Robotic-assisted vascular intervention (RAVI) demands high precision, continuous motion, and uncompromising safety for delicate interventional instruments like guidewires and catheters. Traditional chuck-clamping slave manipulators often suffer from discrete motion discontinuities and bulky footprints, while rigid roller-driven systems risk damaging the hydrophilic coatings of the instruments due to high localized stress. This paper presents a novel, compact, and highly integrated friction-wheel-based manipulation mechanism driven by an orthogonal configuration of two high-precision linear motors. The active friction wheel is driven by a rotary motor to provide continuous translation, while its lateral linear motor stage enables sub-micron adjustment of the gap to actively regulate the clamping force. On the opposing side, the passive follower wheel is mounted on a longitudinal linear motor to execute vertical displacement, generating a differential shear force across the instrument’s circumference to achieve radial rotation, thereby simulating the hand-twirling action of a clinical surgeon. To ensure instrument safety, both rollers are encapsulated in a flexible rubber layer, which increases the contact area through elastic deformation to enhance the friction coefficient while preventing localized stress concentrations. Experimental evaluations conducted via an electromagnetic tracking system demonstrate that the mechanism achieves precise motion control. The mean linear motion error remains under 0.35 mm across speeds up to 30 mm/s (with a maximum observed error of 0.39 mm), and the mean rotational error is kept below 3.0° with a maximum error bounded within 3.4°. The proposed architecture successfully decouples clamping, translation, and rotation within a highly compact footprint (215 × 180 × 268 mm), offering a continuous-drive, safe, and robust slave manipulator solution for the next generation of modular vascular robotic platforms.

As a highly disabling neurological disorder, stroke often causes persistent wrist motor dysfunction in patients. Exoskeleton robots can effectively assist in providing efficient rehabilitation training for them. This study addresses the issue that serial-structured wrist rehabilitation exoskeletons incur human-machine axis misalignment during rehabilitation training and cannot match the differentiated rehabilitation demands of stroke patients in varying recovery phases, thus a 3-DOF home-based parallel wrist rehabilitation system with active and passive rehabilitation modes is developed based on the 6-UCU Gough-Stewart platform. The passive rehabilitation mode adopts the preset trajectory training method, while the active rehabilitation mode employs the bilateral rehabilitation training method. Results indicate the passive movement achieves an MAE of 6.481°, RMSE of 7.624°, R2 of 0.924 on average, while the active movement group yields averaged MAE is 1.162°, RMSE is 1.248°, R2 is 0.990. Experimental errors follow a homogeneous distribution without conspicuous extreme deviations, which verify the validity of the control system configured for the proposed parallel home-based wrist rehabilitation system, and the developed device possesses promising potential for clinical application.

This paper presents a fast sparse 3D–2D topology feature-matching framework for intraoperative navigation of vascular interventional robots. Synthetic vascular routes are first rendered into a multi-view dataset for controlled method comparison, and the selected matcher is then transferred to a physical vascular phantom. Representative topology points are extracted from a vascular skeleton model on the 3D side, while vessel structures are detected through U-Net segmentation, skeletonization, and topology-point screening on the 2D side. An affine alternating optimization scheme is formulated by iterating between least-squares affine fitting and Hungarian assignment, and a PCA-guided fast guided-restart strategy is introduced to improve robustness against poor initialization. Synthetic-dataset experiments and phantom-based robustness tests show a favorable trade-off among accuracy, stability, and runtime for interpretable vascular navigation.

Multi-level coronary stenosis detection in X-ray coronary angiography is difficult because adjacent severity grades are separated by subtle lumen changes, while conventional one-stage detectors treat grades as independent classes. We present YOLO-SOD, a severity-aware ordinal decoupling framework that preserves the native YOLO detection objective and injects stenosis-order priors through a continuous-spectrum severity branch, quality-adaptive positive-sample gating, and a lightweight earth mover’s distance (EMD) regularizer on a normalized class distribution derived from classification scores. On CADICA, we merge the scarce p99 and p100 labels into a near-occlusion class and evaluate a six-class detection task under a fixed severity-stratified image-level split. Across four YOLO variants, YOLO-SOD improves mAP@0.5:0.95 by relative gains of +2.09%, +0.50%, +0.77%, and +1.33% on YOLO11n, YOLO11s, YOLO8n, and YOLO26s, respectively. The best paired result is obtained by YOLO26s-SOD, increasing mAP@0.5:0.95 from 58.47±0.64% to 59.25±0.46% with a +1.33% relative gain.

In robot - assisted vascular interventional surgery, the complex anatomical structure and fragility of human blood vessels impose high demands on the operating accuracy and force - control capabilities of interventional instruments. The clamping mechanisms of current robots struggle to replicate the flexibility of human fingers, and their rigid structures may heighten surgical risks. For instance, when a robot grasps a catheter or guide - wire, an excessive clamping force could damage the surface of the tube - wire, causing fragments to enter the blood vessel. Conversely, insufficient clamping force might result in relative slippage between the tube - wire and the robot, compromising the operation's accuracy. Achieving an ideal clamping state that balances safety and effectiveness while preventing instrument damage is one of the key challenges in the field of vascular - interventional surgical robots. To address this issue, this paper presents an electromagnetic flexible clamping mechanism and designs a surgical robot system based on it. The robot can not only meet the axial delivery force and rotational torque requirements of surgery but also achieve adaptive and non - destructive clamping of catheters and guide - wires of various sizes, thus ensuring the safety and accuracy of surgical operations.

With the increasing number of patients with upper limb hemiplegia, robot-assisted rehabilitation has attracted increasing attention. Compared with traditional face-to-face training, telerehabilitation with therapist-in-theloop is an effective alternative. During the telerehabilitation training, the complementary assessment methods are essential, which form the basis for formulating rehabilitation treatment plans. However, at present, most assessments are either subjective evaluations by physicians or objective methods that cannot be combined with robots, which are not suitable for the telerehabilitation training based on upper limb exoskeleton. This paper builds an exoskeleton-assisted upper limb telerehabilitation system leveraging a cloud server. Based on the telerehabilitation system, an assessment method using multi-sensor information fusion is proposed. Experiments were conducted involving passive movements (simulating complete hemiplegia) and active movements (simulating no hemiplegia). Further, data preprocessing and data derivation were carried out. Finally, the collected multi-sensor data were fused through machine learning algorithms to achieve the classification of different rehabilitation states. Although the method proposed in this study is designed for simplified scenarios, it has the potential to develop assessment methods compatible with rehabilitation exoskeletons.

Guidewire perception is an important visual task in robot-assisted vascular intervention because it supports subsequent geometric analysis and safety-oriented assistance. However, guidewire segmentation remains difficult due to the thin, elongated, and low-contrast nature of the target. To address this problem, this paper proposes a multi-scale U-Net method for guidewire segmentation in vascular interventional scenes. The proposed framework enhances a U-Net backbone with a multi-scale bottleneck module and a guidewire-oriented hybrid Dice-BCE loss to improve structural preservation. Experiments are conducted on a laboratory silicone vascular platform dataset containing 324 annotated samples. On the held-out test set, the proposed method achieves a Dice coefficient of 0.7786 ± 0.1962 and an IoU of 0.6644 ± 0.1747. These results indicate that the proposed method can effectively segment the guidewire and provide a useful visual front-end for subsequent geometric analysis in robot-assisted vascular intervention perception.

This paper presents a magnetic sensing-based master-side guidewire pose capture system for master-slave controlled vascular interventional robots. Conventional systems relying on crank mechanisms or contact encoders decouple surgeons from natural tactile feedback and introduce mechanical damping. To address this, the proposed system employs the guidewire as the direct input, leveraging non-contact magnetic sensing for real-time capture of axial displacement and rotational angle without resistance. A motion mapping framework is established based on ball screw and gear transmissions. Experimental results demonstrate that master-side mapping accuracies reach 99.2% (translational) and 98.6% (rotational), while slave-side accuracies achieve 98.3% and 97.8%, with errors within ±0.18mm and rotational errors within ±0.5°. Vascular phantom navigation confirms operability, achieving a 90.0% success rate with a 76.2s mean completion time under complex conditions. Compared to traditional encoders, this system significantly reduces damping and enhances tracking accuracy, substantially improving the fidelity of the surgeon’s operative experience.

Permanent magnet tracking (PMT) is widely used in minimally invasive surgical navigation but is sensitive to soft-magnetic interference from surgical instruments. Existing compensation-based approaches often introduce computational overhead and may affect real-time performance. This paper presents a lightweight anomaly detection and directional warning method for PMT. Singular Value Decomposition (SVD) is used to identify structured magnetic disturbances, and a Residual Center of Mass (CoM) method provides 2D directional feedback. Experiments demonstrate reliable detection and stable directional estimation under dynamic conditions, with sub-millisecond execution time, supporting real-time deployment.

To address the demand for lightweight intention recognition algorithms in home-based portable upper limb rehabilitation robots, this paper presents a lightweight upper limb motion intention recognition scheme based on bimodal fusion of surface electromyography (sEMG) and inertial measurement unit (IMU) signals. Time-domain features are extracted from sEMG and statistical features from IMU, followed by feature-level concatenation. Principal component analysis (PCA) is then applied for dimensionality reduction, and a Support Vector Machine (SVM) is employed to classify five wrist movements. Experiments on the publicly available Ninapro DB5 dataset validate that bimodal fusion enhances recognition accuracy, while PCA effectively compresses the feature space with minimal performance degradation. Key design choices, hyperparameter configurations, and performance boundaries are reported to facilitate practical deployment on resource-constrained portable rehabilitation devices.

Due to the growing incidence of vascular-related conditions, the accurate identification of arterial calcifications has become a critical step in preoperative planning for intravascular interventions. While coronary calcification segmentation methods have been extensively studied, calcification detection in lower limb arteries remains underexplored due to complex anatomical structures and insufficient public datasets. This paper applied a method based on convolutional neural network (CNN) to segment calcified plaques in lower limb CT angiography (CTA) using the nnUNet framework. A custom dataset of 23 annotated lower limb CTA scans was prepared, and three nnUNet variants(2D UNet, 3D fullres UNet, and 3D lowres UNet) were evaluated. In the experiment, the performance of 3D fullres UNet was better than that of 2D UNet and 3D lowres UNet, with a Pseudo Dice of 0.4793 and an IoU of 0.3442. The results suggest that 3D fullres models are more effective in capturing the spatial complexity of lower limb vascular calcifications and hold promise for improving automated vascular assessment in clinical workflows.

In robot-assisted vascular interventions, the surgeons' operating skills have a significant impact on the outcome of the surgery. However, there is limited research on operation recognition and skill learning of surgeons' operating skills. Most of the studies rely on data collected by multiple external sensors, and the data collection process and recognition algorithms are complex and time consuming, which is not conducive to long term development. In this study, we propose a method for multilevel classification of surgeons' operations and a MultiRocket based classification algorithm. This method can recognize the surgeons' surgical operations while relying only on the data collected by a vascular interventional surgery robot. The results of experiment simulating real surgeries demonstrated the effectiveness of our method with a classification accuracy of 92.56%. This method can qualitative analysis surgeons' operating skills, and can be used in interventionalist training, evaluation of surgical skills, and improvement of the intelligence of robotic systems.

With the rising incidence of cardiovascular and cerebrovascular diseases, vascular interventional surgery has become a primary treatment for conditions such as stroke, vascular occlusion, and aneurysms. However, the need for physicians to wear heavy lead aprons and endure prolonged radiation exposure, combined with the high technical demands of the procedure, has led to a declining number of willing practitioners. To address these challenges, vascular interventional surgery robots have emerged, offering effective radiation isolation and improved procedural standardization and precision. Nevertheless, due to the need for remote operation, current robotic systems lack real-time environmental perception, resulting in a loss of embodied intelligence and increased procedural risk. To alleviate the operational burden on doctors and enhance robotic intelligence while reducing dependence on frequent angiographic imaging, this study proposed an autonomous guidewire delivery system based on the curvature variation of interventional instruments. Experimental results demonstrated that the system can effectively detect collisions between the instrument and vessel wall, and promptly halt advancement upon risk identification, thereby improving autonomous safety and clinical adaptability.

The application of wrist exoskeletons in the rehabilitation of stroke and hand dysfunction is becoming increasingly widespread, as they can assist patients in undergoing high-intensity, repetitive, targeted and interactive rehabilitation training. However, the existing wrist exoskeletons cannot effectively conform to the structure of the human wrist and cannot cover the entire physiological movement space. Here, we present a the wrist exoskeleton based on the 6-UCU Stewart for Home-based rehabilitation by analyzing the physiological structure of the human wrist. We conducted simulation experiments on forward kinematics and reverse kinematics for it. The parallel wrist exoskeleton can effectively achieve three wrist movements, flexio n/ extensio n(F /E), radial/ulnar devia tio n(R/U) pronation/supination (P/S) and angles are -80°-80°, -30°-30°, -60°-60° respectively.

Angiography image segmentation is critical for cardiovascular disease diagnosis but faces challenges due to complex vascular structures and costly manual annotation. While deep learning offers solutions, supervised methods require large labeled datasets. Interactive segmentation (IIS) methods, using clicks or scribbles, provide a promising alternative by balancing efficiency and accuracy. However, the performance of existing IIS algorithms on angiography images remains underexplored. This study evaluates representative IIS algorithms on multiple angiography datasets, analyzing their performance characteristics and highlighting domain-specific challenges such as vascular complexity and imaging noise.

Vascular interventional surgery (VIS) robots still face limitations in enhancing treatment quality. The primary challenges come from the accurate morphology control and miniaturization of the distal instruments. To address these issues, a novel 4 degrees of freedom steerable catheter robotic system is developed. The catheter is composed of two segments continua connected in series. By actuating the two segments independently, the distal end of the catheter has both position and wide-range orientation control capabilities. To accurately estimate the continuum deformation, an improved tendon tension propagation model considering the path friction is proposed. Then a continuum morphology estimation model is derived cell by cell based on the chain beam assumption. Based on the estimation model, the kinetostatic model is established and a model-based nonlinear numerical solution method is developed to obtain the inverse solution of the actuation space. Experimental results demonstrate that the improved tendon tension propagation model reduces the control errors in position and orientation by 45.6% and 57.6% respectively compared with the simplified model. Additionally, the dexterity and operability of the robotic system are demonstrated by 4 navigation experiments.

Upper limb hemiplegia is a common functional disorder among stroke patients, significantly affecting their quality of life. To address this issue, robot-assisted upper limb rehabilitation training has emerged as a new therapeutic approach, breaking through time and space limitations of traditional rehabilitation. Based on the above, a home-based dual-mode upper limb rehabilitation system is built, including teleoperation mode based on a cloud server and bilateral mode with fusion of Surface Electromyography (sEMG) and Inertial Measurement Unit (IMU). In the telerehabilitation mode, patients can receive professional guidance and regular training at home, greatly enhancing the accessibility of rehabilitation services. The experiments with the master side in Beijing City (China) and the slave side in three different cities are conducted through a cloud server. The slave side is controlled by the master side, and the contact force is sent back to the master side. In the bilateral mode, the intention of continuous movements across subjects can be accurately predicted via the fusion of sEMG and IMU, improving the naturalness of human-robot interaction. In the subject-independent modeling, the Root Mean Square Error (RMSE) under fusion showed a relative decrease of 15.0329% (p <10-4) compared to IMU data alone, and a significantly greater reduction of 61.9376% (p <10-4) in comparison with sEMG data alone. Robot-assisted upper limb exoskeleton, cloud-based teleoperation and bilateral training based on sEMG and IMU collectively form a new rehabilitation system, representing part of the future rehabilitation trend.

Hand tremor seriously reduce the accuracy and safety of robot-assisted vascular interventional surgery. However, the existing methods have some deficiencies, such as phase delay and distortion of operation signal. Therefore, a more accurate and universal method for tremor signal recognition is needed. This paper proposes a recognition method based on residual neural networks. And we use a Geomagic TouchX to collect tremor signal. The sliding window algorithm is used to process the data and make the dataset. Finally, the validation results indicate that the method can recognize normal operation signal and tremor signal, and ensure that the normal operation signal of the surgeon is not affected.

The scarcity and difficulty in acquiring vascular interventional surgery data have hindered the development of this field. To address this challenge, we propose a novel application of the Deep Digitally Reconstructed Radiographs(DeepDRR) method for synthesizing X-ray images of vascular interventional surgery with guidewires. By merging predefined-shaped guidewires with computed tomography(CT) scan images, composite CT images containing guidewires can be obtained. The DeepDRR model is then used to project these images, producing corresponding X-ray images. Experimental results demonstrate that the synthesized X-ray images with guidewires possess a certain degree of realism and can be generated at a relatively fast speed, achieving 2.69 frames per second (FPS). This study establishes a pipeline for synthesizing vascular interventional images, laying a foundation for future research. This will provide high-quality and diverse data support for the development of vascular interventional surgery robots and promote further advancements in this field.

Robots capable of traversing multiple environments impose stringent design requirements on their propulsion units. In order to add flying capabilities to the existing amphibious spherical robot and enhance its maneuverability while optimizing propulsion strategies, this paper proposes a three-dimension triphibian robot propulsion unit based on electric ducted fan engine. The paper begins by introducing the concept of the robot and its significance. Subsequently, it analyzes the design requirements for the propulsion unit based on the functional upgrade. Following this, an evaluation system is designed to conduct thrust tests and verify the power of the propulsion unit, demonstrating the effectiveness of its design. This lays the groundwork for the comprehensive design and experimentation of the three-dimensional triphibian robot propulsion unit.