Driven by Industry 4.0, smart factories are transitioning from discrete automation to hyper-connected intelligence. As the logistical core, cranes are evolving from human-operated tools into autonomous intelligent robots. However, this transformation is hindered by high-order underactuated dynamics, environmental disturbances, and NP-hard scheduling. Digital Twin (DT) technology offers a disruptive solution by integrating cyber-physical systems (CPS) through high-fidelity mirrors and closed-loop interaction. This paper systematically reviews DT systems for autonomous cranes, proposing a "Four-Layer and One-Network" reference architecture. We specifically address critical implementation challenges: ensuring sub-millisecond control latency to prevent "twin-drift," implementing blockchain-secured communication within the Asset Administration Shell (AAS) to mitigate cyber-threats, and providing a "Gateway-Wrap" strategy for retrofitting legacy equipment. Furthermore, we analyze four theoretical pillars: multimodal perception, adaptive anti-sway control, DRL-based scheduling, and XR-driven interaction. Experimental discussions highlight the impact of RTK antenna positioning and environmental factors like signal blockage on system reliability. Finally, we envision the evolution toward "cognitive autonomous twins" powered by Generative AI.

Foreign object detection on underground coal mine conveyor belts is essential for ensuring production safety, yet challenges such as uneven illumination, dust interference, and motion blur significantly affect detection accuracy and real-time performance. To address these issues, this paper proposes a lightweight detection model, CBFOD-YOLO, based on an improved YOLOv11 architecture. The model incorporates a C3K2_PConv module, which embeds partial convolution (PConv) into the C3K2 structure to expand the receptive field while reducing parameters and computational cost, thereby improving feature extraction efficiency. In addition, a Content-Guided Attention (CGA) mechanism is introduced to generate channel-specific spatial importance maps, enhancing key feature representation and suppressing background noise. Experimental results on a public coal mine foreign object dataset demonstrate that CBFOD-YOLO achieves 90.4

Continuous tool operation with a myoelectric prosthetic hand is considerably more complex than discrete grasping tasks. This complexity arises because the control system must maintain stable, adaptive, and coordinated motions under varying loads and unpredictable interactions. In human motor control, this stability is achieved through a biological sensorimotor closed loop, where tactile feedback continuously modulates neural signals to adapt to environmental changes. Inspired by these mechanisms for reducing grasp instability caused by external shocks, this study designed a multimodal controller termed the tactile, kinesthetic, and electromyography (EMG) bionic gripping controller (TKE-BGC). It integrates tactile, kinematic, and EMG information. Initially, multimodal data—encompassing tactile signals, joint angles, and EMG patterns—were collected from able-bodied users during tool manipulation via a data glove. Subsequently, the TKE-BGC model was trained on these data, utilizing a Transformer encoder to extract high-level features and a multilayer perceptron to predict joint angles in real time. Based on this controller, this paper presents a prosthetic control framework developed through human skill transfer. Unlike conventional fixed force or force follows strategies that struggle with dynamic impacts or tracking delays, this framework enables robust end-to-end adaptive control. Tested across 4 seen and unseen tool operation tasks, the proposed method demonstrated precise detailed performance. Specifically, it significantly reduced the number of tool drops and shortened task completion times compared to the baseline methods. Furthermore, it achieved human-like average contact forces and substantially lowered the user’s physical workload, requiring noticeably less muscle effort than the force follows strategy (e.g., average EMG amplitude, 0.0023 versus 0.0124). By rapidly adjusting grip force through feedback and effectively mitigating instability, this research holds significant practical value in enhancing the daily independence of amputees and supporting their vocational rehabilitation and reemployment.

Accurate weld seam segmentation is essential for reliable defect detection. This study addresses the segmentation of weld seams on the top covers of prismatic batteries using point cloud data, where the primary challenge arises from the minimal height difference between the weld seam and the surrounding surface. To tackle this issue, we propose a segmentation framework that integrates multi-stage geometric processing. Voxel filtering is first applied to reduce redundancy while preserving structural integrity. Subsequently, the top and side shell planes are fitted via the least squares method, and their intersection is used for coarse localization of the weld region. Finally, an adaptive normal-based segmentation algorithm is employed to achieve precise seam extraction. Experimental results verify that the proposed method enables accurate and robust weld region segmentation, providing a reliable foundation for subsequent defect detection in battery manufacturing.

Ground segmentation-based point cloud processing methods have gained significant attention in simultaneous localization and mapping (SLAM) for LiDAR-inertial odometry applications, particularly in autonomous vehicles. However, existing methods often lack robustness in complex environments, such as uneven terrains or obstructed areas. To address these challenges, we propose a novel point cloud segmentation algorithm that combines vector iteration processing and ground expansion techniques. This method iteratively updates the reference vector using the normal vectors of successfully fitted planes, while expanding the ground points by incorporating neighboring planes. Furthermore, to mitigate redundant calculations in conventional approaches, we introduce a grid-based local plane fitting method. By partitioning the ground into smaller regions and employing a custom node structure, this technique updates plane parameters iteratively, improving both fitting accuracy and computational efficiency. For non-ground point clouds, we propose an iterative principal component analysis method combined with outlier filtering. Experimental evaluations were conducted using 20 sequences from four public datasets and a custom campus dataset. The proposed method achieved an average time savings of 26.7% compared to state-of-the-art approaches, while maintaining superior map accuracy. This work contributes to advancing robust and efficient point cloud processing for autonomous vehicle applications in challenging environments.

In the field of LiDAR-based Simultaneous Localization and Mapping, the potential of ground point clouds to enhance pose estimation in mobile robots has yet to be fully realized. This paper focuses on leveraging ground point clouds to improve the performance of LiDAR-Inertial Odometry (LIO) systems for ground-based mobile robots. We begin by analyzing the characteristics of ground point clouds and the typical types of noise that affect their extraction and utilization. Ground point clouds are then extracted from denoised data. Given the generally local planar nature of ground point clouds, we propose a segmentation-and-refitting approach to process them. This method reduces the computational burden of residual calculation in pose estimation by avoiding redundant plane fitting. Additionally, we introduce a data structure designed for the efficient management and utilization of ground point clouds obtained through segmentation and refitting. This structure is particularly suited to the ground point cloud data produced by our method, enabling efficient access and registration through the continuous maintenance and consolidation of local plane parameters. Our method has been integrated into advanced LIO systems (Bai et al 2022 IEEE Robot. Autom. Lett. 7 4861-8), and experimental results on challenging datasets demonstrate its promising performance.

Image segmentation plays a crucial role in target recognition as it provides an essential prerequisite for identifying targets. However, UNet cannot acquire comprehensive information about the image, which limits its ability to effectively segment small and discrete objects. A new network structure, called MorNet, is proposed. It applies dilation and erosion operations in morphological opening and closing for connected and discrete region segmentation. MorNet effectively integrates global segmentation images, emphasizing small and discrete area segmentation and enhancing the segmentation of large continuous areas. This method enhances the understanding of the location of organs and diseased sites, enabling accurate segmentation. A side-out (SO) monitoring mechanism is introduced into the decoder to provide monitoring information for the original image and the predicted results. Compared with current UNet-like networks, our proposed method achieved the highest Dice scores (96.4635

sEMG (surface electromyography) signal control of bionic prostheses has been widely studied over the past few years. In particular, sparse sEMG signals are rapidly developing in the field of gesture recognition for their convenience, noninvasiveness, and ease of access. However, compared with high-density EMG signals, sparse EMG signals lack rich feature information, which in turn affects gesture recognition accuracy. In order to reduce the loss of feature information of sparse EMG signals in the spatio-temporal dimension, this paper proposes a hybrid neural network Res-LSTM combining residual network and long short-term memory network. The ordinary convolutional blocks in the CNN network are replaced with residual blocks, and the final fully connected layer is removed and a constant mapping layer is added to adequate extraction of spatial feature information of the data. The Res layer’s output is utilized as the input for the long and short-term memory (LSTM) network, which further extracts the features of the data in the temporal dimension, and finally completes the categorization output through a fully connected layer. The average accuracy of 91.11

The inverse kinematics of robotic manipulators involves determining an appropriate joint configuration to achieve a specified end-effector position. This problem is challenging because the inverse kinematics of manipulators are highly nonlinear and complexly coupled. To address this challenge, the bald eagle search optimization algorithm is introduced. This algorithm combines the advantages of evolutionary and swarm techniques, making it more effective at solving nonlinear problems and improving search efficiency. Due to the tendency of the algorithm to fall into local optima, the Lévy flight strategy is introduced to enhance its performance. This strategy adopts a heavy-tailed distribution to generate long-distance jumps, thereby preventing the algorithm from becoming trapped in local optima and enhancing its global search efficiency. The experiments first evaluated the accuracy and robustness of the proposed algorithm based on the inverse kinematics problem of manipulators, achieving a solution accuracy of up to 10−18 m. Subsequently, the proposed algorithm was compared with other algorithms using the CEC2017 test functions. The results showed that the improved algorithm significantly outperformed the original in accuracy, convergence speed, and stability. Specifically, it achieved over 70% improvement in both standard deviation and mean for several test functions, demonstrating the effectiveness of the Lévy flight strategy in enhancing global search capabilities. Furthermore, the practicality of the proposed algorithm was verified through two real engineering optimization problems.

Autonomous driving has gradually become a research hotspot in recent years. Visual Simultaneous Localization and Mapping (SLAM) technology can help unmanned vehicles accurately explore the environment at a lower cost, and the readability of the map can be improved by integrating target detection algorithms. However, the location and 3D shape of the object in the map were not obtained. The method of RGBD-SLAM based on object detection with two-stream YOLOv4-MobileNetv3 convolutional neural network is proposed in this paper. RGBD SLAM algorithm and target detection algorithm are combined to build an algorithm model that can generate the global sparse map and build target dense map quickly. The two-stream network is integrated to obtain 2D information about the target, and further combined with the camera pose after the front-end key frame detection of the SLAM algorithm in this paper, and the dense 3D point cloud of the target and the center point position of the object is obtained. Then, the sparse point cloud of the SLAM system and the dense point cloud of the target can be obtained. The experimental results show that the number of point clouds decreases by about 50% and the time for mapping accounts for about 60% of the global dense mapping time. The method of this paper can efficiently decrease the computational space and improve the speed of semantic mapping, which verifies its feasibility and superiority. It can be used to achieve large-area mapping and the ability to update maps during autonomous driving.

The assumptions of a static environment and scene rigidity are important theoretical underpinnings of traditional visual simultaneous localization and mapping (SLAM) algorithms. However, these assumptions are difficult to work in dynamic environments containing non-rigid objects, and cannot effectively handle the characteristics of local areas of non-rigid moving objects, seriously affecting the robustness and accuracy of the SLAM system in localization and mapping. To address these problems, we improved ORB-SLAM3 and proposed a real-time RGB-D visual SLAM framework for dynamic environments based on StrongSORT-Strong-SLAM. First, we combine YOLOv7-tiny with StrongSORT to match the semantic information of dynamic targets. Optical flow and epipolar constraints are then used to initially extract geometric and motion information between adjacent frames. Subsequently, based on an improved adaptive threshold segmentation algorithm and geometric residuals, a background model and a Gaussian residual model are constructed to further extract the geometric information of dynamic targets. Finally, semantic and geometric information are integrated to perform global feature motion level classification, and motion probabilities and optimization weights are defined to participate in global pose estimation and optimization. Experimental results on the publicly available TUM RGB-D dataset show that Strong-SLAM reduces the absolute trajectory error and relative pose error by at least 90% compared to ORB-SLAM3, achieving performance comparable to the most advanced dynamic SLAM solutions.

Collision detection is very important for robot motion planning. The existing accurate collision detection algorithms regard the evaluation of each node as a discrete event, ignoring the correlation between nodes, resulting in low efficiency. In this paper, we propose a novel approach that transforms collision detection into a binary classification problem. In particular, the proposed method searches the k-nearest neighbor (KNN) of the new node and estimates its collision probability by the prior node. We perform the hierarchical navigable small world (HNSW) method to query the nearest neighbor data and store the detected nodes to build the database incrementally. In addition, this research develops a KNN query technique tailored for linear data, incorporating threshold segmentation to facilitate collision detection along continuous paths. Moreover, it refines the distance function of the collision classifier to enhance the precision of probability estimations. Simulation results demonstrate the effectiveness of the proposed method.

The seven-degree-of-freedom space manipulator, characterized by its redundant and aspheric wrist structure, is extensively used in space missions due to its exceptional dexterity and multi-joint capabilities. However, the non-spherical wrist structure presents challenges in solving inverse kinematics, as it cannot decouple joints using the Pieper criterion, unlike spherical wrist structures. To address this issue, this paper presents a closed-form analytical method for solving the inverse kinematics of seven-degree-of-freedom aspheric wrist space manipulators. The method begins by identifying the redundant joint through comparing the volumes of the workspace with different joints fixed. The redundant joint angle is then treated as a parametric joint angle, enabling the derivation of closed-form expressions for the non-parametric joint angles using screw theory. The optimal solution branch is identified through a comparative analysis of various self-motion manifold branches. Additionally, a hybrid approach, combining analytical and numerical methods, is proposed to optimize the parametric joint angle for a trajectory tracking task. Simulation results confirm the effectiveness of the proposed method.

sEMG(surface electromyography) signals have been widely used in rehabilitation medicine in the past decades because of their non-invasive, convenient and informative features, especially in human action recognition, which has developed rapidly. However, the research on sparse EMG in multi-view fusion has made less progress compared to high-density EMG signals, and for the problem of how to enrich sparse EMG feature information, a method that can effectively reduce the information loss of feature signals in the channel dimension is needed. In this paper, a novel IMSE (Inception-MaxPooling-Squeeze- Excitation) network module is proposed to reduce the loss of feature information during deep learning. Then, multiple feature encoders are constructed to enrich the information of sparse sEMG feature maps based on the multi-core parallel processing method in multi-view fusion networks, while SwT (Swin Transformer) is used as the classification backbone network. By comparing the feature fusion effects of different decision layers of the multi-view fusion network, it is experimentally obtained that the fusion of decision layers can better improve the classification performance of the network. In NinaPro DB1, the proposed network achieves 93.96% average accuracy in gesture action classification with the feature maps obtained in 300ms time window, and the maximum variation range of action recognition rate of individuals is less than 11.2%. The results show that the proposed framework of multi-view learning plays a good role in reducing individuality differences and augmenting channel feature information, which provides a certain reference for non-dense biosignal pattern recognition.

Visual odometry constitutes a critical component in enabling autonomous navigation. However, the existing methods are limited by the feature extraction and matching accuracy, and cannot show good real-time performance while combining accuracy and robustness. In this paper, we propose a novel monocular visual odometry framework based on cross-correlation. The framework starts with a parameter-sharing Siamese network to build feature extractors that can simultaneously process multiple images as inputs. Moreover, we design cross-correlation modules and define a cross-correlation matrix to describe the strength of correlation between different parts of the input feature maps, reflecting the rotational and translational transformations of the input images. Furthermore, a novel loss function is introduced to impose constraints on the network. Additionally, a fully convolutional network is designed for pose estimation, computing poses alterations from the structure of the cross-correlation matrix. Channel attention and spatial attention mechanisms are introduced to improve the performance. More importantly, our method innovatively uses time intervals as labels, enables self-supervised training, and relies only on a monocular camera. Experimental results on the KITTI visual odometry dataset and the Oxford Robotcar Dataset show that our method produces competitive performance, demonstrating the superiority of the proposed method.

Advances in the field of measurement science and technology have improved the detection of defects in industrial production. One of the key challenges in steel plate surface defect detection is the need to quickly detect a small number of defects in an overwhelmingly defect-free sample. Unlike supervised learning, which relies heavily on precise sample labeling, unsupervised learning leverages its inherent learning capabilities for detection. This paper introduces an innovative method for smart steel diagnosis, integrating joint optimization of feature extraction and clustering. The proposed approach merges mini-batch K-means clustering with a feature extraction network to acquire pseudo-label information for current images. It employs a multi-view transformation strategy, enabling classification through the optimized feedback from pseudo-labels. This method allows the network to self-optimize the distinction of image features through backpropagation. The method exhibits a mere 4% classification failure rate for steel surface images. This significant reduction in additional data processing requirements enhances the inspection system's efficiency and accuracy. Furthermore, the versatility of this method extends beyond steel defect diagnosis. It holds potential for application in various engineering domains, particularly in scenarios characterized by data imbalance.

Self-supervised depth estimation has achieved remarkable results in sunny weather. However, in the foggy scenes, their performance is limited because of the low contrast and limited visibility caused by the fog. To address this problem, an end-to-end feature separation network for self-supervised depth estimation of fog images is proposed. We take paired clear and synthetic foggy images as input, separate the image information into interference information (illumination, fog, etc.) and invariant information (structure, texture, etc.) by a feature extractor with orthogonality loss. The invariant information is used to estimate depth. Meanwhile, similarity loss is introduced to constrain the fog image depth using the depth of the clear image as a pseudo-label, and an attention module and reconstruction loss are added to refine the output depth, so that better depth maps can be obtained. Then, real-world fog images are used for fine-tuning, which effectively reduces the domain gap between synthetic data and real data. Experiments show that our approach produces advanced results on both synthetic datasets and Cityscape datasets, demonstrating the superiority of our approach. (c) 2022 Society of Photo-Optical Instrumentation Engineers (SPIE)

This paper presents the collision-free solution to the inverse kinematics of a mobile manipulator. The main contribution is to propose a novel inverse kinematics solution framework combining unique domains. The inverse kinematics problem is formulated as an optimization problem in this work. Firstly, the objective function is simplified by the kinematics decoupling method based on improved disconnection and re-connection. Then, the joint variables are decoupled and constrained, and an approach to initialize the variables is proposed. An inverse kinematic solution framework for the manipulator body to avoid obstacles is further presented, and CMA-ES is applied to the framework. Finally, the simulation results demonstrate the effectiveness of the proposed method.