The 3-D model and spatial pose of a workpiece are essential for robotic spray painting. However, in multivariety and small-batch painting scenarios, the structural diversity of workpieces makes it difficult to establish a comprehensive model library. Even when prior models are available, the uncertain poses of workpieces during transportation necessitate online modeling and pose calculation. To address these issues, an online modeling method based on skeleton-guided decomposition and model-driven assembly is proposed. First, a skeleton-guided point cloud decomposition method is developed, where the skeleton is extracted from the mapped image of the workpiece and used to construct decomposition-oriented bounding boxes, enabling structured decomposition of the workpiece point cloud. Subsequently, a model-driven feature parameter extraction and regularization method is developed, where directional-consistency-based point cloud segmentation and feature parameter extraction are performed, and connection-structure constraints are incorporated to achieve global correction and alignment of feature parameters. Finally, parametric modeling is applied to generate and assemble the structural units into a complete CAD model of the workpiece. A dual-view line-laser vision system was built for validation. Experimental results demonstrate that the proposed method achieves complex workpiece modeling within 1000 ms on a standard computing platform, including a GPU, while maintaining geometric and pose accuracies sufficient for spray painting applications.

PurposeDue to the increasing automation requirements for robotic welding, automatic seam extraction has become a research hotspot. The welding objects in this paper are a type of multi-variety gate-shaped workpieces, the shapes of workpieces and the distribution of seams are uncertain, the current seam extraction methods are difficult to deal with. For this challenge, this paper aims to propose an automatic seam extraction method based on the improved DLP vision system.Design/methodology/approachIn this paper, an improved DLP (Digital Light Processing) structured light vision system is first set up, which integrates 2D seam region detection results into the 3D point cloud reconstruction process. By this solution, several independent point clouds of different seam regions can be generated, and the type of seam on each point cloud can also be obtained. Subsequently, for different types of welding seams, the corresponding seam extraction algorithms are designed by analyzing the local structural characteristics.FindingsThe experimental results demonstrate the proposed improved DLP structured light vision system can automatically, accurately and efficiently complete the seam extraction for the workpieces with multiple seams of different types in the field of view.Originality/valueThe proposed method enhances robotic welding automation for workpieces with multiple seams of different types, overcoming the limitations of the current methods.

Planar butt seams are common in industrial sites, while the current seam extraction schemes are difficult to cope with butt seams with varying widths. Based on DLP (Digital Light Processing) vision technology, this paper proposes an automatic seam extraction method to solve this challenge. Firstly, we calculate different-granularity contour points based on the Alpha Shape algorithm, then extract seam feature points by calculating specific distance information. Secondly, we construct a type of 2D reference coordinate frame by analyzing the structural characteristics of the workpiece, then conduct a dimensionality reduction transformation on the feature points. Finally, we design an algorithm for curve fitting and discrete interpolation, which can generate the ordered interpolation points for welding. The experimental results demonstrate that the proposed seam extraction method can accurately, robustly, and efficiently extract the butt seams with varying widths.

This paper proposes a method for identifying optical component misalignments based on relative ray vector (RRV) observation. Firstly, the RRV is calculated from the gradient direction of the wavefront. Through ray tracing, the model between the RRV and misaligned optical systems can be established. The optimization objective is to minimize the angle error between the theoretical and observed RRV. Using the error backpropagation method, the misalignment parameters of each lens surface can be optimized layer by layer. And then, calculate the lens’s misalignment parameters. Simulation and experiment verified the high performance of this method in dealing with large initial misalignment.

This study systematically investigates the effects of thermally orthotropic wall material on the combustion performance and thermal management of a catalytic microchannel reactor using two-dimensional numerical simulations. Under a constant inlet condition of preheated propane/air mixture at 569 K and an initial microreactor temperature of 300 K, the axial ( λ xx ) and transverse ( λ yy ) thermal conductivities are varied independently while holding the other constant. Numerical results demonstrate that axial thermal conductivity significantly influences the combustion behavior. Increasing λ xx from 5 to 20 W/(m·K) reduces the peak temperature from approximately 1800 K to 1700 K and delays ignition due to enhanced axial heat recirculation and upstream preheating. In contrast, variations in transverse conductivity show negligible effects on maximum temperature, ignition time, and HTR (heterogeneous reaction) contribution. Wall temperature profiles further confirm that higher λ xx broadens and flattens the axial temperature distribution, whereas λ yy has minimal impact. These findings highlight the critical role of axial thermal conductivity in regulating thermal stability and ignition initiation, providing key insights for optimizing microreactor design to prevent thermal runaway while maintaining combustion performance.

Multi-modal three-dimensional (3D) roadside object detection is a challenging yet critical topic for Vehicle-Infrastructure Cooperated Autonomous Driving (VICAD). Recently, the Birds-Eye View (BEV) framework has emerged as a promising solution. However, current approaches that process BEV features with isotropic convolutions fail to account for the error characteristics of sensors, leading to inefficiency and instability in disturbed scenarios. This paper addresses this issue by developing the Sensor Characteristics Adaptive Bird's Eye View (SCA-BEV), a simple yet effective roadside 3D object detection framework based on improved deformable convolution and anisotropic attention mechanism. Specifically, we designed the error Dispersion Adaptive Convolution (DAC) to enhance the network's anti-interference ability. Moreover, a novel axial attention mechanism was proposed to improve multi-modality fusion. Additionally, we introduce a knowledge distillation framework to boost the performance in camera-only scenarios. The proposed method was validated on the DAIR-V2X dataset and compared against state-of-the-art networks. The experimental results demonstrate that our method achieves higher detection accuracy with remarkable real-time performance in disturbed scenarios, suggesting a broad application prospect in intelligent roadside systems.

Micro-combustion stability is critically influenced by wall material properties. This study investigates the impact of wall density ( ρ w ) and specific heat capacity ( c w ) on the ignition and combustion characteristics of a propane/air mixture within a micro-channel via transient numerical simulation. The microreactor, with an initial temperature of 300 K, is fed with a preheated mixture at 569 K. The results demonstrate that increasing either wall density or specific heat capacity significantly delays ignition, exhibiting a pronounced linear trend. The ignition time increased from 17 s to 42 s as ρ w escalates from 4000 to 10,000 kg/m 3 , and from 17 s to 132 s as c w rises from 250 to 2000 J/(kg·K). However, these properties are found to have a negligible effect on the maximum combustion temperature and the HTR (heterogeneous reaction) contribution. Furthermore, when the volumetric heat capacity ( ρ w c w ) is held constant, variations in ρ w and c w show no influence on ignition time, peak temperature, and HTR contribution. The transient evolution of the wall temperature distribution confirms that ignition consistently originates near the channel inlet. This work underscores the pivotal role of wall material properties in governing ignition dynamics and provides essential insights for the thermal design and material selection of micro-combustors.

Extrinsic calibration between cameras and LiDAR (Light Detection and Ranging) is a fundamental step for numerous autonomous driving applications, such as 3D object detection, lane detection, and trajectory planning. However, conventional calibration methods are laborious and require dedicated data collection. Although recent research has shown the potential of learning-based solutions for end-to-end use, existing methods tend to focus on feature matching rather than geometric constraints, leading to inefficiency and instability in traffic scenarios with various disturbances. In this paper, we introduce an improved deep-learning-based joint calibration framework for LiDAR-camera systems, termed the Complementary Calibration Network (Co-CalibNet). The main contribution of our work lies in proposing a novel dual-channel geometric supervision calibration framework that integrates both depth and height supervision to achieve robust extrinsic calibration. Additionally, we introduce an Attention-based Fusion Module (AFM) for efficient feature fusion. Furthermore, we incorporate time compensation and iterative calibration techniques to further enhance the robustness of the algorithm in handling initial alignment errors. Evaluations on the DAIR-V2X and KITTI datasets demonstrate that Co-CalibNet achieves state-of-the-art calibration performance while exhibiting greater robustness to initial misalignment. Since it requires no targets or human effort, Co-CalibNet can be seamlessly integrated into any LiDAR-camera architecture, suggesting significant value and broad application prospects in autonomous driving systems.

Dielectric elastomer actuators (DEAs) are increasingly recognized for their potential in robotic applications due to their ability to undergo significant deformation when subjected to an electric field. However, they are often limited by their low output power, which can make their integration into dynamic systems like hopping robots particularly challenging. This research optimizes the performance by introducing a cone DEA with a novel type of semi-diamond preload mechanism. This type of preload mechanism can meet the requirements of a negative-stiffness preload and a light weight. According to the experiments, the DEA can provide 3.62 mW power and its mass is only about 17.5 g. In order to drive hopping robots based on a cone DEA, this research introduces an energy accumulation mechanism coupled with a constant-torque cam for a hopping robot. The hopping robot weighs approximately 30.3 g and stands 10 cm tall in its upright position. Its energy accumulation mechanism involves a gear and cam transmission system, which is the key to store and release energy efficiently. The primary components of this mechanism include a torsion spring that stores mechanical energy when twisted, a constant-torque actuation cam that ensures the consistent application of torque during the energy storage phase, and a conical DEA that acts as an actuator. When the conical DEA is activated, it pushes a one-way clutch to the rocker, rotating the gear and cam mechanism and subsequently twisting the torsion spring to store energy. Upon release, the stored energy in the torsion spring is rapidly converted into kinetic energy, propelling the robot into the air. The experiments reveal that the designed DEA can drive the hopping robot by using the energy storage mechanism. Its hopping height is related to the pre-compression angle of the torsion spring. The DEA can drive the rigid hopping mechanism, and the maximum hopping height of the robot is up to 2.5 times its height. DEA hopping robots have obvious advantages, such as easy control, quietness and safety.

This study aims to accurately predict the nonlinear characteristics of transmission efficiency of cycloid reducers under different operating conditions. Firstly, equivalent modeling of the multi-source errors (MSEs) in the designed cycloid reducer is conducted. Force analysis algorithms considering MSEs are proposed for the cycloid drive mechanism, the output mechanism, and the bearings. Secondly, mathematical models are established for the load-dependent power losses, while an equivalent test is used for modeling load-independent power losses. Subsequently, an improved transmission efficiency prediction (TEP) method for cycloid reducers is proposed, which is then applied to the performance prediction of a prototype under different operating conditions. The advantages of the improved TEP method over the conventional method are discussed, and the influences of MSEs and load-independent power losses on the nonlinear characteristics of transmission efficiency are summarized. Finally, tests are conducted for the reducer prototype, and the test results are found to be in good agreement with the results obtained by the proposed TEP method. The main contribution of this study is to establish a solid algorithmic and modeling foundation for the optimal design of nonlinear transmission efficiency in cycloid reducers and provide reliable guidance for their engineering applications.

Roadside 3-D object detection allows for a drastic expansion of the visibility range and a reduction in occlusions for autonomous vehicles. Recent approaches are based on bird's-eye view (BEV) fusion, which unifies multimodal features in the shared BEV representation space. However, the camera-to-BEV projection throws away the geometric information of camera features, hindering the effectiveness of such methods. Besides, depth-based camera lifting results in inefficiency and instability in disrupted roadside scenarios. To address these challenges, this article introduces a novel 3-D object detection framework based on height-aware scene reconstruction, dubbed HSRDet. Specifically, we leverage height-aware 3-D reconstruction to ensure geometric consistency in BEV feature mapping and employ a fast camera-to-BEV transformation based on feature distillation to boost efficiency without compromising performance. In addition, we integrate a novel data augmentation method, namely, View Shake (VS), to further improve the performance of our model. Extensive experiments on the DAIR-V2X dataset demonstrate that HSRDet not only achieves state-of-the-art detection accuracy but also exhibits strong robustness in disturbance scenarios. Further experiments on the intelligent roadside units (RSUs) have revealed that our method runs stably at 11.8 frames/s on a RTX 3090 Ti GPU, thus promising vast engineering application prospects.

As users' expectations for precision cycloid reducer positioning accuracy rise, accurate positioning accuracy prediction of products before assembly and design solutions to improve product qualification rates are very important. Considering the actual manufacturing conditions of components, a new multi-source error (MSE) equivalence modeling assumption and pin gear's MSE measurement method are proposed, and the rationality of the assumption is verified through experiments. Based on the geometric characteristics of each component's MSEs, a new analytical model of bi-directional drive transmission error (BDDTE) is established, which organically unifies the prediction of transmission error (TE) and lost motion (LM). The influence of the output mechanism on the positioning accuracy of the entire reducer is found to be non-negligible through Monte Carlo simulation; meanwhile, the general tolerance allocation scheme makes it difficult to ensure that each prototype meets the performance requirements. For this reason, a new design solution for the positioning accuracy of the reducer with the pin gear and output mechanism as the optimization targets is proposed. Finally, the prototype test results show that the BDDTE model based on the new assumption has a higher accuracy in predicting positioning accuracy compared with the model based on the conventional assumptions. Meanwhile, the effectiveness of the proposed improvement design solution for positioning accuracy improvement is verified. The BDDTE model reveals the intrinsic connection between the TE and LM of the cycloid reducer, and the improvement solution has guiding effects on improving the product qualification rate.

3D Local feature description is an essential work in 3D computer vision since a lot of downstream techniques rely on point-to-point correspondences. Most of the existing descriptors perform the description on local surfaces from one single aspect, which inherently results in limited performance. So this paper first proposes a real-valued 3D local feature descriptor named multi-feature fusion histogram (MFFH), which combines five different types of well-designed geometric features to achieve comprehensive descriptions for local surfaces. In addition, to be available for platforms with limited computing and storage resources, we conduct a seamless extension of MFFH to its binary representation B-MFFH by three kinds of directed binarization methods toward different real-valued features. Through extensive evaluation experiments on the benchmark datasets, we prove the superiorities of the proposed MFFH and B-MFFH concerning comprehensive performance. Lastly, the practicability of the proposed MFFH and B-MFFH descriptors is visually demonstrated by the point cloud registration experiments.

3D Local feature description is an active and fundamental task in 3D computer vision. However, most of the existing descriptors fail to simultaneously achieve satisfactory performance among descriptiveness, robustness, efficiency, and compactness. To address these limitations, we first propose a real-valued descriptor named Rotational Voxels Statistics Histogram (RoVo), which exploits the novel 3D multi-pose processing mechanism proposed in this paper to calculate the 3D voxel density distribution in different 3D poses. Moreover, through well-designed binary encoding algorithms, we conduct the seamless extension of the real-valued RoVo descriptor to three binary representations that have different performance characteristics. Extensive evaluation experiments validate the superiority of the real-valued and three binary RoVo descriptors concerning descriptiveness, robustness, and efficiency. Furthermore, the three binary RoVo descriptors extend the performance of high compactness. Lastly, we perform the experiments of 3D scene registration and 3D object recognition to intuitively present the effectiveness of the four proposed RoVo descriptors.

Currently, most seam extraction methods focus on some simple application scenarios. For structured workpieces with medium-thick plates, which have unknown and various structures, the current methods are difficult to deal with. Based on DLP (Digital Light Processing) vision, this paper proposes a novel multi-seam extraction method for structured workpieces. Firstly, the DLP vision system is constructed to achieve single-view data acquisition for workpieces. Next, we propose an improved registration method to fuse point clouds collected from multiple perspectives. Finally, the positions and poses of seams are solved by the designed extraction algorithm. The experimental results demonstrate the proposed multi-seam extraction method can accurately extract seams in a single perspective, and achieve coarse extraction for seams distributed in multiple perspectives, which can be further provided for robot trajectory planning and laser sensing to achieve fully automatic welding.

The computer vision-based roadside occupation surveillance system is a key infrastructure component of Cooperative Intelligent Transport Systems. However, traffic images captured under low-light conditions suffer from low visibility and unexpected noise. Despite the great progress achieved in recent years, the existing night image enhancement algorithms often suffer from color deviation, ghosting, and overexposure problems in practical traffic applications. Thus, we present a novel night color image enhancement approach to overcome this issue by combining multi-sensor fusion and pseudo-multi-exposure fusion techniques. Unlike the traditional exposure adjustment-based approaches, we performed a novel bidirectional region segmentation-based inverse tone mapping operator to generate pseudo-multi-exposure sequences from day and night image pairs. Meanwhile, to solve the problem that moving objects are diluted after fusion, a partial differential equation (PDE)-based luminance stretching is applied to the moving areas to guarantee that the enhanced image always highlights the traffic targets. Instead of image feature-based methods for moving object detection, we generate more accurate moving regions by fusing data from the radar and camera sensors. Finally, a pyramid-based fusion method with an improved weight function is conducted to generate high-quality traffic images. The proposed method and five state-of-the-art methods are evaluated on randomly selected images from the Rope-3D database and nighttime images captured by an Intelligent Roadside Surveillance System. The experimental results demonstrate that our method has significant advantages in enhancing details and making colors more natural for human observation.

The reliability of the prediction model and test scheme for the positioning accuracy of the cycloid reducer is very important. This paper proposes an efficient bidirectional drive tooth contact analysis (BDTCA) model that can simultaneously analyze the forward transmission error (TE), the reverse TE, and the global lost motion (GLM) of a cycloid drive, considering a variety of manufacturing and assembly errors (including pin assembly error). The results of the BDTCA case studies show that the sensitivity of the pin radius error to the GLM considering pin assembly error is twice that of the sensitivity without considering pin assembly error. Therefore, the influence of pin assembly error on the GLM cannot be ignored. The equivalent error model for BDTCA is established based on Latin hypercube sampling and template measurement, and the positioning accuracy of 50,000 reducer virtual prototypes is predicted. According to the positioning accuracy requirements, the tolerance allocation is optimized by using the discrete particle swarm optimization algorithm, and the qualified rate after optimization is significantly improved. A bidirectional drive test (BDT) scheme is proposed according to the BDTCA model. By comparing with the traditional hysteresis curve test, the superiority of the BDT scheme and the rationality of the tolerance allocation optimization are proved. The sensitivity of the pin radius error to the GLM is obtained through a comparison BDT, which verifies the correctness of the BDTCA model considering pin assembly error. This BDTCA model provides more reliable theoretical guidance for the design and manufacture of cycloid reducers.

3D local feature description is now at the core of many 3D vision technologies. However, most of the existing 3D feature descriptors can't strike a balance among descriptiveness, robustness, compactness, and efficiency. To overcome the challenges, we propose a real-valued 3D local feature descriptor named Geometric Feature Statistics Histogram (GFSH) and its binary extension descriptor named B-GFSH. A GFSH descriptor first constructs an improved-weighted covariance matrix to solve a stable and reliable Local Reference Frame (LRF), and then achieves a comprehensive description of the 3D local surface by performing statistics on multiple geometric distribution features, namely voxel density, voxel centroid, and projection density. A particular trait of our GFSH descriptor is its seamless extension to the binary representation to reduce storage consumption and accelerate feature matching. For each sub-feature of GFSH, B-GFSH respectively adopts the corresponding binarization strategy, i.e., improved Gray code quantization, thresholding based on coordinates, and neighbor comparison. Extensive experiments on six public datasets prove that both GFSH and B-GFSH have high descriptiveness, strong robustness, and fast real-time performance. In addition, B-GFSH further has the characteristics of fast matching speed, low memory footprint, and high compactness. Finally, we conduct 3D scene registration and 3D object recognition experiments to visually demonstrate the actual effectiveness of GFSH and B-GFSH.

设计一种新型结构的连续型机器人.该机器人以螺旋弹簧为主干,采用线驱动的方式,可实现平面内单自由度弯曲.通过建立机器人运动学模型,得到驱动线位移-柔性关节弯曲角度的对应关系.提出一种通过改变驱动线张力实现连续型机器人变刚度的方案,对柔性关节分别建立力学模型和Adams模型,进行理论和仿真分析,通过实验对所提出的运动学模型与变刚度方案进行了验证.