Concrete construction intelligent auxiliary robot path planning method and system based on beidou positioning

Abstract

The application provides a concrete construction intelligent auxiliary robot path planning method and system based on Beidou positioning, and belongs to the technical field of robot path planning. In the scheme, the robot pose information and construction environment information are collected through a Beidou high-precision positioning receiver, an inertial navigation unit (IMU), a laser radar and a depth camera; the multi-source heterogeneous data are processed and fused to generate high-precision pose information and an environment map; based on the fused data, the shape features of the construction area are analyzed, the construction area boundary and special structures in the construction area are identified; according to the shape features of the construction area and the concrete characteristics, a global path is generated, and the robot performs concrete construction according to the global path. In this way, the positioning information provided by the Beidou high-precision positioning receiver is used, and the problems that may occur in the complex construction environment of the Beidou signal are effectively compensated by combining the IMU, so that stable, continuous and high-precision positioning of the robot in the complex shape construction area is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot path planning, in particular to a concrete construction intelligent auxiliary robot path planning method and system based on Beidou positioning. BACKGROUND

[0002] With the accelerated advancement of building industrialization and intelligentization, concrete construction automation, as an important means to improve construction quality and efficiency, has gradually become the focus of the industry. However, due to the complex and variable construction environment and the possible positioning deviation and unreasonable path planning in the construction process, traditional automated construction equipment is difficult to achieve ideal results in concrete precise paving and boundary processing.

[0003] The path planning method of existing automated construction equipment usually relies on simple linear planning, preset mode or laser-guided method, however, in complex scenarios (such as irregular construction areas or multi-obstacle environments), these methods often do not work well. At the same time, with the increasing demand of intelligent buildings for construction precision and efficiency, efficient path planning has become a major challenge.

[0004] Therefore, it is necessary to provide an improved technical solution for the existing concrete construction path planning algorithm which has poor adaptability and is difficult to efficiently cover complex-shaped construction areas. SUMMARY

[0005] The purpose of the present application is to provide a concrete construction intelligent auxiliary robot path planning method and system based on Beidou positioning, to solve or alleviate the technical problems of insufficient positioning accuracy, limited environmental perception ability, simple path planning, imprecise boundary processing and insufficient construction quality monitoring in the prior art.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The present application provides a concrete construction intelligent auxiliary robot path planning method based on Beidou positioning, comprising:

[0008] Step S1, data acquisition: collecting robot pose information and construction environment information through a Beidou high-precision positioning receiver, an inertial navigation unit, a laser radar and a depth camera, forming multi-source heterogeneous data;

[0009] Step S2, multi-source heterogeneous data adaptive fusion: processing and fusing the multi-source heterogeneous data to generate high-precision pose information and an environment map;

[0010] Step S3, environment analysis: based on the fused data, analyzing the shape characteristics of the construction area, identifying the construction area boundary and special structures in the construction area;

[0011] Step S4, concrete property-aware global path planning: generating a global path according to the shape features of the construction area and the concrete properties, and performing concrete construction by the robot according to the global path.

[0012] Since the laser radar, depth camera, IMU and Beidou data belong to different modal, frequency and accuracy data, through adaptive fusion algorithm (such as Kalman filter), more accurate and robust pose estimation and environment map can be extracted, and three-dimensional point cloud map or grid map of the construction environment can be constructed in real time, so as to improve the perception ability and adaptability of the robot to the construction environment. Shape feature analysis of the construction area can provide prior constraint conditions for path planning, support the path planning system to avoid dangerous areas and adapt to complex construction layout, and improve the construction efficiency and safety; different concrete mix, fluidity, initial setting time and other properties will affect the construction method and robot motion strategy, in the embodiment, the path planning not only considers the spatial accessibility, but also dynamically adjusts the material properties to improve the consistency and automation level of the concrete construction quality.

[0013] In combination with the first aspect, in some possible implementation manners, step S2 comprises:

[0014] Step S21, space-time alignment: time alignment and space alignment of multi-source heterogeneous data;

[0015] Step S22, pose fusion: fusing the positioning data obtained by the Beidou high-precision positioning receiver and the inertial navigation data by using an improved extended Kalman filter (EKF) algorithm;

[0016] Step S23, environment map fusion: registering and fusing the point cloud data of the laser radar and the depth camera by using a deep learning point cloud registration algorithm to construct a high-precision environment map;

[0017] Step S24, surface feature extraction: extracting the flatness, roughness, humidity and hardening degree features of the concrete surface, and using the extracted surface features to represent the properties of the concrete.

[0018] In combination with the first aspect, in some possible implementation manners, in step S22, the improved EKF algorithm comprises:

[0019] Adaptive noise estimation: dynamically adjusting the measurement noise covariance matrix R and the process noise covariance matrix Q according to the Beidou signal quality and the IMU data stability;

[0020] Nonlinear state constraint: based on the kinematics model of the construction robot, using the nonlinear expression of the state vector and the nonlinear equation of the constraint condition as the constraint, projecting the estimated state to the state space satisfying the constraint;

[0021] Fault detection and isolation: design chi-square distribution The detector identifies abnormal measurement values, and when a Beidou signal fault is detected, short-time pose calculation is performed through IMU data.

[0022] In combination with the first aspect, in some possible implementation manners, in step S23, a deep neural network used by the deep learning point cloud registration algorithm is constructed based on PointNet++, and the deep neural network includes a planarity feature extraction layer, a planarity regularization loss, and an edge detection layer.

[0023] The planarity feature extraction layer performs planarity feature extraction on the concrete based on a cloud local planarity index, and the cloud local planarity index is calculated according to eigenvalues of a local covariance matrix of the point cloud.

[0024] The planarity regularization loss is obtained by calculating a distance between a planarity prediction value and a planarity labeled value.

[0025] The edge detection layer performs edge detection based on an edge saliency index, and the edge saliency index is calculated according to a local curvature eigenvalue feature of the point cloud.

[0026] In combination with the first aspect, in some possible implementation manners, step S3 includes:

[0027] Construction area segmentation: using region growing and edge detection algorithms to segment the construction area, and dividing the environment point cloud into a construction area and a non-construction area;

[0028] Special structure identification: using a point cloud-based target detection algorithm to identify special structures in the construction area;

[0029] Region classification: classifying the construction area into multiple categories according to geometric features and surface characteristics, and the multiple categories include: an open area, a narrow area, and a boundary area.

[0030] In combination with the first aspect, in some possible implementation manners, step S4 includes:

[0031] Path mode selection: automatically selecting a basic path mode according to geometric shape features of the construction area, and the path mode includes: an S-shaped swing path, a spiral path, and an adaptive grid path;

[0032] Region decomposition: using an improved convex decomposition algorithm to decompose a complex-shaped construction area into multiple simple sub-regions, and generating a path for each sub-region;

[0033] Multi-objective optimization: applying a genetic algorithm to optimize multiple objectives to generate a global path; the multiple objectives include: minimizing path length, minimizing the number of turns, maximizing construction uniformity, and time constraints.

[0034] In combination with the first aspect, in some possible implementations, the method further includes:

[0035] Step S5, local path optimization: during the concrete construction process of the robot following the global path, dynamically adjusting the local path according to the real-time sensed concrete state.

[0036] In combination with the first aspect, in some possible implementations, the method further includes:

[0037] Step S6, boundary refinement: during the concrete construction process of the robot following the global path, refining the boundary of the construction area in the global paving path.

[0038] In combination with the first aspect, in some possible implementations, step S6 includes:

[0039] Boundary identification and modeling: extracting a set of boundary points by an edge detection algorithm and establishing a parametric mathematical model using a piecewise curve fitting algorithm to obtain a parametric boundary curve;

[0040] Boundary path planning: designing a boundary parallel path that moves parallel to the inside of the boundary and a vertical boundary path that extends vertically from the center area to the boundary;

[0041] Boundary effect compensation: implementing a boundary effect compensation strategy including path position compensation, speed compensation, and force compensation dynamically adjusted according to boundary curvature;

[0042] Boundary quality evaluation: real-time monitoring of boundary flatness, compactness, and connection quality, and iterative optimization based on monitoring results.

[0043] Second aspect, the embodiment provides a concrete construction intelligent auxiliary robot path planning system based on Beidou positioning, comprising:

[0044] Data acquisition unit: acquiring robot pose information and construction environment information through a Beidou high-precision positioning receiver, an inertial navigation unit, a laser radar, and a depth camera, forming multi-source heterogeneous data;

[0045] Multi-source heterogeneous data adaptive fusion unit: processing and fusing multi-source heterogeneous data to generate high-precision pose information and an environment map;

[0046] Environment analysis unit: based on the fused data, analyzing the shape characteristics of the construction area, identifying the construction area boundary and special structures in the construction area;

[0047] The concrete property perception global path planning unit generates a global path according to the shape characteristics of the construction area and the concrete properties, so that the robot performs concrete construction according to the global path. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A flowchart of a method for planning a path of a concrete construction intelligent auxiliary robot based on Beidou positioning according to some embodiments of the present application is provided.

[0049] Figure 2 A detailed step diagram of step S2, adaptive fusion of multi-source heterogeneous data, according to some embodiments of the present application is provided.

[0050] Figure 3 A technical step diagram of a deep learning point cloud registration algorithm according to some embodiments of the present application is provided.

[0051] Figure 4 A schematic diagram of different path modes in global path planning according to some embodiments of the present application is provided.

[0052] Figure 5 A detailed flowchart of step S6, boundary refinement processing, is provided. DETAILED DESCRIPTION

[0053] The terms "first", "second", "third", and "fourth" and the like in the description and in the claims of the present application and the accompanying drawings are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. The terms "include", and "have", and any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device that comprises a list of steps or units are not necessarily limited to the listed steps or units, but can optionally include additional steps or units not expressly listed or inherent to such process, method, product, or device.

[0054] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive or alternative embodiments. It is expressly understood that the embodiments described herein are combinable.

[0055] "Multiple" means two or more. "And / or", which describes the association relationship of associated objects, means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after it.

[0056] Embodiments of the present application are described below with reference to the accompanying drawings.

[0057] Research shows that there are mainly two aspects affecting the effect of concrete construction path planning, one is the complexity of the environment: due to the actual construction area often exists irregular shape, obstacles and special structure and other complex factors, these factors will lead to the construction path is difficult to optimize, thereby affecting the construction uniformity and efficiency. The second is the concrete construction boundary processing.

[0058] The concrete boundary processing method can be mainly divided into search-based path planning method and sampling-based path planning method. The search-based path planning method reduces the search range by optimizing the heuristic function, improves the path planning efficiency, and uses the method of dynamically adjusting the weight to make the path smoother, which is suitable for indoor robot navigation. However, this method may be affected by local optimal solution in complex environment, and depends on the setting of heuristic function, and has limited adaptability to different scenes, which is difficult to be directly applied to complex and changeable concrete construction environment. The sampling-based path planning method, such as improved rapid exploration random tree (RRT) algorithm, although this method is suitable for complex dynamic environment and can quickly find a feasible path, but RRT is essentially a random sampling, the path may be relatively tortuous, which needs subsequent smoothing optimization, and the calculation resource consumption is large, which is suitable for larger space, and may not perform well in narrow environment.

[0059] Therefore, the present application provides a concrete construction intelligent auxiliary robot path planning method and system based on Beidou positioning, which uses Beidou high-precision positioning receiver, inertial navigation unit, laser radar and depth camera to collect data, obtains robot pose and construction environment information, adopts multi-source heterogeneous data adaptive fusion algorithm (such as improved extended Kalman filter algorithm), processes and fuses multi-source heterogeneous data, generates high-precision pose information and environment map,

[0060] Step S1, collecting data

[0061] The robot pose information and construction environment information are collected by the Beidou high-precision positioning receiver, inertial navigation unit, laser radar and depth camera to form multi-source heterogeneous data.

[0062] Specifically, the robot pose information and construction environment information are obtained by multiple sensors, including: the robot pose information and construction environment information are collected by the Beidou high-precision positioning receiver, inertial navigation unit, laser radar and depth camera. Among them:

[0063] (1) Beidou high-precision positioning receiver: using the high-precision service of Beidou No. 3 system, obtaining the global position coordinates of the robot, and the precision can reach centimeter level;

[0064] (2) Inertial navigation unit (IMU): contains accelerometers, gyroscopes and magnetometers, obtains the attitude information of the robot, including pitch angle, roll angle and heading angle;

[0065] (3) Laser radar: 360° scans the construction environment, generates three-dimensional point cloud data, which is used for environment modeling and obstacle detection;

[0066] (4) Depth camera: obtains RGB-D data of the construction environment; used for concrete surface feature analysis and detailed environment perception.

[0067] It should be noted that the above sensors have different sampling frequencies: Beidou positioning is usually 1-10Hz, inertial navigation is 100-200Hz, laser radar is 5-20Hz, and depth camera is 30Hz. In order to ensure the timeliness and integrity of the data, a high-precision time synchronization mechanism is adopted to add time stamps to all sensor data.

[0068] Step S2, adaptive fusion of multi-source heterogeneous data (data fusion)

[0069] The multi-source heterogeneous data is processed and fused to generate high-precision pose information and environment map.

[0070] In this embodiment, the data fusion step processes and fuses multi-source sensor data to generate high-precision pose information and environment map. Specifically, as shown in Figure 2 This step includes four sub-steps:

[0071] Step S21, space-time alignment: time alignment and space alignment of multi-source heterogeneous data.

[0072] The multi-source heterogeneous sensor data is time-aligned and adaptively fused to generate high-precision pose information and environment map, including:

[0073] The time alignment of different sensor data to the same time reference and the space alignment of different sensor data to the same coordinate system, wherein the linear interpolation algorithm is used for time alignment to synchronize all sensor data to a unified time stamp. Space alignment establishes a unified coordinate transformation relationship to convert the local coordinate system data of each sensor to the robot body coordinate system and then to the global coordinate system.

[0074] Step S22, pose fusion: fusion of Beidou positioning and inertial navigation data using improved extended Kalman filter (EKF) algorithm, the improvements include: adaptive noise estimation, nonlinear state constraint, fault detection and isolation.

[0075] A, adaptive noise estimation, refers to dynamically adjusting the measurement noise covariance matrix R and the process noise covariance matrix Q according to the Beidou signal quality and the IMU data stability, and the specific implementation method is as follows:

[0076] Step S22A1, Beidou signal quality evaluation: construct a signal quality index by monitoring the number of satellites, signal strength ratio (SNR) and geometric dilution of precision (GDOP) in real time:

[0077] (1)

[0078] Among them, indicates the signal quality at time k, is the number of visible satellites at time k, is the average signal strength at time k, is the geometric dilution of precision at time k. The larger the value, the worse the Beidou signal quality.

[0079] Step S22A2, IMU data stability evaluation: use the sliding window variance analysis method to calculate the stability index of the gyroscope and accelerometer data:

[0080] (2)

[0081] (3)

[0082] Among them, , are the stability indexes of the gyroscope and accelerometer at time k respectively, W is the size of the sliding window, in this embodiment, W is set to 50 sampling points, and are the original data of the gyroscope and accelerometer in window i respectively, and are the average values of the gyroscope and accelerometer data at the end of the window at the kth sampling point.

[0083] Step S22A3, dynamic covariance matrix adjustment

[0084] According to the signal quality index and the stability index, adjust the noise covariance matrix:

[0085] (4)

[0086] (5)

[0087] Among them, , are the measurement noise covariance matrix and the process noise covariance matrix at time k respectively, and is the nominal covariance matrix, diag denotes a diagonal matrix, is the last time's BDS signal quality index, , is the last time's BDS signal quality index,

[0088] In the above adjustment, the worse the BDS signal quality is (the larger ), the greater the measurement noise covariance is, and the lower the trust in BDS positioning measurement values is. When the stability of the gyroscope and the accelerometer decreases (i.e., the index increases), the process noise increases, so as to more strongly rely on the observation values and weaken the state prediction.

[0089] B, nonlinear state constraint, by introducing the kinematic model of the construction robot as a constraint, the estimation accuracy in complex environment is improved. The specific implementation method is as follows:

[0090] Step S22B1, state vector definition

[0091] The 6-dimensional or 9-dimensional state vector usually used in the traditional EKF is extended to 15-dimensional, and the 15-dimensional state vector is used to represent the system state:

[0092] (6)

[0093] wherein the subscript k represents the k time, is the position coordinate, is the velocity, is the quaternion representing the attitude, and are the zero offsets of the gyroscope and the accelerometer respectively, and T represents the matrix transposition operation.

[0094] By explicitly modeling the gyroscope and accelerometer bias , ), state drift is avoided; the quaternion of the attitude is used to represent the attitude, the singularity of the Euler angle is avoided, and the filtering stability is improved; the 15-dimensional state vector is used to represent the system state, which is conducive to improving the estimation accuracy.

[0095] Step S22B2, kinematic constraint

[0096] Considering that the concrete spreading robot mainly moves on a plane and is mostly driven by differential, the following constraint (physical constraint) is introduced:

[0097] (7)

[0098] (8)

[0099] wherein, is the vertical velocity at the k time; yaw angle at time k, , velocity in y direction and x direction at time k, respectively.

[0100] Equation (7) constrains the vertical velocity to be 0, and equation (8) represents the sideslip constraint (i.e., differential drive has no lateral degree of freedom), which can provide physical rationality for state estimation. When the Beidou signal is invalid or the observation is incomplete, the kinematic constraint can still compensate for the missing information.

[0101] Step S22B3, constraint implementation method

[0102] The improved extended Kalman filter (EKF) algorithm is used to iteratively calculate the state estimation at time k (i.e., the estimated state), and then the state projection method is used to project the estimated state into the state space that satisfies the constraint:

[0103] (9)

[0104] wherein, is the state estimation value at time k, which combines the observation value at time k; A is a constraint matrix, is a constraint vector, which is defined as 0 in this paper, and P is a weighting matrix.

[0105] The constraint is implemented after the state update, ensuring that the final estimation value meets the kinematic characteristics of the robot.

[0106] C, fault detection and isolation, design The detector identifies abnormal measurement values, and when the Beidou signal is temporarily lost, the system performs short-time pose calculation through IMU data. The specific implementation method is as follows:

[0107] Step S22C1, abnormality detection

[0108] An innovative sequence test method (also known as: innovation test method, chi-square test method, residual sum of squares test method) is used to identify abnormal measurements:

[0109] (10)

[0110] wherein, is the observation vector at time k, is the observation equation, is the innovation covariance matrix at time k (i.e., the covariance matrix in the innovation sequence test method), is the state prediction at time k-1 for time k. When exceeds the preset threshold, it is determined as an abnormal measurement.

[0111] ​Step S22C2, fault detection preparation, calculate the observation residual by the following formula:

[0112] (11)

[0113] where, is the observation residual of Beidou at time k, is the observation vector of Beidou at time k, is the IMU observation matrix at time k.

[0114] Step S22C3, calculate the standardized residual:

[0115] (12)

[0116] where, is the observation dimension of Beidou.

[0117] By comparing the standardized residual with the chi-square distribution threshold, it is determined whether a fault has occurred. If , it is determined that Beidou has failed.

[0118] Step S22C4, degradation processing strategy

[0119] When the system detects a Beidou signal failure, appropriate degradation processing measures will be taken to ensure that the system can still maintain stable pose estimation performance in the event of a failure.

[0120] Specifically, if it is determined that Beidou has failed, the system will switch to an IMU calculation mode, i.e., using the position, velocity, and attitude at the previous time as the initial value, and calculating the current position and attitude by integrating the acceleration and angular velocity measured by the IMU:

[0121] (13)

[0122] where, are the position and velocity vectors at time k, is the attitude quaternion at time k, is the rotation matrix from the body coordinate system to the navigation coordinate system, and are the acceleration and angular velocity measurements, g is the gravity acceleration vector, is the sampling time interval, represents the position vector at the time. is the position estimated or calculated at the previous time.

[0123] Step S22C5, in the IMU estimation mode, the system continuously monitors the Beidou signal state, preparing for signal recovery. When the Beidou signal returns to normal, the system adopts a smooth transition strategy to ensure the continuity and smoothness of state estimation. The smooth transition strategy is executed according to the following formula:

[0124] (14)

[0125] wherein, represents the final state estimation result, represents the dynamically adjusted weight, and are the state estimation results based on IMU estimation and EKF, respectively. and are obtained according to the following formula:

[0126] (15)

[0127] wherein, is the Kalman gain, is the observation vector at time k, is the observation equation.

[0128] When the Beidou signal quality is poor, is small, and the IMU estimation result is relied on more; when the Beidou signal quality is good, is large, and the EKF estimation result is relied on more.

[0129] Through this multi-level fault handling mechanism, the system can maintain stable pose estimation performance in the case of abnormal or missing Beidou signal, providing reliable position and attitude information for concrete paving operations. At the same time, the system records detailed fault handling processes, providing data support for subsequent system optimization and maintenance.

[0130] Step S23, environment map fusion: use a deep learning point cloud registration algorithm to register and fuse the point cloud data of the lidar and depth camera, and construct a high-precision environment map. Specifically, as shown in Figure 3 , the algorithm includes four sub-steps:

[0131] Step S231, point cloud feature extraction

[0132] Point cloud feature extraction extracts features from lidar and depth camera point cloud data using a deep neural network. The deep neural network is based on PointNet++, which captures multi-scale features of point clouds through sampling, grouping, and feature extraction operations, and constructs a hierarchical structure of point clouds. According to the characteristics of the concrete construction environment, a concrete plane feature enhancement module is introduced into the deep neural network, and a planarity feature extraction layer is designed to calculate the local planarity index of the point cloud:

[0133] (16)

[0134] wherein, is the eigenvalue of the local covariance matrix of the point cloud.

[0135] The deep neural network introduces a plane regularization loss to strengthen the model's ability to recognize concrete planes:

[0136] (17)

[0137] wherein, is the labeled value of planarity.

[0138] The deep neural network also includes a construction site edge feature enhancement module that calculates an edge saliency index through an edge detection layer:

[0139] (18)

[0140] wherein, , is the local curvature eigenvalue of the point cloud.

[0141] Adding concrete edge samples strengthens training and improves the network's sensitivity to project boundaries.

[0142] Step S232, feature matching

[0143] Feature matching achieves matching of point cloud features from different sources through attention mechanisms and similarity measures.

[0144] In this embodiment, the attention mechanism uses a Transformer structure and is integrated after the last set abstraction (SA) module of the PointNet++ network structure. Specifically, after completing multi-layer feature extraction and aggregation, PointNet++ inputs the global features output by the last SA module into the designed concrete texture-sensitive attention mechanism module. This module takes the features output by the SA module as the query (Q) and key (K) matrices, respectively, and combines the surface texture similarity matrix MtextureMtexture for weighted calculation. The specific formula is as follows:

[0145] (19)

[0146] where Q, K are query and key matrix, is the surface texture similarity matrix, is the feature dimension.

[0147] The concrete surface texture features are weighted by the regional characteristic adaptive attention module, which enhances the sensitivity of the model to subtle texture differences on the concrete surface. Unlike the original PointNet++ network, the embodiment adds a Transformer-based attention mechanism module after the SA module, and introduces a surface texture similarity matrix in the attention calculation, thereby improving the accuracy of point cloud feature matching and the ability to distinguish concrete surface textures.

[0148] A geometric and semantic fusion dual-modal similarity measurement method is proposed, which uses a combination of cosine similarity and L2 distance, and adopts a nearest neighbor and mutual nearest neighbor bidirectional matching strategy, and the expression is as follows:

[0149] (20)

[0150] where, is the feature vector of the i, j object, and α(d) and β(d) are distance adaptive parameters.

[0151] To achieve adaptive adjustment of parameters, in the embodiment, a neural network or a differentiable parameterization function is used to model α(d) and β(d). Specifically, with distance d as input, the corresponding α(d) and β(d) values are output through the neural network or parameterization function, and the related parameters are automatically optimized in the network training process, realizing adaptive adjustment of similarity weights at different distances. This method effectively solves the problem of uneven feature matching accuracy at different distances, and improves the reliability and accuracy of matching.

[0152] Step S233, transformation matrix estimation

[0153] The transformation matrix estimation calculates the rigid transformation between point clouds based on matched feature pairs.

[0154] To address the interference of vibration, occlusion and other disturbances in the construction environment, an adaptive RANSAC threshold adjustment algorithm is developed, which eliminates abnormal matching points and calculates the optimal transformation matrix through singular value decomposition method. First, the inlier threshold is dynamically adjusted according to the vibration condition of the construction site:

[0155] (21)

[0156] where, is the basic threshold, is an adjustment coefficient, is the vibration index at time k. This improvement improves the effect of removing outliers.

[0157] The adjustment coefficient is not a fixed value set by artificial, but is dynamically adjusted by an adaptive feedback mechanism. The specific implementation is as follows: the system monitors the vibration index of the construction site and the proportion of inliers matched by RANSAC in real time, and counts the current proportion of inliers in each sliding window. If the proportion of inliers is lower than the preset threshold, the system automatically increases the value of γ to relax the threshold of inlier determination, and improves the tolerance to outliers; otherwise, the value of γ is reduced to tighten the threshold and improve the matching accuracy. This adaptive adjustment mechanism can automatically optimize the inlier determination standard of RANSAC according to the dynamic changes of the actual construction environment, thereby improving the robustness of outlier removal and the stability of overall matching.

[0158] At the same time, a confidence weighted SVD solving method is designed to solve the optimal rigid transformation matrix (i.e. rotation and translation parameters) between point clouds, and a matching quality scoring mechanism based on consistency check is introduced:

[0159] (22)

[0160] wherein, is the weight of the ith matching point pair. It is used for subsequent calculations such as weighted SVD, and reflects the confidence of the point pair in the estimation of the transformation matrix, is the feature point confidence, is the matching error, is the scale parameter, is the spatial consistency score.

[0161] The confidence weighted SVD solving method refers to introducing the weight of each matching point pair (as shown in formula (22)) to all point pairs for weighted processing based on traditional SVD (singular value decomposition). In this way, the confidence, matching error and spatial consistency information of each point pair can be fully utilized to improve the robustness and accuracy of point cloud registration in complex environments.

[0162] The above improvement significantly improves the stability and accuracy of the transformation matrix in complex construction environments.

[0163] Step S234, point cloud fusion

[0164] ​To improve the quality of point cloud fusion, voxel filtering and statistical outlier removal algorithm are used in turn. Voxel filtering is used to downsample the point cloud, remove redundant points and improve processing efficiency; the statistical outlier removal algorithm further removes noise points and abnormal points, improving the overall quality and reliability of the point cloud. The combination of the two methods can effectively optimize the data structure and accuracy of the fused point cloud, providing a high-quality data basis for subsequent environment mapping and path planning.

[0165] On the basis of optimizing the quality of point cloud by using voxel filtering and statistical outlier removal algorithm, in this embodiment, an adaptive weight calculation method based on the surface characteristics of concrete is further developed:

[0166] (23)

[0167] In the formula, is the adaptive weight of each point, is the basic weight, d is the distance from the point to the sensor, is the attenuation coefficient, is the surface texture richness.

[0168] The calculated as the weight factor of each point in the point cloud fusion and filtering process, directly participates in the subsequent weighted processing. Specifically, in the point cloud fusion, the spatial coordinates, intensity and other attributes of all points are weighted and averaged or filtered according to their weights. The higher the weight of a point, the greater the impact on the final fusion result, thereby improving the overall quality and robustness of the fused point cloud. In addition, it can also be used in subsequent point cloud resampling and other steps to further optimize the accuracy and reliability of the environment map.

[0169] The adaptive implementation method is as follows: the system real-time statistics the depth change and surface texture distribution of each region during the point cloud fusion process, and dynamically adjusts the feature change trend in the sliding window. When it is detected that the surface texture of a certain region is rich or the depth change is severe, the system automatically increases the value of to improve the weight of the points in that region; otherwise, it is reduced. Through the above mechanism, the adaptive dynamic adjustment of the weight parameter is realized. By introducing the adaptive weight calculation method based on the surface characteristics of concrete, the problem of inaccurate measurement of the smooth surface of concrete by the depth camera is solved.

[0170] Step S24, surface feature extraction: based on sensor data (such as depth camera data), a concrete surface model is constructed and its features such as flatness, roughness, humidity and hardening degree are analyzed. Details are as follows:

[0171] Step S241, flatness: evaluate the flatness of the surface through plane fitting and deviation analysis;

[0172] Step S242, roughness: analyze the surface texture features based on wavelet transform and gray level co-occurrence matrix;

[0173] Step S243, humidity: estimate the concrete surface humidity using the RGB information and reflection characteristics of the depth camera;

[0174] Step S244, hardening degree: evaluate the hardening state of the concrete according to color change and surface characteristics.

[0175] The above features are used for subsequent construction quality evaluation and path optimization to ensure that the construction quality meets the requirements.

[0176] Step S3, environment analysis

[0177] The environment analysis is based on the fused data to analyze the construction area features and identify boundaries and special structures. Specifically, it includes:

[0178] Step S31, construction area segmentation: use region growing and edge detection algorithms to segment the construction area, dividing the environment point cloud into construction and non-construction areas;

[0179] Step S32, boundary extraction: combine Canny edge detection and RANSAC curve fitting algorithms to accurately extract the boundary line of the construction area;

[0180] Step S33, special structure identification: use point cloud-based object detection algorithms to identify special structures such as columns, corners, and pipes in the construction area;

[0181] Step S34, region classification: classify regions according to geometric features and surface characteristics, classify construction areas into open areas, narrow areas, boundary areas, etc., and provide a basis for path planning.

[0182] Step S4, global path planning based on concrete characteristics (global path planning)

[0183] Global path planning generates a global paving path based on the shape of the construction area and the characteristics of the concrete. This step includes three sub-steps:

[0184] Step S41, path mode selection

[0185] As shown in Figure 4 , path mode selection automatically selects basic path modes according to the geometric shape features of the construction area, including automatically selecting S-shaped swing path, spiral path, or adaptive grid path, etc. basic path mode. Among them:

[0186] (1) S-shaped swing path: suitable for rectangular or approximately rectangular areas, reciprocating along the long side direction, with fewer turns and high coverage efficiency;

[0187] (2) Spiral path: suitable for circular or regular polygon areas, spiraling from the center outward or from the outside to the center, avoiding sharp turns and ensuring uniformity of construction;

[0188] (3) Adaptive grid path: suitable for irregularly shaped areas, dynamically adjusting grid size and direction according to area characteristics to ensure complete coverage. The system analyzes the geometric shape characteristics of the construction area (such as circularity, rectangularity, aspect ratio, etc.), automatically selects the most suitable path mode, and does not require manual intervention.

[0189] Step S42, region decomposition

[0190] For complex-shaped construction areas, an improved convex decomposition algorithm is used to decompose them into multiple simple sub-regions, and paths are generated for each sub-region. The improved convex decomposition algorithm first identifies the concave points of the region, then determines the optimal split line based on visibility and area minimization principles, and decomposes the complex region into multiple convex polygon sub-regions. The specific implementation is as follows:

[0191] Step S421, concave point identification

[0192] The concave point is identified according to the following formula:

[0193] (25)

[0194] In the formula, is a point on the boundary, represents the internal angle formed by the adjacent three points. When the internal angle is less than 180°, is the concave point.

[0195] Step S422, optimal split line selection:

[0196] (26)

[0197] In the formula, L represents the candidate split line, is the visibility index (0-1 range, the larger the value, the less obstructed the split line), is the area balance degree after splitting, is the weight coefficient.

[0198] Step S423, recursive decomposition termination condition:

[0199] (27)

[0200] In the formula, Area(R) represents the current region, determining whether the region is a convex polygon, an area threshold.

[0201] For each sub-region, the system individually plans a path, and finally connects the paths of each sub-region by designing connection points and transition paths, forming a complete global path. The sub-region connection adopts the principle of minimum path length:

[0202] (28)

[0203] wherein, is the exit point of sub-region i, is the entrance point of sub-region j, and MinPath calculates the shortest collision-free path.

[0204] Through this decomposition and connection strategy, the system can efficiently handle various complex-shaped construction regions, ensuring complete coverage while minimizing path length and turning frequency.

[0205] Step S43, multi-objective optimization

[0206] The genetic algorithm is applied for multi-objective optimization, considering multiple objectives such as path length minimization, turning frequency minimization, construction uniformity maximization, and time constraint satisfaction, etc. Each objective is described as follows:

[0207] (1) Path length minimization: reduce total travel distance and improve efficiency;

[0208] (2) Turning frequency minimization: reduce construction unevenness caused by turning;

[0209] (3) Construction uniformity maximization: ensure uniform distribution of concrete;

[0210] (4) Time constraint satisfaction: ensure completion of construction before concrete initial setting.

[0211] Multi-objective optimization is achieved using NSGA-II (Non-dominated Sorting Genetic Algorithm II), which selects the path scheme that best meets the actual construction requirements through the Pareto optimal solution set and decision maker preferences. The fitness function in the algorithm considers the above four objectives and sets weights according to actual construction requirements, balancing the relationship between each objective.

[0212] Step S5, local path optimization

[0213] Real-time path optimization based on concrete state prediction, i.e., during the execution of the global path, dynamic adjustment of the local path is performed according to the real-time perception of the concrete state. This step includes two sub-steps:

[0214] Step S51, Model Predictive Control (MPC) algorithm

[0215] Based on the real-time collected concrete state parameters, the model predictive control (MPC) algorithm is used to perform forward-looking optimization on the local path. The system dynamically monitors the key parameters such as the fluidity, temperature, and humidity of the concrete, and takes them as the input variables of the MPC optimization. The MPC optimization problem is defined as:

[0216] (29)

[0217] where J is the optimization objective function value, representing the cumulative cost in the entire prediction horizon, x represents the system state (including position, velocity, and concrete state), u represents the control input, and weight matrix, N is the prediction horizon length, is the expected system state vector at the k+i step, is the expected (reference) control input vector at the k+i-1 step.

[0218] By solving this optimization problem, the system can optimize the robot's motion trajectory and construction parameters while ensuring construction quality.

[0219] Step S52, Adaptive Dynamic Window Approach (ADWA)

[0220] The ADWA method dynamically adjusts the local path parameters to avoid obstacles based on real-time sensor data. This method first determines the feasible speed space according to the robot's dynamics constraints, and then evaluates the objective function value of different speed combinations in the speed-space search space:

[0221] (30)

[0222] where v is the linear speed, w is the angular speed, heading represents the heading score (consistency with the target direction), dist represents the obstacle distance score, velocity represents the speed score (proximity to the expected speed), is the weight coefficient.

[0223] The system selects the speed combination with the maximum objective function value to achieve dynamic optimization of the local path.

[0224] Step S6, boundary refinement processing

[0225] Boundary quality assurance refinement processing is used to perform refined processing on the construction area boundary to improve boundary construction accuracy, such as Figure 5 As shown, this step includes four sub-steps:

[0226] Step S61, boundary recognition and modeling

[0227] Boundary recognition and modeling accurately identify the boundary of the construction area and establish a parameterized mathematical model using piecewise curve fitting algorithm. First, the edge detection algorithm is used to extract the boundary point set, and then the piecewise curve fitting algorithm (such as B-spline curve) is used to represent the boundary as a parameterized curve:

[0228] (31)

[0229] where, is the control point, is the k-order B-spline basis function.

[0230] The accurate description of the boundary parameterized curve facilitates the subsequent generation of boundary parallel path and calculation of normal vector, providing a basis for boundary path planning.

[0231] Step S62, boundary path planning

[0232] Design the boundary parallel path that moves along the inside of the boundary and the vertical boundary path that extends from the center area to the boundary vertically, where:

[0233] (1) Boundary parallel path: move a certain distance d along the inside of the boundary, generated by the inside shrinkage of the boundary curve:

[0234] (32)

[0235] where, n(t) is the unit normal vector of the boundary curve at parameter t, pointing to the inside of the area.

[0236] (2) Vertical boundary path: extend from the center area to the boundary vertically, ensuring smooth transition of the boundary and the internal area of the concrete connection.

[0237] Through the complementary use of boundary parallel path and vertical boundary path, the boundary parallel path is mainly used for the initial processing of the boundary, and the vertical boundary path is used to ensure the connection quality of the boundary and the internal area.

[0238] Step S63, boundary effect compensation

[0239] Implement the boundary effect compensation strategy including path position compensation, speed compensation and force compensation according to the dynamic adjustment of boundary curvature, as follows:

[0240] (1) Path position compensation: considering the boundary geometric characteristics and the fluidity of concrete, appropriately adjust the path position at the boundary:

[0241] (33)

[0242] where, is the curvature of the curve at t, is the compensation function, the greater the curvature, the greater the compensation.

[0243] (2) Speed compensation: appropriately reduce the running speed at the boundary, increase the construction time:

[0244] (34)

[0245] where, is the speed of the normal area, is the normalized distance to the boundary, is the compensation coefficient.

[0246] (3) Force compensation: adjust the construction pressure and vibration parameters to ensure the compactness of the concrete at the boundary:

[0247] (35)

[0248] where, is the construction pressure of the normal area, is the force compensation coefficient.

[0249] Step S64, boundary quality evaluation

[0250] The boundary quality evaluation is used to monitor the flatness, compactness and connection quality of the boundary in real time, and iteratively optimize if necessary. The evaluation is based on the following indicators:

[0251] (1) Flatness, characterized by the standard deviation of the height change in the boundary area.

[0252] (2) Compactness, estimated by vibration feedback and surface features.

[0253] (3) Connection quality, representing the smoothness of the transition between the boundary and the internal area.

[0254] If the quality of a certain section of the boundary is detected to be substandard, the system will re-plan the construction path for that section of the boundary and adjust the construction parameters to ensure that the boundary quality meets the requirements.

[0255] Through fine boundary processing, the quality of the concrete construction boundary can be accurately controlled, so as to realize the high-precision forming of the building structure, such as wall flatness, floor levelness and the connection between structure and decoration, etc. Key technologies, not only improve the construction quality, but also provide a more accurate basis for subsequent decoration and installation work.

[0256] In summary, in this embodiment, data is collected using a Beidou high-precision positioning receiver, an inertial navigation unit, a laser radar, and a depth camera to obtain the robot pose and construction environment information. An improved extended Kalman filter algorithm is used for adaptive pose fusion, a deep learning point cloud registration algorithm is used for laser radar and depth camera data registration, a high-precision environment map is constructed, and the concrete surface features are analyzed. Through region growing, edge detection, and other algorithms, the environment is analyzed, the construction area boundary and special structures are identified and classified. According to the shape of the construction area and the characteristics of the concrete, the path mode is automatically selected, the improved convex decomposition algorithm and NSGA-II non-dominated sorting genetic algorithm are combined to generate a global paving path. The model predictive control algorithm and the adaptive dynamic window method are used to optimize the local path according to the real-time concrete state. The boundary is accurately identified and modeled, a special path is designed, a boundary effect compensation strategy is implemented, the boundary quality is monitored in real time, and iterative optimization is performed.

[0257] The high-precision automation of concrete construction is realized. Actual application proves that the construction quality is significantly improved, such as ensuring the flatness, compactness, and connection quality of the boundary, and improving the uniformity of concrete construction. The construction efficiency is effectively improved, the path length and the number of turns are reduced, and it is ensured that the construction is completed before the initial setting of the concrete. The labor intensity is reduced, and manual intervention is reduced. A reliable technical support is provided for the automation and intelligentization of concrete construction, which can be widely applied to the construction of concrete floors, road paving, bridge construction, and other fields in the field of building engineering, and has significant industrial practical value.

[0258] In summary, the method provided in this embodiment effectively solves the key technical problems in concrete construction through multi-sensor data fusion, deep learning point cloud processing, hierarchical path planning, and boundary refinement. In particular, the accuracy of boundary processing is improved. Actual application proves that the method provided in this embodiment not only improves the construction quality, but also improves the construction efficiency and reduces the labor intensity, providing a reliable technical support for the automation and intelligentization of concrete construction.

[0259] From another point of view, the method performs multi-source data acquisition through Beidou high-precision positioning, an inertial navigation unit, a laser radar and a depth camera, adopts an improved extended Kalman filtering algorithm for pose fusion, uses a deep learning point cloud registration algorithm to realize accurate registration of laser radar and depth camera data, and generates a high-precision environment map; based on the fused data, a coverage path planning algorithm and a multi-objective optimization algorithm are used to generate a global construction path, and a model predictive control and an adaptive dynamic window method are used to optimize a local path in real time; for the boundary of the construction area, a boundary refinement processing algorithm is applied to improve the construction precision of the boundary. Through multi-sensor data fusion, point cloud deep learning processing, hierarchical path planning and boundary refinement processing, high-precision automation of concrete construction is realized, and the construction quality and efficiency are significantly improved, and in particular, the problem of inaccurate boundary processing in traditional construction is solved.

[0260] Based on the same inventive concept, the embodiment also provides a Beidou positioning-based intelligent auxiliary robot path planning system for concrete construction, comprising:

[0261] A data acquisition unit acquires robot pose information and construction environment information through a Beidou high-precision positioning receiver, an inertial navigation unit, a laser radar and a depth camera, and forms multi-source heterogeneous data;

[0262] A multi-source heterogeneous data adaptive fusion unit processes and fuses the multi-source heterogeneous data to generate high-precision pose information and an environment map;

[0263] An environment analysis unit analyzes the shape features of the construction area, identifies the boundary of the construction area and special structures in the construction area based on the fused data;

[0264] A concrete property perception global path planning unit generates a global path according to the shape features of the construction area and the properties of the concrete, so that the robot performs concrete construction according to the global path.

[0265] The Beidou positioning-based intelligent auxiliary robot path planning system for concrete construction provided by the embodiment can realize the steps and processes of the Beidou positioning-based intelligent auxiliary robot path planning method for concrete construction provided by any of the above embodiments, and achieve the same technical effects, which will not be repeated here.

[0266] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Those skilled in the art can make various modifications and changes to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A path planning method for a concrete construction intelligent auxiliary robot based on Beidou positioning, characterized in that, The method comprises the following steps: Step S1, data acquisition: collecting robot pose information and construction environment information through a Beidou high-precision positioning receiver, an inertial navigation unit, a laser radar and a depth camera to form multi-source heterogeneous data; Step S2, adaptive fusion of multi-source heterogeneous data: processing and fusing the multi-source heterogeneous data to generate high-precision pose information and an environment map; Step S3, environment analysis: based on the fused data, analyzing the shape characteristics of the construction area, and identifying the construction area boundary and special structures in the construction area; Step S4, concrete characteristic perception global path planning: generating a global path according to the shape characteristics of the construction area and the concrete characteristics, and performing concrete construction according to the global path; Step S2 comprises: Step S21, space-time alignment: time alignment and space alignment of multi-source heterogeneous data; Step S22, pose fusion: fusing the positioning data obtained by the Beidou high-precision positioning receiver and the inertial navigation data by using an improved extended Kalman filter (EKF) algorithm; Step S23, environment map fusion: registering and fusing the point cloud data of the laser radar and the depth camera by using a deep learning point cloud registration algorithm to construct a high-precision environment map; Step S24, surface feature extraction: extracting the flatness, roughness, humidity and hardening degree characteristics of the concrete surface, and using the extracted surface features to represent the characteristics of the concrete; In step S22, the improved extended Kalman filter (EKF) algorithm comprises adaptive noise estimation, nonlinear state constraint and fault detection and isolation; Adaptive noise estimation: dynamically adjusting the measurement noise covariance matrix R and the process noise covariance matrix Q according to the Beidou signal quality and the IMU data stability; the specific implementation method is as follows: Step S22A1, Beidou signal quality evaluation: constructing a signal quality index by monitoring the number of satellites, signal strength ratio (SNR) and geometric dilution of precision (GDOP) in real time: ， wherein, represents the signal quality at time k, is the number of visible satellites at time k, is the average signal strength at time k, is the geometric dilution of precision at time k; Step S22A2, IMU data stability evaluation: calculating the stability index of the gyroscope and accelerometer data by using a sliding window variance analysis method: ， ， wherein, , are the stability index of the gyroscope and accelerometer at time k, respectively, and W is the size of the sliding window, and are the raw data of the gyroscope and accelerometer within window i, respectively, and are the mean of the gyroscope and accelerometer data at the end of the window for the kth sample point, respectively. Step S22A3, dynamic covariance matrix adjustment: adjusting the noise covariance matrix according to the signal quality index and the stability index: ， ， wherein, , are the measurement noise covariance matrix and the process noise covariance matrix at time k, respectively, and is a nominal covariance matrix, diag denotes a diagonal matrix, is the last time's BDS signal quality index, , are the stability indexes of the gyroscope and the accelerometer at time k, respectively; Nonlinear state constraint: based on the kinematic model of the construction robot, using the nonlinear expression of the state vector and the nonlinear equation of the constraint condition as the constraint to project the estimated state to the state space satisfying the constraint; Fault detection and isolation: design chi-square distribution The detector identifies abnormal measurements and when a fault in the Beidou signal is detected, a short-time pose estimation is performed using IMU data.

2. The method of claim 1, wherein, In step S23, the deep neural network used by the deep learning point cloud registration algorithm is constructed based on PointNet++, and the deep neural network comprises a flatness feature extraction layer, a flatness regularization loss and an edge detection layer; The flatness feature extraction layer extracts the flatness features of the concrete based on a cloud local flatness index, which is calculated according to the eigenvalues of the local covariance matrix of the point cloud; The flatness regularization loss is obtained by calculating the distance between the flatness predicted value and the flatness labeled value; The edge detection layer performs edge detection based on an edge saliency index, which is calculated according to the local curvature eigenvalues of the point cloud.

3. The method of claim 1, wherein, Step S3 includes: Construction area segmentation: using region growing and edge detection algorithms to segment the construction area, dividing the environment point cloud into construction and non-construction areas; Special structure identification: using a point cloud-based target detection algorithm to identify special structures in the construction area; Region classification: classifying the construction area into multiple categories according to geometric features and surface characteristics, including open areas, narrow areas, and boundary areas.

4. The method of claim 1, wherein, Step S4 includes: Path pattern selection: automatically selecting a basic path pattern according to the geometric shape characteristics of the construction area, including S-shaped swing path, spiral path, and adaptive grid path; Region decomposition: using an improved convex decomposition algorithm to decompose complex-shaped construction areas into multiple simple sub-regions, and generating paths for each sub-region; Multi-objective optimization: applying a genetic algorithm for multi-objective optimization to generate a global path; the multiple objectives include minimizing path length, minimizing the number of turns, maximizing construction uniformity, and time constraints.

5. The method of claim 1, wherein, The method further includes: Step S5, local path optimization: during the robot's concrete construction process according to the global path, dynamically adjusting the local path according to the real-time perception of the concrete state.

6. The method of claim 1, wherein, The method further includes: Step S6, boundary refinement: during the robot's concrete construction process according to the global path, refining the construction area boundary in the global paving path.

7. The method of claim 6, wherein, Step S6 includes: Boundary identification and modeling: extracting boundary point sets through edge detection algorithms and establishing a parameterized mathematical model using piecewise curve fitting algorithms to obtain a boundary parameterized curve; Boundary path planning: designing a boundary parallel path that moves parallel to the inside of the boundary and a vertical boundary path that extends from the center area to the boundary; Boundary effect compensation: implementing a boundary effect compensation strategy that includes dynamically adjusting path position compensation, speed compensation, and force compensation according to boundary curvature; Boundary quality evaluation: real-time monitoring of boundary flatness, compactness, and connection quality, and iterative optimization based on monitoring results.

8. A Beidou positioning-based concrete construction intelligent auxiliary robot path planning system, characterized in that, Including: Data acquisition unit: collecting robot pose information and construction environment information through Beidou high-precision positioning receivers, inertial navigation units, laser radars, and depth cameras to form multi-source heterogeneous data; Multi-source heterogeneous data adaptive fusion unit: processing and fusing multi-source heterogeneous data to generate high-precision pose information and environment maps; Environment analysis unit: based on the fused data, analyzing the shape characteristics of the construction area, identifying the construction area boundary and special structures in the construction area; Concrete property perception global path planning unit: generating a global path according to the shape characteristics of the construction area and the properties of the concrete to enable the robot to perform concrete construction according to the global path; The multi-source heterogeneous data adaptive fusion unit is further configured to: Space-time alignment: time-aligning and space-aligning multi-source heterogeneous data; Pose fusion: using an improved extended Kalman filter (EKF) algorithm to fuse positioning data obtained by a Beidou high-precision positioning receiver with inertial navigation data; Environment map fusion: Use deep learning point cloud registration algorithm to register and fuse the point cloud data of lidar and depth camera, and construct high-precision environment map; Surface feature extraction: Extract the flatness, roughness, humidity and hardening degree features of the concrete surface, and use the extracted surface features to represent the characteristics of the concrete; Improved extended Kalman filter (EKF) algorithm includes: adaptive noise estimation, nonlinear state constraint, fault detection and isolation; Adaptive noise estimation: According to the quality of Beidou signal and the stability of IMU data, dynamically adjust the measurement noise covariance matrix R and the process noise covariance matrix Q; The specific implementation method is as follows: Step S22A1, Beidou signal quality evaluation: Construct signal quality index by real-time monitoring of satellite number, signal strength ratio (SNR) and geometric dilution of precision (GDOP): ， wherein, represents the signal quality at time k, is the number of visible satellites at time k, is the average signal strength at time k, is the geometric dilution of precision at time k; Step S22A2, IMU data stability evaluation: Calculate the stability index of gyroscope and accelerometer data by using sliding window variance analysis method: ， ， wherein, , are the stability index of the gyroscope and accelerometer at time k respectively, W is the size of the sliding window, and are the raw data of the gyroscope and accelerometer within window i respectively, and are the mean of the gyroscope and accelerometer data at the kth sample point which is the end of the window respectively. Step S22A3, dynamic covariance matrix adjustment: According to the signal quality index and stability index, adjust the noise covariance matrix: ， ， wherein, , are the measurement noise covariance matrix and the process noise covariance matrix at time k, respectively, and is a nominal covariance matrix, diag denotes a diagonal matrix, is the last time's BDS signal quality index, , are the stability indexes of the gyroscope and the accelerometer at time k, respectively; Nonlinear state constraint: Based on the kinematics model of construction robot, use the nonlinear expression of state vector and the nonlinear equation of constraint condition as constraint, project the estimated state to the state space satisfying the constraint; Fault detection and isolation: design chi-square distribution The detector identifies abnormal measurements and when a fault in the Beidou signal is detected, a short-time pose estimation is performed using IMU data.

Images