertainty. The controller regulates the mechanical impedance of the end effector to manage physical interaction by forcing it as a mass-spring-damper system. This ensures the effective absorption of impact energy and damping of relative velocity without disturbing the base motion. To make the system robust to model uncertainties, a stochastic prediction model, such as a GP, is employed to estimate unmodeled parameters and environmental disturbances. The robustness of the control framework against model uncertainties is demonstrated through reductions in input impedance force and relative acceleration to maintain contact stability. Furthermore, SGP is employed to reduce the computation cost and speed up the estimation process.

Keywords: Biologically-Inspired Robots and Systems, Multi-Agent Systems, Cooperative Systems and Control
Abstract: The paper proposes a novel bio-inspired control strategy for co-operative control of heterogeneous multi-robot systems (MRS) in industrial indoor settings under low lighting conditions. The design of the methodology is based on the hierarchical task assignment process observed in the leaf-cutter ant colony. The proposed methodology provides a decentralized coordination for multiple robots without any physical interconnection among them in a dynamic indoor environment. The model integrates an optimal task assignment process between the heterogeneous robots, a decentralized path-planning framework based on the Non-linear MPC control policy (NMPC). This framework incorporates mapping, localization, trajectory tracking, non-linear control, and decision making to attain complete autonomy. The feasibility of this method is evaluated via simulation and experimental testing. Based on simulation tests, the NMPC is found to show good performance at point stabilization. It has steady-state error found to be is as low as 0.96% after 4 seconds. During obstacle avoidance, the robot is able to converge towards the target position in 5 seconds with a steady-state error of 0.81%. Leader follower and trajectory tracking experiments exhibit efficient performance for common trajectories, with an error of 1.14% attained within 25 seconds.

Keywords: Image Processing, Deep Learning, Neural Networks
Abstract: Materials characterization imaging—X-ray and electron diffraction, scanning electron microscopy—routinely operates in the small-sample regime, where conventional augmentation either distorts physically meaningful structure (pixel mixing) or collapses under data scarcity (generative models). We present Diffusion-Informed Feature Smoothing (DIFS), a training-free operator that smooths the empirical distribution of frozen foundation-model features under an optimal-transport geometry. DIFS casts augmentation as maximum entropy over a Wasserstein ball, whose Lagrange dual yields a Gibbs kernel sampled by Mahalanobis rejection inside the ball. Across three real materials datasets and five small-sample scenarios, DIFS is best or statistically tied on Matthews correlation and accuracy, and significantly improves over no augmentation in the hardest data-scarce and cross-modal settings. Its expansion variance follows a provable law with intrinsic dimension, exposing a low-dimensional feature manifold.

Keywords: Kinematics, Modeling
Abstract: In industrial applications, such as robotic welding and assembly operations, the absolute positioning accuracy of robots is crucial for production efficiency and product quality. To improve the absolute positioning accuracy of serial robots, optimization efforts mainly focus on two aspects: error model construction and parameter identification algorithms. First, a comprehensive geometric error model is established based on the product of exponentials (POE) formula, and the identifiability of the geometric parameter errors is analyzed using a self-developed 9-degree-of-freedom (9-DOF) redundant manipulator. Second, a hybrid identification approach integrating the Levenberg–Marquardt (LM) algorithm and the bald eagle search (BES) algorithm is proposed. The proposed method integrates the fast convergence of gradient-based algorithms with the random search capability of metaheuristic algorithms, thereby improving the accuracy and robustness of parameter identification. Finally, the effectiveness of the proposed calibration model and identification algorithm is validated through experiments on a self-developed 9-DOF redundant manipulator. The experimental results demonstrate that the robot achieves a positioning root mean square error within 0.3 mm and an orientation error within 0.2◦ after calibration.

Keywords: Robotics and Automation Applications, Modeling, Wheeled Mobile Robots
Abstract: This study proposes a multi-sensor fusion-based Simultaneous Localization and Mapping framework (LIOG- SLAM) to address positioning errors and drift issues encountered by substation inspection robots in large-scale substations. By integrating 3D LiDAR, Inertial Measurement Unit (IMU), Wheel Odometry (ODO), and Global Navigation Satellite System (GNSS) data, combined with ground- optimized odometry constraints, IMU-ODO joint preintegration, sliding window marginalization, and loop closure detection, positioning accuracy and system robustness have been enhanced. In the KITTI dataset and Gazebo simulation environment, LIOG-SLAM demonstrated higher accuracy and lower error compared to traditional algorithms such as A- LOAM and LIO-SAM. Specifically, the root mean square error of APE decreased by 9.1 m and 5.207 m on the same sequence dataset, while maintaining efficient real-time performance in complex environments. This method offers a novel solution for efficient inspection in substations.

Keywords: Embodied AI, Robot Vision
Abstract: Diffusion Policy has demonstrated strong performance in long-horizon robotic manipulation by generating smooth and executable action trajectories through conditional denoising. However, practical deployment remains limited by two key challenges. First, inadequate modeling of pose and position features for grasping and placing tasks reduces generalization under randomized initial states. Second, frequent violations of physical execution constraints (e.g., joint position and torque limits) lead to joint overshoot and control oscillations during long-horizon inference. To address these issues, we propose an integrated framework that combines perception enhancement with execution-constraint enforcement. On the perception side, we introduce CA-ResNet by inserting a Coordinate Attention module into Layer 4 feature maps of ResNet, enabling direction-aware attention along both spatial axes. On the execution side, we design a training-free Overshoot Recovery Safety Layer with three stages: near-limit detection, adaptive recentering, and safety clipping. Experiments on NVIDIA Isaac Sim with a Franka Panda platform show that our method significantly outperforms DP-UNet and DP-Transformer baselines in success rate, positioning accuracy, and pose stability. Ablation studies further confirm the effectiveness of both components and their complementary gains.

Keywords: Robotics and Automation Applications, Haptics, Methodologies for Robotics and Automation

Keywords: Cybernetics Automation and Control, Computational Intelligence
Abstract: This paper investigates the discrete-time solution of time-varying quadratic programming (TVQP) problems with linear equality constraints in noisy environments. Starting from the Karush--Kuhn--Tucker conditions, a perturbation-suppressed zeroing neural dynamics model is established to describe the online evolution of the optimal solution. For sampled-data implementation, a seven-instant discretization scheme is developed, yielding the proposed SE-DT-TVQP algorithm. For comparison, Euler-type and Taylor-type discrete-time formulations are also considered. Numerical experiments consisting of a baseline case without noise under the fixed sampling interval g=0.01, a constant-noise case, and a linearly time-varying noise case show that the proposed algorithm consistently achieves lower steady-state residuals and stronger noise resilience than the benchmark methods. These results demonstrate that the proposed seven-instant sampled-data design provides an effective sampled-data approach for online TVQP computation under perturbations.

Keywords: Robot Vision, Neural Networks, Deep Learning
Abstract: Event cameras can perceive the continuous motion of a scene at high temporal resolution, thereby providing effective motion cues for optical flow estimation. In recent years, some methods have attempted to jointly leverage images and events to improve performance. However, most of them directly fuse the two modalities, overlooking the differences between image appearance and event boundaries, which easily leads to modal interference and limits the utilization of complementary information. Moreover, to meet the practical deployment requirement of event-only input, existing distillation methods still struggle to effectively handle the heterogeneous modality gap between images and events. To this end, we propose an Appearance-Boundary Fusion Distillation framework (ABFD) for event-based optical flow estimation. Specifically, the proposed method first employs an Appearance-Boundary Decoupled Fusion module (ABF) to explicitly model image appearance information and event boundary information, and further generates high-quality fused features through asymmetric cross-modal interaction. Subsequently, a Motion-Aware Distillation strategy (MAD) is introduced to transfer effective motion representations from the fused teacher to the event-only student model under the supervision of the optical flow task. During inference, only the event-based model is retained, enabling efficient and robust optical flow estimation without requiring additional image input. Extensive experiments demonstrate that the proposed method achieves superior performance over existing methods.

Keywords: Medical Robots and Systems, Human/Machine Systems, Deep Learning
Abstract: Walking, going upstairs, and going downstairs are fundamental gait patterns in lower limb exoskeleton assistance systems, and accurate recognition of these activities is essential for adaptive control. However, distinguishing these patterns remains challenging due to the high similarity of acceleration signals, limited training samples, and class imbalance, especially between walking and going downstairs. To address these issues, this paper proposes a CNN-LSTM deep learning method that combines handcrafted features and an attention mechanism for gait activity recognition. The core focus of this work is on constructing a robust feature engineering pipeline, including signal preprocessing and multi-dimensional feature extraction, as well as optimizing the deep learning architecture for enhanced discriminative ability. Specifically, time-domain, frequency-domain, and kinematic parameters are extracted from inertial sensor data to form a feature vector with clear physical meaning. A CNN is used to capture local motion patterns, while an LSTM models the temporal evolution of gait cycles. A Squeeze-and-Excitation (SE) channel attention mechanism is further introduced to adaptively weight different feature channels, improving the model’s sensitivity to subtle differences between similar activities. Experimental results indicate that the proposed method attains an impressive 99.48% overall accuracy in classification, coupled with a 0.9930 macro-averaged F1-score. effectively reducing the confusion between similar gait activities. This work provides a reliable perceptual foundation for applications such as motion intention recognition and personalized assistance control in lower limb exoskeletons. Keywords—activity recognition, lower limb exoskeleton, handcrafted features, attention mechanism, CNN-LSTM

Keywords: Transportation Systems
Abstract: During long-haul freight transportation, driver distraction and drowsiness are major causes of road traffic accidents. To address this safety hazard, this study designs and implements a vision-based real-time driver status monitoring and warning system tailored for long-haul truck driving scenarios. The system captures video streams via cameras and utilizes deep learning algorithms to identify non-driving behaviors and assess risk levels specific to long-haul truck drivers. It automatically issues warnings to drivers based on the risk assessment results, thereby reducing accident risks. This experiment not only completed a full closed loop of data collection, status recognition, risk assessment, and warning feedback in a laboratory setting, but also provided systematic performance evaluations in both simulated and real-world scenarios, verifying its engineering feasibility. The innovation of this study lies in combining traditional computer vision methods with cutting-edge vision-language models. Through time-series analysis based on multi-feature fusion, the system achieves a comprehensive assessment of long-haul truck driver status, offering an effective technical solution for preventing accidents in long-haul truck driving.

Keywords: Neural Networks, Image Processing
Abstract: To address the problem of tracking drift and target loss in UAV target tracking caused by occlusion and wind-induced disturbances, this paper proposes a DaSiamRPN-based target tracking method with shallow neural network model updating. By comparing the precision and success rates of seven classical tracking algorithms, DaSiamRPN, which exhibits the best overall performance, is selected as the baseline algorithm. A shallow network-based updating strategy that integrates the initial reference template is then designed, and a multi-stage iterative training scheme is adopted to enhance model robustness. Experimental results show that the improved algorithm significantly outperforms the original method on both VOT2018 public benchmark and three self-constructed power inspection sequences. In particular, under illumination variation and low-resolution conditions, both the precision and success rate reach 1.0. In scenarios involving small targets and large-scale variations, the proposed method improves precision and success rate by 0.048 and 0.059, respectively, over the original algorithm. These results indicate that the proposed method can provide reliable technical support for live-line UAV inspection operations.

Keywords: Deep Learning, Neural Networks
Abstract: To address the scarcity of cable-joint surface defect data for deep learning detection, this paper proposes a two-stage defect synthesis method based on diffusion models

and texture priors. An image prompt adapter is introduced into a pre-trained latent diffusion model to inject physical texture priors from a few real defect samples, enabling diverse and

texture-consistent defect generation without fine-tuning. Poisson image editing is then used for seamless defect background fusion, and a local dynamic differential enhancement algorithm combining Gamma correction and Otsu thresholding is designed for automatic pixel-level mask generation. Experiments on the CJDD dataset show that the proposed method achieves an FID of 29.28 and an LPIPS of 0.456. Using the synthesized data to train YOLOv8s improves

mAP50 from 0.487 to 0.612 and precision from 0.421 to 0.755. These results demonstrate the effectiveness of the proposed

method for few-shot defect dataset construction and downstream detection enhancement.

Keywords: Human/Machine Systems, Large Language Models, Artificial Intelligence Generated Content
Abstract: To investigate the collaborative mechanisms of generative artificial intelligence in programming instruction, this study conducted a 12week quasiexperiment with 80 higher vocational students. The experimental group used AI assistance, while the control group relied solely on conventional search engines. The results revealed heterogeneous effects of AI usage, with significantly greater intragroup variance than intergroup variance. Three interaction patterns were identified—lowengagement copying, passive debugging, and active constructing—among which only the activeconstructing pattern facilitated the internalization of programming thinking. Learning outcomes were optimal when the independent modification ratio was maintained within the 40%–60% range. Students who engaged in active error attribution demonstrated significantly better independent programming performance. Based on these findings, this paper proposes an “attributionfirst” humanAI collaborative teaching framework to inform programming education reform.

Keywords: Modeling, Artificial Intelligence Generated Content, Computational Intelligence
Abstract: To address the challenges of low production efficiency, heterogeneous AI tools, and non-standardized workflows in traditional micro-lecture production, this paper proposes a generative AI-driven intelligent micro-lecture production model with an AI toolchain scheduling mechanism. The proposed framework consists of five stages: instructional design, script generation, asset creation, knowledge explanation, and video synthesis. To improve automation and interoperability among heterogeneous AI tools, a task–tool weighted matching algorithm and standardized data interface are designed to support intelligent scheduling and seamless data transfer. In addition, an AI general education knowledge graph and educational prompt engineering strategy are introduced to ensure content consistency and pedagogical quality. Experimental validation conducted in university AI general education courses demonstrates that the proposed model reduces the micro-lecture production cycle by 68% and improves student learning effectiveness by more than 10%. The results indicate that the proposed framework provides an effective intelligent system solution for educational resource production and contributes to the application of AI-enabled automation technologies in digital education.

Keywords: Aerial Robotics, Robotics and Automation in Unstructured Environment, Robotics and Automation Applications
Abstract: Language-guided unmanned aerial vehicles (UAVs) often fail not from bad reasoning or perception, but from execution mismatch: the gap between a planned trajectory and the controller’s ability to track it when the real dynamics differ from training (mass changes, drag shifts, actuator delay, wind). We propose AeroBridge-TTA, a language-conditioned control pipeline that targets this gap with test-time adaptation. It has three parts: a language encoder that maps the command into a subgoal, an adaptive policy conditioned on the subgoal and a learned latent, and a test-time adaptation (TTA) module that updates the latent online from observed transitions. On five language-conditioned UAV tasks under 13 mismatch conditions with the same domain randomization, AeroBridge-TTA ties a strong PPO-MLP baseline in-distribution and wins all 5 out-of-distribution (OOD) conditions, +22.0 pts on average (62.7% vs. 40.7%); the +8.5 pt overall gain comes entirely from the OOD regime. A same-weights ablation that only changes the step size α shows the latent update itself is responsible for a 4.6×OOD lift.

Keywords: Image Processing, Deep Learning, Computational Intelligence
Abstract: The replacement of high-quality Toor Dal (Pigeon Pea) with toxic Khesari Dal (Grass Pea) and surreptitiously using the non-permitted azo dye Metanil Yellow as food adulterants has become a very serious threat to global food safety and health. This comprehensive review brings together the findings of recent scientific studies to critically analyze the toxicological effects of these food adulterants, as well as the effectiveness of innovative, non-destructive methods of their detection. Khesari Dal is found to contain the neurotoxin amino acid β-N-oxalyl-L- α, β-diaminopropionic acid (β-ODAP), proven to be the main agent responsible for neurolathyrism—a permanent paralytic syndrome. Metanil Yellow, on the other hand, a Category II toxic agent, is recognized as a powerful genotoxic and carcinogenic agent, causing the development of hepatocellular carcinomas and chromosomal The latest research underlines the paradigm shift from tradi- tional, destructive laboratory assays to rapid, intelligent sensing systems. Application of NIR spectroscopy (900–1700 nm), line- scan Hyperspectral Imaging (HSI) (900–2500 nm), and thermal imaging have effectively identified unique chemical and thermal “fingerprints” of adulterated samples. Integration of AI and Deep Learning architectures, including one-dimensional Convo- lutional Neural Networks (1D-CNN), ResNet-50, SqueezeNet, and YOLOv5, enabled automated classification and quantification with accuracies exceeding 94–98%, in general. Notably, the 1D- CNN models achieved a correlation coefficient of 0.992 with Metanil Yellow quantification (0–2%), without any manually performed spectral pre-processing. This review further considers portable edge-computing prototypes, like the Raspberry Pi- based “FoodExpert,” enabling democratization of high precision screening of 96% accuracy for end-users under marketplace conditions.Despite such recent progress, gaps still exist in model generalization across diverse varieties of pulses and environmen- tal sensitivity to ambient lighting and dust conditions. The paper concludes by emphasizing the need for data fusion in multimodal dimensionality and blockchainintegrated traceability to assure long-term integrity of the globa

Keywords: Planning and Control, Localization & Tracking, Wheeled Mobile Robots
Abstract: Collision avoidance for industrial AGVs operating in dynamic, shared environments remains challenging because reactive local planners treat moving obstacles as static, lead- ing to conservative or unsafe behavior. We address this by enabling the local planner to reason over predicted obstacle motion rather than instantaneous positions. Specifically, an Ensemble Kalman Filter (EnKF)-based multi-object tracker provides filtered position and velocity estimates that a modified Dynamic Window Approach (DWA) uses to evaluate candidate trajectories against projected future obstacle states. A radius- informed ensemble generation scheme adapts filter uncertainty to the observed object size from 2D LiDAR, and a geometric- center representation provides stable bounding estimates under partial occlusion. The system is implemented as three modular, pipelined ROS 2 nodes. Experiments in Gazebo, Stage (up to five concurrent robots), and on a real industrial forklift demonstrate collision-free navigation in same-direction, crossing, and head- on scenarios with sub-10 ms controller response times.

Keywords: Image Processing, Sensor Design, Integration and Fusion
Abstract: The grinding quality of cable joints directly affects the safe operation of power grids. However, traditional detection methods suffer from limitations such as poor environmental adaptability, strong subjectivity, or high costs, leading to heavy reliance on manual experience for on-site judgment. This paper proposes a field-oriented engineering framework based on 3D models and displacement maps. First, a blue-light speckle structured-light depth camera is used to capture surface images of cable joints, synchronously acquiring 3D point clouds. Second, an improved normal map calculation method is utilized to generate high-fidelity displacement maps, which are then fitted to the 3D models to supplement micro-geometric details and calculate various types of roughness parameters. Finally, a quality evaluation system based on parameter thresholds is established to realize quantitative judgment of grinding quality. Experiments on tested samples separate unqualified and qualified grinding states, supporting practical non-contact inspection.