Police robots are used to assist police officers in performing tasks in complex environments, so as to improve the efficiency of law enforcement, ensure the safety of police officers and maintain social stability. With the rapid development of science and technology, police robots are widely used in the field of public security, such as alarm reception, patrol, explosive disposal, reconnaissance and so on. However, police robots still have the problem of analysis deviation in the process of receiving the alarm, which leads to the low efficiency of police dispatch. This study aims to enhance the police alarm automatic analysis ability of the police robots to assist in the dispatch of police. In this paper, we propose a novel method (FSTC-LLM) for sample augmentation based on large language model and noise reduction. The experimental evaluations are carried out on the alarm data set and the THUC News data set. The results show that the proposed FSTC-LLM has excellent performance in few shot text augmentation tasks, and can assist police robots to complete the task of automatic analysis of alarm with high quality, which is of great significance to enhance public security.

In order to enhance the efficiency of visual inspection and effectively carry out 3D visual coverage tasks, this paper focuses on the 3D view planning problem concerning the visual coverage of an airplane’s surface using unmanned aerial vehicles (UAV). Our objective is to attain a sufficiently high coverage rate with the least number of viewpoints. The contributions of this work are enumerated as follows. Firstly, the 3D model of the target aircraft is spatially extended in accordance with the depth range of the camera mounted on the drone, thereby confining the sampling range of 3D viewpoints. Next, a candidate set of viewpoints is generated through random sampling and the probabilistic potential field technique. Subsequently, we propose a novel hyper-heuristic algorithm. In this algorithm, a genetic algorithm serves as a high-level heuristic strategy, in tandem with multiple low-level heuristic operators devised for combinatorial optimization. This not only augments the capacity to seek the global optimal solution but also expedites the convergence rate, aiming to ascertain the optimal subset of viewpoints. Moreover, we devise a new fitness function for appraising candidate solution vectors in the set covering problem (SCP), strengthening the evolutionary guidance for genetic algorithms. Eventually, experimental findings on the simulated and real airplanes corroborate the efficacy of the proposed method, i.e., it markedly diminishes the requisite number of viewpoints and augments inspection efficiency.

Humanoid robot joints require real-time torque detection to provide accurate force feedback information for the control system. To meet the measurement requirements and realize the miniaturization of the sensor, a torque sensor based on the magnetoelastic effect is developed, utilizing planar spiral coils as detection probes. In this work, a planar spiral coil mutual inductance calculation model is established to solve the mutual inductance coefficient, and the mechanical structure and circuit design of the sensor are completed. Finally, a torque loading platform is built to perform calibration experiments, and the hysteresis model is improved to compensate for the hysteresis phenomenon. The calibration results indicate that the sensor shows excellent loaded nonlinearity of 3.08%F.S., unloaded nonlinearity of 2.71%F.S., loaded repeatability of 2.48%F.S., unloaded repeatability of 1.89%F.S. and hysteresis of 1.9%F.S., at a compact probe size of 13.8×9.9×1.8 mm.

In the field of hexapod robot control, the application of central pattern generators (CPG) and deep reinforcement learning (DRL) is becoming increasingly common. Compared to traditional control methods that rely on dynamic models, both the CPG and the end-to-end DRL approaches significantly simplify the complexity of designing control models. However, relying solely on DRL for control also has its drawbacks, such as slow convergence speed and low exploration efficiency. Moreover, although the CPG can produce rhythmic gaits, its control strategy is relatively singular, limiting the robot’s ability to adapt to complex terrains. To overcome these limitations, this study proposes a three-layer DRL control architecture. The high-level reinforcement learning controller is responsible for learning the parameters of the middle-level CPG and the low-level mapping functions, while the middle and low level controllers coordinate the joint movements within and between legs. By integrating the learning capabilities of DRL with the gait generation characteristics of CPG, this method significantly enhances the stability and adaptability of hexapod robots in complex terrains. Experimental results show that, compared to pure DRL approaches, this method significantly improves learning efficiency and control performance, when dealing with complex terrains, it considerably enhances the robot’s stability and adaptability compared to pure CPG control.

Humans can quickly perform adaptive grasping of soft objects by using visual perception and judgment of the grasping angle, which helps prevent the objects from sliding or deforming excessively. However, this easy task remains a challenge for robots. The grasping states of soft objects can be categorized into four types: sliding, appropriate, excessive and extreme. Effective recognition of different states is crucial for achieving adaptive grasping of soft objects. To address this problem, a novel visual-curvature fusion network based on YOLOv8 (VCFN-YOLOv8) is proposed to evaluate the grasping state of various soft objects. In this framework, the robotic arm equipped with the wrist camera and the curvature sensor is established to perform generalization grasping and lifting experiments on 11 different objects. Meanwhile, the dataset is built for training and testing the proposed method. The results show a classification accuracy of 99.51% on four different grasping states. A series of grasping evaluation experiments is conducted based on the proposed framework, along with tests for the model’s generality. The experiment results demonstrate that VCFN-YOLOv8 is accurate and efficient in evaluating the grasping state of soft objects and shows a certain degree of generalization for non-soft objects. It can be widely applied in fields such as automatic control, adaptive grasping and surgical robot.

In cerebrovascular interventional surgery, spatial position prediction navigation (SPPN) provides 3D spatial information of the vascular lumen, reducing the spatial dimension loss from digital subtraction angiography (DSA) and improving surgical precision. However, it is limited in its adaptability to complex vascular environments and prone to error accumulation. To address these issues, we propose spatial position prediction-based multimodal navigation (SPPMN), integrating minimal intraoperative X-ray images to enhance SPPN accuracy. In the first phase, a feature-weighted dynamic time warping (FDTW)-based branch matching algorithm is introduced for 3D topological positioning under non-registered conditions, with a dynamic location repositioning module for real-time corrections. In the second phase, an occlusion correction module, based on the elastic potential energy of the instrument tip, dynamically adjusts the tip’s angle to achieve low-projection occlusion control. Experimental validation using a high-precision electromagnetic tracking system (EMTS) on a 3D vascular model shows that the proposed method achieves an average 3D positioning accuracy of 9.36 mm in intracranial vascular regions, with a 78% reduction in radiation exposure, significantly enhancing both precision and safety in interventional surgeries.

Surgical image segmentation serves as the foundation for laparoscopic surgical navigation technology. The indistinct local features of biological tissues in laparoscopic image pose challenges for image segmentation. To address this issue, we develop an image segmentation network tailored for laparoscopic surgery. Firstly, we introduce the Mixed Attention Enhancement (MAE) module that sequentially conducts the Channel Attention Enhancement (CAE) module and the Global Feature Enhancement (GFE) module linked in series. The CAE module enhances the network’s perception of prominent channels, allowing feature maps to exhibit clear local features. The GFE module is capable of extracting global features from both the height and width dimensions of images and integrating them into three-dimensional features. This enhancement improves the network’s ability to capture global features, thereby facilitating the inference of regions with indistinct local features. Secondly, we propose the Multi-scale Feature Fusion (MFF) module. This module expands the feature map into various scales, further enlarging the network’s receptive field and enhancing perception of features at multiple scales. In addition, we tested the proposed network on the EndoVis 2018 and a human minimally invasive liver resection image segmentation dataset, comparing it against six other advanced image segmentation networks. The comparative test results demonstrate that the proposed network achieves the most advanced performance on both datasets, proving its potential in improving surgical image segmentation outcome. The codes of MAMNet are available at: https://github.com/Pang1234567/MAMNet.

Optimal path planning involves finding a feasible state sequence between a start and a goal that optimizes an objective. This process relies on heuristic functions to guide the search direction. While a robust function can improve search efficiency and solution quality, current methods often overlook available environmental data and simplify the function structure due to the complexity of information relationships. This study introduces Genetic Informed Trees (GIT*), which improves upon Effort Informed Trees (EIT*) by integrating a wider array of environmental data, such as repulsive forces from obstacles and the dynamic importance of vertices, to refine heuristic functions for better guidance. Furthermore, we integrated reinforced genetic programming (RGP), which combines genetic programming with reward system feedback to mutate genotype-generative heuristic functions for GIT*. RGP leverages a multitude of data types, thereby improving computational efficiency and solution quality within a set timeframe. Comparative analyses demonstrate that GIT* surpasses existing single-query, sampling-based planners in problems ranging from R⁴ to R¹⁶ to and was tested on a real-world mobile manipulation task. A video showcasing our experimental results is available at https://youtu.be/URjXbc_BiYg.

The hyper-redundant manipulator (HRM) can explore narrow and curved pipelines by leveraging its high flexibility and redundancy. However, planning collision-free motion trajectories for HRMs in confined environments remains a significant challenge. To address this issue, a pipeline inspection approach that combines nonlinear model predictive control (NMPC) with the snake-inspired crawling algorithm(SCA) is proposed. The approach consists of three processes: insertion, inspection, and exit. The insertion and exit processes utilize the SCA, inspired by snake motion, to significantly reduce path planning time. The inspection process employs NMPC to generate collision-free motion. The prototype HRM is developed, and inspection experiments are conducted in various complex pipeline scenarios to validate the effectiveness and feasibility of the proposed method. Experimental results demonstrate that the approach effectively minimizes the computational cost of path planning, offering a practical solution for HRM applications in pipeline inspection.

The human hand, essential for performing daily tasks and facilitating social interaction, is indispensable to everyday life. Millions worldwide experience varying levels of amputation, profoundly affecting their physical, emotional, and psychological well-being, limiting independence, and reducing quality of life. Myoelectric prosthetics, the most advanced active prosthetic hands, use surface electromyography (sEMG) signals and pattern recognition to translate user intentions into control signals. Despite these advancements, high rejection rates persist due to the non-stationarity of sEMG signals, leading to inconsistent and often frustrating user experiences. As a result, clinical and academic research has increasingly focused on improving myoelectric hand gesture recognition under real-world conditions to reduce rejection rates and enhance user acceptance of myoelectric prostheses. Given the vast and diverse range of methods applied in previous research, this survey aims to systematically highlight key studies and provide an overview of the field’s current achievements. Furthermore, research on machine learning for myoelectric hand gesture recognition has been largely influenced by unrelated fields of computer science, such as computer vision and natural language processing. However, myoelectric hand gesture recognition presents unique challenges, particularly severe and unpredictable covariate shifts in sEMG signals, which require specialized approaches. To address these challenges, we propose a new taxonomy for categorizing machine learning models based on feature extraction methods and decision boundary strategies. Additionally, this paper highlights the need for benchmark datasets that accurately reflect real-world conditions and emphasizes the importance of re-evaluating real-time performance, particularly when using long temporal contextual windows. This study concludes with research challenges and future research directions to enhance the accuracy of myoelectric hand gesture recognition using machine learning techniques.

This paper presents a framework for applying origami-kirigami techniques to design kirigami analogies for remote center-of-motion (RCM) mechanisms, specifically targeting minimally invasive keyhole procedures. The proposed kirigami RCM analogs emulate the motions of existing bar-linkage RCMs, offering advantages in deployability, transportability, and simplified fabrication. A workflow is introduced to transition from initial crease patterns to functional kirigami equivalents, demonstrating the potential for customizability and scalability. Furthermore, a proof-of-concept kirigami RCM under magnetic actuation is presented, showcasing its ability to reduce structural profile during transportation and improve device deployment. Three representative parallelogram-based RCM mechanisms: coupled dual parallelogram, back-drivable, and triple parallelogram, are transformed into kirigami analogs, highlighting the versatility of the design approach. The discussion includes computational modeling, fabrication considerations, and potential applications in MIS robots. This work contributes to the development of compact, deployable, and cost-effective RCM mechanisms for robotic keyhole procedures. This approach can also further facilitate the education of RCM mechanisms and the hands-on demonstration of small-scale RCM concepts.

Modern military drills and conventional training, performed under all-weather conditions, impose exacting challenges on soldiers. This has motivated the development of exoskeleton robot systems, leveraging advanced technology and material innovation. These systems have demonstrated their effectiveness at assisting movement, enhancing protection, promoting rehabilitation, and providing comprehensive support to soldiers. This groundbreaking technology not only reduces a soldier’s physical exertion significantly but also effectively diminishes the risk of injury during training, infusing new vitality into the enhancement of military capabilities. Different types of exoskeleton robots differ in their focus. Lower-limb exoskeleton robots are designed to increase the soldier’s endurance. Upper-limb exoskeleton robots enhance strength. This paper provides a detailed explanation of the key technologies of various types of exoskeleton robots, covering their mechanical design, electromechanical transmission structures, sensors, and actuation methods. It also explores the diverse application scenarios of exoskeleton robots in the military field, systematically introducing their development trajectory, milestone achievements, and the cutting-edge technologies currently employed, as well as the challenges faced. The conclusion offers a prospective discussion of future development pathways, anticipating the broad prospects for exoskeleton robots in the military domain.

Wheeled-legged robots integrate the mobility efficiency of wheeled platforms with the terrain adaptability of legged robots, making them ideal for complex, unstructured environments. However, balancing high payload capacity with agile multimodal locomotion remains a major challenge. This paper presents a field study conducted in the high-altitude region of Golmud, Qinghai, with elevations ranging from 2800 m to 4000 m. We evaluate three wheeled-legged robot platforms of different scales on diverse terrains including Gobi, desert, grassland, and wetlands. Our experiments demonstrate the robot’s robust locomotion performance across multimodal tasks such as obstacle crossing, slope climbing, and terrain classification. Moreover, we validate the performance of autonomous perception systems, including real-time localization and 3D mapping, under harsh plateau conditions. The results provide valuable insights into the deployment of wheeled-legged robots in extreme natural environments and lay a solid foundation for future applications in inspection, rescue, and transport missions in high-altitude regions.