Mobile robot positioning and navigation method and device, electronic device and storage medium

Abstract

The application provides a mobile robot positioning and navigation method and device, electronic equipment and storage medium, and relates to the technical field of robot positioning and navigation. The method comprises the following steps: performing fusion processing on multi-dimensional characteristics to generate a metal interference factor, and dynamically configuring the weights of observation factors corresponding to different sensors arranged on the mobile robot based on the metal interference factor to obtain a weight adjustment result; inputting the weight adjustment result into a factor graph model and performing optimization solving to obtain an optimal pose estimation result, so as to control the mobile robot to perform positioning and navigation. The method aims to solve the problem that in a complex environment such as industrial inspection, metal structures interfere with multi-source sensor data, resulting in a decrease in the positioning and navigation accuracy of the mobile robot, and the existing fixed weight sensor fusion method cannot adapt to dynamic interference changes, thereby effectively improving the positioning accuracy and robustness of the mobile robot in a complex industrial environment.

Description

Technical Field

[0001] This invention relates to the field of robot positioning and navigation technology, and more specifically, to a mobile robot positioning and navigation method, device, electronic device, and storage medium. Background Technology

[0002] With the continuous development of robotics technology, mobile robots have been widely used in the field of industrial inspection. In typical industrial inspection scenarios such as petrochemical warehouses, docks, and power substations, mobile robots undertake key tasks such as daily equipment inspection, abnormal working condition investigation, and data information collection. These tasks have extremely high requirements for the positioning and navigation accuracy of mobile robots. Only by obtaining accurate and reliable pose estimation results can mobile robots accurately reach the designated inspection location and complete the corresponding work.

[0003] However, in these typical industrial inspection scenarios, there are a large number of pipes, tanks, production equipment, and support structures made of metal materials. The surface physical characteristics of these metal structures are special and can cause strong interference to the detection signals of lidar and vision sensors. The most prominent interference is the specular reflection and multipath effect caused by the metal surface. This interference can seriously affect the normal operation of synchronous positioning and mapping technology based on multi-sensor fusion, causing the positioning and navigation system of mobile robots to face serious performance degradation problems and fail to meet the accuracy requirements of industrial inspection operations.

[0004] From the perspective of specific technical defects, firstly, the quality of the point cloud degrades. The specular reflection of the metal surface causes anomalies in the detection signal received by the lidar, resulting in a large number of discrete flying points or false planes that do not conform to the structure of the real scene in the generated lidar point cloud. During point cloud registration, these anomalies and erroneous planes will severely damage the convergence of commonly used point cloud registration algorithms such as ICP and NDT, and may even cause the registration process to diverge, failing to obtain correct point cloud registration results, and thus leading to pose estimation errors. Secondly, point cloud feature extraction is prone to failure. Most existing positioning and navigation methods rely on the reflection intensity features or geometric features of the point cloud, such as edge features and planar features, to complete inter-frame matching or map association. However, in areas with metal interference, the reflection intensity distribution and geometric structure of the point cloud have already been distorted. The extracted features cannot reflect the real scene structure, thus causing the feature association method to fail and failing to achieve correct data association. Furthermore, traditional sensor fusion methods suffer from the limitation of fixed-weight fusion. Traditional tightly coupled visual-inertial odometry (VIO) or laser-inertial odometry (LIO) often employs pre-calibrated fixed sensor noise models, i.e., fixed covariance matrices, to assign weights to observations from different sensors. However, the intensity of metallic interference dynamically changes in time and space depending on the inspection scenario and the robot's pose and viewpoint. Fixed weights cannot adapt to this dynamic change, failing to adaptively reduce the weight of observations from areas with strong interference or increase the contribution of observations from reliable areas. This still introduces erroneous observations due to interference into the optimization process, leading to deviations in pose estimation. Finally, navigation performance deteriorates drastically in densely packed metal areas. In areas with abundant metal structures, these defects accumulate, ultimately causing the positioning error of the entire system to far exceed the permissible range. This prevents the robot from accurately stopping at designated inspection points, hindering precise point-to-point detection tasks and the normal operation of daily inspections, significantly impeding industrial production and maintenance.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] The purpose of this invention is to provide a mobile robot positioning and navigation method, device, electronic device, and storage medium, aiming to solve the problem that in complex environments such as industrial inspection, metal structures interfere with multi-source sensor data, leading to a decrease in the positioning and navigation accuracy of mobile robots, and that existing fixed-weight sensor fusion methods cannot adapt to dynamic interference changes, thereby effectively improving the positioning accuracy and robustness of mobile robots in complex industrial environments.

[0007] In a first aspect, the present invention provides a mobile robot positioning and navigation method, comprising the following steps: S1. Acquire multi-source sensor data of the mobile robot in the current environment; the multi-source sensor data includes scanned point cloud data, image data, and inertial measurement data; S2. Divide the scanned point cloud data into several local regions and extract the multi-dimensional features of each local region; S3. The multi-dimensional features are fused to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local area; S4. Based on the metal interference factor, dynamically configure the weights of the observation factors corresponding to different sensors deployed on the mobile robot to obtain the weight adjustment result for the metal interference factor; the dynamic weight configuration includes: adjusting the covariance matrix of the observation factors corresponding to the scanned point cloud data and the observation factors corresponding to the image data in real time according to the value of the metal interference factor, so as to reduce the weight of the data in the interfered area. S5. Obtain the observation factor corresponding to the inertial measurement data; S6. Input the weight adjustment result and the observation factor corresponding to the inertial measurement data into a preset factor graph model, and optimize the factor graph model to obtain the optimal pose estimation result of the mobile robot; S7. Based on the optimal pose estimation result, control the mobile robot to perform positioning and navigation.

[0008] The mobile robot positioning and navigation method provided by this invention can effectively suppress the influence of metal interference on positioning and navigation by dynamically adjusting the weight of sensor observation factors, thereby improving the positioning accuracy and robustness of mobile robots in complex industrial environments.

[0009] In a second aspect, the present invention provides a mobile robot positioning and navigation device, comprising: The first acquisition module is used to acquire multi-source sensor data of the mobile robot in the current environment; the multi-source sensor data includes scanned point cloud data, image data and inertial measurement data; The extraction module is used to divide the scanned point cloud data into several local regions and extract multi-dimensional features of each local region; The generation module is used to fuse the multi-dimensional features to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local area. The configuration module is used to dynamically configure the weights of the observation factors corresponding to different sensors deployed on the mobile robot based on the metal interference factor, and obtain the weight adjustment result for the metal interference factor; the dynamic weight configuration includes: adjusting the covariance matrix of the observation factors corresponding to the scanned point cloud data and the observation factors corresponding to the image data in real time according to the value of the metal interference factor, so as to reduce the weight of the data in the interfered area. The second acquisition module is used to acquire the observation factor corresponding to the inertial measurement data; The solution module is used to input the weight adjustment results and the observation factors corresponding to the inertial measurement data into a preset factor graph model, and to optimize and solve the factor graph model to obtain the optimal pose estimation result of the mobile robot. The control module is used to control the mobile robot to perform positioning and navigation based on the optimal pose estimation result.

[0010] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the mobile robot positioning and navigation method provided in the first aspect above.

[0011] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the mobile robot positioning and navigation method provided in the first aspect above.

[0012] As can be seen from the above, the mobile robot positioning and navigation method provided by this invention effectively solves the problems in existing technologies where metal structures in industrial inspection scenarios cause strong interference to LiDAR and visual sensor signals, leading to point cloud quality degradation, poor convergence of registration algorithms, and the inability of fixed sensor noise models to adapt to dynamic interference changes. This method introduces a metal interference factor, achieving refined perception and quantification of environmental interference, and dynamically adjusting the weights of sensor observation factors accordingly. This allows the positioning and navigation system to adaptively reduce the contribution of data from interfered areas and increase the weight of data from reliable areas, thereby overcoming the limitations of existing fixed-weight methods. Therefore, this application can significantly improve the positioning and navigation accuracy and robustness of mobile robots in complex industrial environments with metal interference, ensuring their accurate completion of inspection tasks and possessing significant practical application value.

[0013] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0014] Figure 1 This is a flowchart of a mobile robot positioning and navigation method provided in an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of a mobile robot positioning and navigation device provided in an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0017] Label Explanation: 100. First acquisition module; 200. Extraction module; 300. Generation module; 400. Configuration module; 500. Second acquisition module; 600. Solving module; 700. Control module; 13. Electronic device; 1301. Processor; 1302. Memory; 1303. Communication bus. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] In traditional industrial inspection scenarios, when mobile robots use LiDAR for synchronous localization and mapping, specular reflection and multipath effects generated by the metal structure surface cause discrete flying points and false planes in the scanned point cloud data, disrupting the convergence of the point cloud registration algorithm. In particular, feature extraction methods based on reflection intensity or geometric features fail in metal areas due to signal distortion. Furthermore, fixed-weight sensor fusion strategies cannot adapt to the dynamic changes of metal interference in time and space, resulting in a decline in positioning and navigation performance.

[0021] For example, in the inspection task of a petrochemical storage area, when the mobile robot moves near the metal storage tank, the lidar detection signal is specularly reflected on the metal surface, resulting in abnormal discrete points in the scanned point cloud data. At the same time, the image captured by the vision sensor is blurred due to strong reflection. In this specific scenario, the point cloud registration algorithm cannot correctly match the data between consecutive frames, the feature extraction module outputs incorrect association results, and the fixed weight fusion mechanism fails to reduce the contribution of the interfered sensors, causing the robot pose estimation to drift and making it unable to accurately reach the designated detection point to perform instrument reading or valve status detection tasks.

[0022] If the above problems are not resolved, the point cloud quality degradation and feature extraction failure caused by metal interference will continue to affect the stability of the positioning system, causing the robot to accumulate positioning errors during critical task execution, and thus be unable to complete high-precision fixed-point detection operations. This not only reduces inspection efficiency, but may also cause safety risks due to positioning deviations, such as misjudging equipment status or missing fault points.

[0023] For this, please refer to Figure 1 , Figure 1 This is a flowchart of a mobile robot localization and navigation method. The mobile robot localization and navigation method provided by this invention includes the following steps: S1. Acquire multi-source sensor data of the mobile robot in the current environment; multi-source sensor data includes scan point cloud data, image data, and inertial measurement data; S2. Divide the scanned point cloud data into several local regions and extract multi-dimensional features of each local region; the multi-dimensional features include intensity features reflecting the distribution of signal reflection energy, geometric features reflecting the local geometric topology, density features reflecting the spatial distribution density of the point cloud, and stability features reflecting the changes in point cloud pose between consecutive frames. S3. Perform fusion processing on multi-dimensional features to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local area (environmental interference mainly refers to the interference caused by metal structures to sensors in the inspection scenario, such as the specular reflection of metal surfaces). S4. Based on the metal interference factor, the observation factors corresponding to different sensors deployed on the mobile robot are dynamically weighted to obtain the weight adjustment results for the metal interference factor. The dynamic weight configuration includes: adjusting the covariance matrix of the observation factors corresponding to the scanned point cloud data (i.e., the observation factors corresponding to the LiDAR) and the observation factors corresponding to the image data (i.e., the observation factors corresponding to the vision sensor) in real time according to the value of the metal interference factor, so as to reduce the weight of the data in the interfered area (environmental interference will only affect the LiDAR and vision sensor, that is, the metal interference factor will only affect the observation factors of these two sensors). S5. Obtain the observation factor corresponding to the inertial measurement data (i.e., the observation factor corresponding to the inertial sensor). S6. Input the weight adjustment results and the observation factors corresponding to the inertial measurement data (environmental interference will not affect the inertial sensor, so the observation factors corresponding to the inertial sensor remain unchanged, i.e., their covariance matrix is ​​fixed) into the preset factor graph model, and optimize the factor graph model (the goal is to minimize the weighted sum of squared residuals of all observation factors; in factor graph optimization, each sensor observation factor has an information matrix). ( covariance matrix The invention dynamically adjusts the covariance of the observation factors corresponding to the lidar and the observation factors corresponding to the vision sensor (inverse matrix); in addition, the optimization problem is solved in real time using algorithms such as incremental smoothing and mapping (iSAM2) to obtain the optimal pose estimation result of the mobile robot. S7. Based on the optimal pose estimation result, control the mobile robot to perform localization and navigation.

[0024] For ease of understanding, the following explains some key terms in this embodiment: Mobile robots are robotic systems capable of autonomous movement, environmental perception, and the execution of specific tasks. They are typically equipped with various sensors, such as lidar, vision sensors, and inertial measurement units, to achieve localization, navigation, and environmental awareness.

[0025] Multi-source sensor data: refers to the raw data set acquired by the mobile robot through different types of sensors. In this embodiment, it includes scanned point cloud data (acquired by LiDAR, reflecting the three-dimensional geometric information of the environment), image data (acquired by vision sensors, providing two-dimensional visual information of the environment), and inertial measurement data (acquired by inertial measurement unit, providing the robot's attitude, angular velocity, and acceleration information).

[0026] Local region: refers to the spatial division of scanned point cloud data into several smaller, independent sub-regions. This division helps in the refined analysis of the environment, especially when assessing local environmental disturbances.

[0027] Multi-dimensional features refer to various quantitative indicators extracted from scanned point cloud data of a local area to characterize the physical properties of that area. In this embodiment, these include intensity features, geometric features, density features, and stability features, which reflect the possibility of environmental interference with the point cloud data from different perspectives.

[0028] Metal interference factor: This is a scalar value used to quantitatively characterize the degree of environmental interference (especially interference caused by metal structures) in various local areas. This factor is usually between 0 and 1, with the closer the value is to 1, the stronger the interference, and the closer it is to 0, the weaker the interference.

[0029] Observation factors: In the factor graphical model, observation factors represent the mathematical relationship between sensor measurements and robot state variables. Each observation factor is associated with a covariance matrix, which describes the uncertainty of measurement noise.

[0030] Dynamic weight configuration refers to the process of adjusting the covariance matrix of the observation factors corresponding to different sensors in real time based on the value of the metal interference factor. By adjusting the covariance matrix, the weight of sensor data in factor graph optimization can be changed, thereby suppressing the influence of interference data.

[0031] Factor graphical model: A mathematical tool for representing probabilistic graphical models, commonly used in robot localization and mapping (SLAM) problems. It represents robot pose, sensor observations, and motion models as nodes and factors, and estimates the robot's optimal pose through optimization.

[0032] Optimal pose estimation result: refers to the best estimate of the position and orientation of the mobile robot at the current moment, obtained by optimizing the solution through the factor graph model.

[0033] This embodiment aims to address the problem that metal structures in complex industrial inspection scenarios interfere with LiDAR and vision sensors, leading to point cloud quality degradation, feature extraction failure, and the inability of fixed weights to adapt to dynamic changes in interference, ultimately resulting in a sharp decline in the localization and navigation performance of mobile robots. To address this, this application proposes a mobile robot localization and navigation method. Through a closed-loop framework of perception, evaluation, and adjustment, it achieves real-time quantification, dynamic weight adjustment, and robust optimization of metal interference, thereby significantly improving the localization reliability of inspection robots in complex industrial environments.

[0034] In one implementation, the mobile robot localization and navigation method includes the following steps: First, acquire multi-source sensor data from the mobile robot in the current environment. This multi-source sensor data can include scanned point cloud data, image data, and inertial measurement data. For example, the mobile robot can be equipped with a LiDAR to acquire scanned point cloud data, a camera to acquire image data, and an inertial measurement unit (IMU) to acquire inertial measurement data. These sensors can operate independently or transmit data to the processing unit through an integrated interface. In practical applications, different models and levels of sensor accuracy can be selected based on the specific scenario requirements. For example, in scenarios where high accuracy is not required, lower-cost sensors can be used; while in scenarios requiring high-precision positioning, high-precision sensors are necessary.

[0035] Secondly, the scanned point cloud data is divided into several local regions, and multi-dimensional features of each local region are extracted. For example, the scanned point cloud data can be segmented into multiple non-overlapping or partially overlapping local regions using spatial partitioning methods such as voxelization, Kd-trees, or octrees. Each local region can contain a fixed number of point clouds or have a fixed spatial size. When extracting multi-dimensional features, intensity features, geometric features, density features, and stability features can be calculated for each local region. For example, intensity features can be obtained by statistically analyzing the mean, variance, or outliers of the point cloud reflection intensity within the local region; geometric features can be analyzed using methods such as principal component analysis (PCA) to determine the shape and orientation of the local point cloud; density features can be obtained by calculating the ratio of the number of point clouds within the local region to the region volume; and stability features can be evaluated by comparing the pose changes of the local region point cloud in consecutive frames.

[0036] Next, the multi-dimensional features are fused to generate a metal interference factor that quantitatively characterizes the degree of environmental interference in each local area. For example, the extracted intensity, geometric, density, and stability features can be combined into a feature vector. Then, machine learning models, such as logistic regression, support vector machine (SVM), or neural networks, can be used to process this feature vector and output a scalar value between 0 and 1 as the metal interference factor. This model can be trained on a real-world point cloud dataset containing both metallic and non-metallic regions. For example, during training, point cloud data of metallic regions can be labeled as high interference (close to 1), and non-metallic regions can be labeled as low interference (close to 0).

[0037] Then, based on the metal interference factor, the observation factors corresponding to different sensors deployed on the mobile robot are dynamically weighted to obtain the weight adjustment results for the metal interference factor. Dynamic weight configuration may include: adjusting the covariance matrices of the observation factors corresponding to the scanned point cloud data (i.e., the observation factors corresponding to the LiDAR) and the observation factors corresponding to the image data (i.e., the observation factors corresponding to the vision sensor) in real time according to the value of the metal interference factor, in order to reduce the weight of the data in the interfered area. For example, when the metal interference factor value is high, the element values ​​of the covariance matrix of the LiDAR and vision sensor observation factors can be increased, thereby reducing the weight of these observation factors in subsequent optimization. Conversely, when the metal interference factor value is low, the element values ​​of the covariance matrix can be decreased to increase its weight. This adjustment can be linear or non-linear, for example, using an exponential function or a sigmoid function to achieve finer weight control.

[0038] Subsequently, the observation factors corresponding to the inertial measurement data are obtained. For example, accelerometer and gyroscope data provided by the inertial measurement unit (IMU) can be pre-integrated to generate the observation factors corresponding to the inertial measurement data. Since inertial sensors are generally unaffected by metal interference, the covariance matrix of their observation factors can remain fixed and does not change with variations in the metal interference factor.

[0039] Finally, the weight adjustment results and the observation factors corresponding to the inertial measurement data are input into a preset factor graph model, and the factor graph model is optimized to obtain the optimal pose estimation result of the mobile robot. For example, the factor graph model can include robot pose nodes, inertial measurement factors, LiDAR observation factors, and visual observation factors. In the optimization process, the objective is to minimize the weighted sum of squared residuals of all observation factors. The weights of the LiDAR observation factors and visual observation factors (determined by the inverse of the covariance matrix) are dynamically adjusted based on the metal interference factor, while the weight of the inertial measurement factors is fixed. The optimization problem can be solved in real time using algorithms such as Incremental Smoothing and Mapping (iSAM2) to meet the real-time requirements of mobile robot localization and navigation.

[0040] Based on the optimal pose estimation results, the mobile robot is controlled to perform localization and navigation. For example, the optimal pose (including position and orientation) of the robot obtained from the optimization solution is used as input and passed to the motion controller of the mobile robot. The motion controller generates corresponding speed and steering commands based on the target path and the current pose, driving the mobile robot to move along the preset path and perform inspection tasks.

[0041] The following example will provide a more detailed explanation of the above technical solution: Imagine an industrial inspection scenario where a mobile robot needs to navigate autonomously through an area containing numerous metal pipes and equipment. Traditional positioning and navigation methods suffer from significant degradation in data quality from lidar and vision sensors when facing metal structures due to specular reflection and multipath effects. This leads to increased positioning errors and even positioning loss. Without addressing these issues, the mobile robot will be unable to accurately perform inspection tasks, such as precisely reading instrument data or detecting valve status, thus impacting the safety and efficiency of industrial production.

[0042] To address this issue, this application proposes a mobile robot localization and navigation method. Specifically, when the mobile robot enters a densely packed metal area, its onboard LiDAR, camera, and inertial measurement unit (IMU) first acquire scanned point cloud data, image data, and inertial measurement data in real time. For example, at time t, the LiDAR acquires one frame of point cloud data, the camera acquires one frame of image data, and the IMU continuously outputs acceleration and angular velocity data.

[0043] Next, the system divides the acquired scanned point cloud data into several local regions. For example, the point cloud data can be divided into cubic voxels with a side length of 0.5 meters. For each local region, the system extracts its multi-dimensional features, including intensity features, geometric features, density features, and stability features. For example, in a local region, if there is a metal pipe, its point cloud reflection intensity may be abnormally high and concentrated, resulting in a high intensity feature value; at the same time, specular reflection on the metal surface may form a seemingly smooth "pseudo-plane," but its normal vector distribution is different from that of the real plane, resulting in a high geometric feature value; in addition, specular reflection may cause the laser beam to be reflected elsewhere, making the point cloud in this local region sparse, resulting in a high density feature value; and metal reflection points may undergo drastic jumps between consecutive frames due to small changes in viewing angle, resulting in a high stability feature value.

[0044] The system then fuses these multi-dimensional features to generate a metal interference factor that quantitatively characterizes the degree of environmental interference in each local area. For example, using a pre-trained logistic regression model, intensity features, geometric features, density features, and stability features are taken as input, and a metal interference factor between 0 and 1 is output. If the metal interference factor of a local area is close to 1, it indicates that the area is subject to strong metal interference; if it is close to 0, it indicates that the interference is weak.

[0045] Subsequently, based on the calculated metal interference factor, the system dynamically configures the weights of the observation factors corresponding to the LiDAR and visual sensors. For example, if the metal interference factor is high in a certain local area (e.g., 0.8), the system will increase the covariance matrix of the observation factors corresponding to the LiDAR and visual sensors in that area in real time, thereby reducing the weight of these observation factors in factor map optimization. This means that the influence of LiDAR point cloud data and image data from this strongly interfering area on the localization calculation will be weakened. Conversely, if the metal interference factor is low in a certain local area (e.g., 0.1), the covariance matrix will remain small, allowing the sensor data from these areas to play a greater role in optimization. Meanwhile, the covariance matrix of the observation factors corresponding to the inertial measurement data remains constant because it is not affected by metal interference.

[0046] Finally, the weighted LiDAR and visual observation factors, along with the fixed inertial measurement observation factors, are input into a pre-defined factor graph model. The factor graph model performs optimization, for example using the iSAM2 algorithm, to minimize the weighted sum of squared residuals of all observation factors, thus obtaining the optimal pose estimation result for the mobile robot. Because the weights of the LiDAR and visual sensors are reduced in areas with strong metal interference, the negative impact of these unreliable data on the final pose estimation is effectively suppressed; while in non-interference areas, reliable sensor data plays its due role. Inertial measurement data consistently provides stable pose constraints. Therefore, the system can output a high-precision robot pose even in dense metal environments. Based on this optimal pose estimation result, the mobile robot can be precisely controlled, achieving reliable localization and navigation to complete its inspection tasks.

[0047] The mobile robot localization and navigation method in this embodiment introduces a metal interference factor to quantitatively assess environmental interference and dynamically adjusts the multi-sensor fusion strategy accordingly, significantly improving localization and navigation performance in complex industrial scenarios. Compared to traditional methods, this embodiment has the following technical contributions: First, traditional methods typically employ fixed sensor noise models, which cannot adapt to the dynamic changes in metallic interference intensity over time and space. This embodiment achieves real-time assessment of environmental interference by online sensing and quantification of metallic interference intensity, generating a metallic interference factor. This "sensing-assessment" mechanism enables the system to understand and respond to interference based on actual environmental conditions, rather than preset static parameters, thereby improving environmental adaptability.

[0048] Secondly, this embodiment dynamically configures the weights of the observation factors corresponding to the LiDAR and visual sensors based on the metal interference factor. This means that in areas with severe metal interference, the weight of interfered data is reduced, while the weight of reliable data is retained. For example, in the example above, when the robot enters a metal pipe area, the weights of the LiDAR and visual sensors are dynamically adjusted, effectively suppressing the negative impact of errors such as "flying points" and "pseudo-planes" on the localization results. In contrast, in dense metal areas, traditional methods suffer from point cloud quality degradation and feature extraction failure, leading to a sharp increase in localization errors or even localization loss. The dynamic weight adjustment strategy of this embodiment effectively solves the limitations of traditional fixed weight fusion, ensuring robustness of localization in interference areas.

[0049] Furthermore, in this embodiment, the dynamically adjusted sensor observation factors and the observation factors corresponding to the undisturbed inertial measurement data are input together into the factor graph model for optimization. The inertial measurement data serves as a stable fundamental constraint, and combined with weighted LiDAR and vision data, the optimization process can more accurately estimate the robot's pose. In the example above, even if the LiDAR and vision data are disturbed in the metal area, the inertial measurement data can still provide reliable pose information, preventing complete positioning failure. This tightly coupled multi-sensor fusion strategy combines the advantages of each sensor and achieves complementary strengths in a disturbed environment, thereby obtaining a high-precision optimal pose estimation result.

[0050] In summary, this embodiment, through a closed-loop framework of "perception-evaluation-adjustment," achieves online quantification, dynamic weight adjustment, and robust optimization of metal interference, fundamentally solving the problem of decreased positioning and navigation performance in metal-intensive industrial scenarios using traditional methods. This approach not only improves the positioning accuracy and robustness of mobile robots in complex environments but also provides more reliable technical support for applications such as industrial inspection.

[0051] In some embodiments, the specific expression for the intensity feature is: ; ; ; in, The intensity anomaly degree corresponding to the intensity characteristics of the j-th local region at time t (reflects the anomaly of the point cloud reflection intensity distribution; the reflection intensity of metal surfaces is usually significantly higher than that of non-metal surfaces and is concentrated). The calculation formula makes the contribution of points with high intensity and deviating from the overall distribution greater. Let be the number of point clouds in the j-th local region at time t. Let be the reflection intensity at the i-th point in the j-th local region at time t. Let be the average reflection intensity of the j-th local region at time t. Let be the standard deviation of the reflection intensity of the j-th local region at time t. Indicated as about Indicator functions, Let t be the metal intensity threshold of the j-th local region at time t (for example, it can be 1.5 times the average environmental intensity).

[0052] This application's solution calculates an intensity anomaly degree corresponding to intensity characteristics, specifically designed for the characteristics of metal surface reflection intensity in industrial inspection scenarios. Specifically, the solution first obtains the number of point clouds in the j-th local region at time t and the reflection intensity of each point i within that region. Based on this, the average reflection intensity and standard deviation of the reflection intensity in that local region are calculated. The average reflection intensity provides a benchmark reference for the current local region's reflection intensity level, allowing it to adapt to differences in the overall reflection intensity level under different environments. The standard deviation of the reflection intensity characterizes the dispersion of the local region's reflection intensity, helping to distinguish the intensity distribution characteristics of metallic and non-metallic regions. When calculating the intensity anomaly degree, this solution introduces a Gaussian exponential term. This design amplifies the contribution of high-intensity points deviating from the average intensity to the final anomaly degree, which aligns with the characteristic of metallic point intensity deviating from ordinary regions, making the intensity anomaly characteristics more prominent. Simultaneously, by using an indicator function to statistically analyze points whose reflection intensity exceeds the metal intensity threshold of the j-th local region at time t, interference from low-intensity ordinary points is effectively filtered out, reducing the impact of invalid information on the intensity anomaly degree calculation. This calculation method, which combines local statistical information, Gaussian weighting, and threshold filtering, ensures that the final intensity anomaly score accurately reflects the possibility of metal interference in local areas. Through these techniques, the proposed solution accurately quantifies the intensity anomaly level of local point cloud regions, providing accurate and reliable input for subsequent metal interference factor calculations and guaranteeing the accuracy of metal interference assessment. This precise intensity anomaly score calculation, as one of the multi-dimensional features, can be fused with other features (such as geometric features, density features, and stability features) to jointly generate a metal interference factor that quantitatively characterizes the degree of environmental interference in each local area. This metal interference factor is then used to dynamically adjust the observation factor weights corresponding to different sensors deployed on the mobile robot, thereby reducing the weight of data from the interfered area during factor graph model optimization, ultimately obtaining a more reliable optimal pose estimation result for the mobile robot, significantly improving the positioning and navigation accuracy and robustness of the mobile robot in complex industrial environments.

[0053] In some embodiments, the specific expression of the geometric feature is as follows: ; ; in, Let be the flatness anomaly degree corresponding to the geometric features of the j-th local region at time t (the reflection of a metal mirror may form a seemingly smooth "pseudo-plane", but its normal vector distribution is different from that of a real plane (such as a wall or the ground). Let t be the number of valid point clouds in the j-th local region at time t (i.e., the total number of points involved in the calculation). Let be the direction of the maximum variance of the i-th point in the j-th local region at time t. Let be the direction of the intermediate variance of the i-th point in the j-th local region at time t. Let be the direction of minimum variance (i.e., normal vector direction) of the i-th point in the j-th local region at time t. This is obtained by performing principal component analysis (PCA) decomposition on the covariance matrix of the neighborhood centered at the i-th point in the j-th local region. The three eigenvalues ​​(i.e., normal vector direction) are then used to decompose this direction. , and These characteristics represent the degree of dispersion of the point's distribution in the three principal directions. This is a preset minimum value (to prevent division by zero).

[0054] This application's scheme introduces flatness anomaly as a geometric feature to accurately distinguish between real planes and pseudo-planes formed by reflections from metallic mirrors. Specifically, the scheme first performs principal component analysis on the neighborhood point cloud of each point i within each local region of the scanned point cloud data. Through principal component analysis, three mutually orthogonal eigenvectors and their corresponding eigenvalues ​​are obtained, i.e. , and These eigenvalues ​​characterize the degree of dispersion of the point cloud in the neighborhood of a given point along different principal directions. For the real plane, the point cloud primarily extends in one direction, therefore... It will be significantly larger than and ,and Approaching zero, at this time The value will approach zero. However, for pseudo-planes or noisy point clouds formed by metallic mirror reflection, the point cloud distribution may exhibit isotropic or irregular characteristics, leading to... and The values ​​are relatively close, making The value approaches 1. By averaging this ratio across all valid point clouds within the local area, the flatness anomaly of that local area can be obtained. This flatness anomaly accurately captures the geometric differences between the metal pseudo-plane and the real plane, thus providing a reliable geometric dimension for subsequently generating metal interference factors to quantitatively characterize the degree of environmental interference in each local area. This refined geometric feature extraction method enables mobile robots to more accurately perceive and assess metal interference in the environment, thereby optimizing sensor data fusion strategies and improving positioning and navigation accuracy in complex industrial scenarios.

[0055] In some embodiments, the specific expression for the density feature is: ; ; ; in, Let t be the point density anomaly degree corresponding to the density feature of the j-th local region at time t (specular reflection may cause the laser beam to be reflected to other places, making the effective point cloud actually received on the metal surface sparse). The closer the value is to 1, the greater the difference between the local point cloud density and the global point cloud density in the j-th local region. In other words, the sparser the local point cloud, the greater the possibility that the point cloud is affected by metal interference. Indicated as to Take the maximum value. Let be the average point density of the j-th local region at time t. Let be the global average point density at time t. Let be the number of point clouds in the j-th local region at time t. Let be the volume of the local bounding box of the j-th local region at time t. Let be the total number of points in the global point cloud at time t. Let be the bounding box volume of the global point cloud at time t.

[0056] This application's solution calculates the point density anomaly by comparing the point cloud density in a local region with the average global point cloud density in the current frame. Specifically, in a mobile robot localization and navigation method, multi-source sensor data from the mobile robot in the current environment is first acquired, including scanned point cloud data. Subsequently, the scanned point cloud data is divided into several local regions, and multi-dimensional features of each local region are extracted, including the density feature. When the laser beam is reflected elsewhere due to mirror reflection from a metal surface, the effective point cloud actually received by the metal surface becomes sparse. This solution calculates the local average point density and the global average point density, and uses a formula to quantitatively characterize this point cloud sparseness anomaly. The closer the value is to 1, the greater the difference between the local point cloud density and the global point cloud density; that is, the sparser the local point cloud, the greater the likelihood of metal interference. This point density anomaly, as one of the multi-dimensional features, is fused to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local area. Based on this metal interference factor, the observation factors corresponding to different sensors deployed on the mobile robot are dynamically weighted, and the covariance matrix of the observation factors corresponding to the scanned point cloud data and the image data is adjusted in real time to reduce the weight of the data in the interfered area. Finally, the weight adjustment results and the observation factors corresponding to the inertial measurement data are input into a preset factor graph model, and the factor graph model is optimized to obtain the optimal pose estimation result of the mobile robot, which is then used to control the mobile robot for localization and navigation. In this way, this scheme can effectively perceive and quantify the point cloud sparsity problem caused by metal interference, providing more accurate input for subsequent sensor fusion and pose estimation, thereby improving the robustness of the mobile robot's localization and navigation in complex industrial environments.

[0057] In some embodiments, the specific expression for the stability feature is: ; in, The temporal stability of the j-th local region at time t is the instability of the point cloud information caused by the sudden change in the position of the metal reflection point in the world coordinate system due to small changes in the viewpoint between consecutive frames. The larger the value, the worse the time-domain stability. The number of valid past frames at time t (i.e., the total number of past frames involved in the calculation). For the pose transformation of the mobile robot from the nk-th valid past frame to the n-th valid past frame, Let i be the coordinates of the i-th point in the j-th local region of the nk-th valid past frame. The coordinates are the coordinates of the i-th point in the j-th local region of the n-th valid past frame.

[0058] The stability feature calculation method proposed in this application aims to quantify the pose changes of a local point cloud across consecutive frames. Specifically, for each point in the j-th local region at time t, its coordinates in the current frame (the n-th frame) are compared with those of the point in the past... The coordinates in each of the nk-th valid frames are compared. To eliminate the influence of the mobile robot's own motion on the point cloud position, the pose transformation of the mobile robot from the nk-th valid past frame to the n-th valid past frame is first used to transform the point coordinates in the historical frames to the coordinate system of the current frame. Then, the Euclidean distance between the transformed historical point coordinates and the current frame point coordinates is calculated, and the distances between all points and frames involved in the calculation are averaged to obtain temporal stability. This calculation method can accurately capture the drastic jump characteristics of metal reflection points in the world coordinate system due to small changes in viewpoint. When there is metal interference in a local area, the relative position of its point cloud between consecutive frames will exhibit significant instability, leading to… The value increases significantly. Conversely, in non-metallic regions, the temporal stability of the point cloud is higher. The value is relatively small. By providing this quantified temporal stability, this scheme provides a crucial input for the subsequent generation of the metal interference factor. As one of the multi-dimensional features, the metal interference factor, when fused with other features (such as intensity, geometric, and density features), can more comprehensively and accurately characterize the intensity of metal interference in the local environment. This accurate metal interference factor is the basis for dynamically configuring the weights of sensor observation factors, enabling the factor graph model to effectively reduce the weight of data in the interfered area during optimization, suppress the influence of unreliable sensor data, and ultimately improve the positioning and navigation accuracy and robustness of mobile robots in complex industrial environments. This solves the problems of point cloud quality degradation, feature extraction failure, and the limitations of fixed-weight fusion caused by metal interference in traditional methods.

[0059] In some embodiments, the specific steps in step S3 include: S31. Calculate the metal interference factor according to the following formula: ; ; in, Let be the metal interference factor of the j-th local region at time t. A value close to 1 indicates strong metallic interference, while a value close to 0 indicates weak interference. Represented as transpose, To and The weight vector corresponding to the dimension (corresponding to) (weight) Let be the feature vector of the j-th local region at time t. For the preset bias parameters ( and Both can be obtained by training logistic regression on a real-world point cloud dataset labeled with "metal" / "non-metal" regions. Indicated as to transpose, Let be the intensity anomaly degree corresponding to the intensity feature of the j-th local region at time t. Let be the flatness anomaly degree corresponding to the geometric features of the j-th local region at time t. Let be the point density anomaly degree corresponding to the density feature of the j-th local region at time t. Let be the temporal stability corresponding to the stability characteristics of the j-th local region at time t.

[0060] This application's solution addresses the challenge of uniformly quantifying the degree of metallic interference by introducing a fusion processing mechanism based on logistic regression and the Sigmoid activation function. Specifically, the solution first integrates four features reflecting different aspects of metallic interference—intensity anomaly, flatness anomaly, point density anomaly, and temporal stability—extracted from scanned point cloud data into a single feature vector. This feature vector comprehensively captures various anomalies that metallic interference may exhibit in point cloud data, providing rich and comprehensive information for subsequent quantization. Subsequently, this feature vector is transposed and multiplied with a pre-trained weight vector, and a bias parameter is added to form a linear combination. The result of this linear combination is then non-linearly mapped using the Sigmoid function, ultimately generating a scalar value between 0 and 1, i.e., the metallic interference factor. This mapping method enables… It can intuitively represent the probability or intensity of metal interference in a current local area, where A value approaching 1 indicates strong metallic interference, while a value approaching 0 indicates weak interference. The ingenuity of this scheme lies in fusing four originally dispersed and physically distinct features through a learnable, unified mathematical model, thereby generating an interference factor with a clear physical meaning and quantization range. This allows for dynamic configuration of sensor observation factor weights in subsequent settings, based on… The values ​​are adjusted in real time and with precision, along with the covariance matrix of the observation factors corresponding to the scanned point cloud data and the observation factors corresponding to the image data, to reduce the weight of the data in the interfered area. This mechanism enables the entire positioning and navigation method to perceive metallic interference in the environment online and adaptively adjust the confidence level of the sensor data according to the interference intensity, thereby effectively suppressing the negative impact of metallic interference on the positioning and navigation accuracy. In this way, the solution of this application provides a specific and operable implementation path for the above step S3, "fusing the multi-dimensional features to generate a metallic interference factor for quantitatively characterizing the degree of environmental interference in each local area," making the abstract fusion process concrete and providing an accurate and reliable quantitative basis for subsequent dynamic weight configuration.

[0061] In some embodiments, the specific steps in step S6 include: S61. Calculate the optimal pose estimation result according to the following formula: ; ; ; ; ; in, This is the optimal pose estimation result. Represents the pose state variable at all times. (For example Minimize optimization. Represented as based on right Find the square of the Mahalanobis distance. The residuals of the observation factors corresponding to the inertial measurement data. Let be the pose state variable of the mobile robot at time t. Let be the pose state variable of the mobile robot at time t+1. The inertial measurement data at time t, This is the covariance matrix of the observation factors corresponding to the inertial measurement data. Indicated as about The Huber robust kernel function (used to further suppress possible false matches), Represented as based on right Find the square of the Mahalanobis distance. This is to scan the residuals of the observation factors corresponding to the point cloud data (e.g., based on NDT or the distance from a point to a surface). Let be the metal interference factor of the j-th local region at time t. Let be the scanned point cloud data of the j-th local region at time t. The covariance matrix of the observation factors corresponding to the scanned point cloud data in the weight adjustment result (the reliability of lidar decreases under metal interference, so its weight should be reduced (i.e., the covariance should be increased), when When = 0, use the nominal covariance; As the value increases, the covariance grows exponentially, and the corresponding information matrix weights decrease. The key innovation lies in the fact that its covariance is dynamically adjusted point-by-point or region-by-region based on local calculations. Represented as based on right Find the square of the Mahalanobis distance. The residuals of the observation factors corresponding to the image data. Let be the image data of the j-th local region at time t. This is the covariance matrix of the observation factors corresponding to the image data in the weighted adjustment results (visual information may be degraded in metallic areas due to reflection, but remains reliable in non-metallic areas). The first nominal (interference-free) covariance matrix is ​​preset (provided by the lidar manufacturer or obtained through offline calibration). The preset sensitivity coefficient ( (e.g., a value of 2.0). It is represented as the inverse of the covariance matrix of the observation factors corresponding to the scanned point cloud data in the weighted adjustment results. Represented as The inverse matrix, The second nominal (interference-free) covariance matrix is ​​preset (provided by the vision sensor manufacturer or obtained through offline calibration). This is the preset maximum magnification (e.g., a value of 1.5). Set the steepness of the preset change curve (e.g., a value of 3.0). The function ensures smooth adjustment. It is represented as the inverse of the covariance matrix of the observed factors corresponding to the image data in the weighted adjustment result. Represented as The inverse matrix.

[0062] The proposed solution achieves accurate pose estimation for a mobile robot by constructing a factor graph optimization objective that dynamically adapts to environmental disturbances. This solution integrates observation factors corresponding to inertial measurement data, scanned point cloud data, and image data into a unified optimization framework. Notably, the observation factors corresponding to inertial measurement data remain unaffected by metallic interference, maintaining a fixed covariance matrix and providing a stable temporal continuity constraint for pose estimation, effectively preventing pose estimation drift.

[0063] To address the issue of metal interference affecting scanned point cloud and image data, this scheme introduces a metal interference factor to dynamically adjust the covariance matrix of its observation factors. Specifically, for lidar observations, the covariance matrix grows exponentially with respect to the metal interference factor; that is, the stronger the metal interference, the lower the weight of the lidar observation. This exponential adjustment method can quickly respond to changes in interference intensity and effectively suppress the negative impact of lidar data from interfered areas on pose estimation. Simultaneously, the introduction of the Huber robust kernel function further enhances the optimization process's resistance to lidar mismatches, improving the system's robustness.

[0064] For image data observations, the covariance matrix is ​​adjusted using a smooth curve incorporating the tanh function, ensuring the smoothness of weight adjustments and avoiding optimization instability caused by abrupt weight changes. This dynamic adjustment mechanism allows visual observations to maintain a high weight in non-metallic interference areas while appropriately reducing the weight in metallic reflective areas, thus fully utilizing the effective portion of visual information.

[0065] Through the aforementioned dynamic weight configuration, this scheme can adaptively adjust the reliability of different sensor observations based on the degree of metal interference in a local area. This allows reliable sensor data to play a greater role in optimization, while unreliable data affected by interference is effectively suppressed. This closed-loop strategy of "perception-evaluation-adjustment" is closely integrated with the processes of acquiring multi-source sensor data, extracting multi-dimensional features, generating metal interference factors, and dynamically configuring observation factor weights in steps S1 to S5, forming a complete and adaptive positioning and navigation system. This system can perceive metal interference in the environment online and adjust the sensor fusion strategy in real time accordingly, thereby achieving high-precision and robust mobile robot positioning and navigation in complex industrial scenarios.

[0066] Please refer to Figure 2 , Figure 2 This is a mobile robot positioning and navigation device according to some embodiments of the present invention (the mobile robot positioning and navigation device adopts the mobile robot positioning and navigation method of the above embodiments, and the specific process is referred to the corresponding steps above). The mobile robot positioning and navigation device is integrated into the back-end control device in the form of a computer program, including: The first acquisition module 100 is used to acquire multi-source sensor data of the mobile robot in the current environment; the multi-source sensor data includes scan point cloud data, image data and inertial measurement data; The extraction module 200 is used to divide the scanned point cloud data into several local regions and extract the multi-dimensional features of each local region; The generation module 300 is used to fuse multi-dimensional features and generate a metal interference factor to quantitatively characterize the degree of environmental interference in each local area. The configuration module 400 is used to dynamically configure the weights of the observation factors corresponding to different sensors deployed on the mobile robot based on the metal interference factor, and obtain the weight adjustment result for the metal interference factor. The dynamic weight configuration includes: adjusting the covariance matrix of the observation factors corresponding to the scanned point cloud data and the observation factors corresponding to the image data in real time according to the value of the metal interference factor, so as to reduce the weight of the data in the interfered area. The second acquisition module 500 is used to acquire the observation factors corresponding to the inertial measurement data; The solver module 600 is used to input the weight adjustment results and the observation factors corresponding to the inertial measurement data into the preset factor graph model, and optimize the factor graph model to obtain the optimal pose estimation result of the mobile robot. The control module 700 is used to control the mobile robot to perform localization and navigation based on the optimal pose estimation result.

[0067] Please refer to Figure 3 , Figure 3This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The present invention provides an electronic device 13, including: a processor 1301 and a memory 1302. The processor 1301 and the memory 1302 are interconnected and communicate with each other via a communication bus 1303 and / or other forms of connection mechanism (not shown). The memory 1302 stores computer-readable instructions executable by the processor 1301. When the electronic device is running, the processor 1301 executes the computer-readable instructions to perform the method in any optional implementation of the above embodiments, thereby achieving the following functions: acquiring multi-source sensor data of a mobile robot in the current environment; the multi-source sensor data includes scanned point cloud data, image data, and inertial measurement data; dividing the scanned point cloud data into several local regions and extracting multi-dimensional features of each local region. The system employs a multi-dimensional feature fusion process to generate a metal interference factor that quantitatively characterizes the degree of environmental interference in each local area. Based on the metal interference factor, the observation factors corresponding to different sensors deployed on the mobile robot are dynamically weighted to obtain the weight adjustment results for the metal interference factor. The dynamic weight configuration includes: adjusting the covariance matrix of the observation factors corresponding to the scanned point cloud data and the observation factors corresponding to the image data in real time according to the value of the metal interference factor to reduce the weight of the data in the interfered area; obtaining the observation factors corresponding to the inertial measurement data; inputting the weight adjustment results and the observation factors corresponding to the inertial measurement data into a preset factor graph model, and optimizing the factor graph model to obtain the optimal pose estimation result of the mobile robot; and controlling the mobile robot to perform localization and navigation based on the optimal pose estimation result.

[0068] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it performs the method in any optional implementation of the above embodiments to achieve the following functions: acquiring multi-source sensor data of a mobile robot in the current environment; the multi-source sensor data includes scanned point cloud data, image data, and inertial measurement data; dividing the scanned point cloud data into several local regions and extracting multi-dimensional features of each local region; fusing the multi-dimensional features to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local region; dynamically configuring the weights of observation factors corresponding to different sensors deployed on the mobile robot based on the metal interference factor to obtain a weight adjustment result for the metal interference factor; the dynamic weight configuration includes: adjusting the covariance matrix of the observation factors corresponding to the scanned point cloud data and the observation factors corresponding to the image data in real time according to the value of the metal interference factor to reduce the weight of the data in the interfered area; acquiring the observation factors corresponding to the inertial measurement data; inputting the weight adjustment result and the observation factors corresponding to the inertial measurement data into a preset factor graph model and optimizing the factor graph model to obtain the optimal pose estimation result of the mobile robot; and controlling the mobile robot to perform positioning and navigation based on the optimal pose estimation result.

[0069] The computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0070] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0071] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] Furthermore, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0073] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0074] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for positioning and navigating a mobile robot, characterized by, Includes the following steps: S1. Acquire multi-source sensor data of the mobile robot in the current environment; the multi-source sensor data includes scanned point cloud data, image data, and inertial measurement data; S2. Divide the scanned point cloud data into several local regions and extract multi-dimensional features of each local region; the multi-dimensional features include intensity features, geometric features, density features and stability features; S3. The multi-dimensional features are fused to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local area; S4. Based on the metal interference factor, dynamically configure the weights of the observation factors corresponding to different sensors deployed on the mobile robot to obtain the weight adjustment results for the metal interference factor; The dynamic weight configuration includes: adjusting the covariance matrix of the observation factor corresponding to the scanned point cloud data and the observation factor corresponding to the image data in real time according to the value of the metal interference factor, so as to reduce the weight of the data in the interfered area. S5. Obtain the observation factor corresponding to the inertial measurement data; S6. Input the weight adjustment result and the observation factor corresponding to the inertial measurement data into a preset factor graph model, and optimize the factor graph model to obtain the optimal pose estimation result of the mobile robot; S7. Based on the optimal pose estimation result, control the mobile robot to perform localization and navigation; The specific steps in step S6 include: S61. Calculate the optimal pose estimation result according to the following formula: ； ； ； in, The optimal pose estimation result is... Represents the pose state variable at all times. Perform minimization optimization. Represented as based on right Find the square of the Mahalanobis distance. The residual of the observation factor corresponding to the inertial measurement data. Let be the pose state variable of the mobile robot at time t. Let be the pose state variable of the mobile robot at time t+1. The inertial measurement data at time t, Let be the covariance matrix of the observation factors corresponding to the inertial measurement data. Indicated as about Huber robust kernel function, Represented as based on right Find the square of the Mahalanobis distance. The residual of the observation factor corresponding to the scanned point cloud data. Let be the metal interference factor of the j-th local region at time t. Let be the scanned point cloud data of the j-th local region at time t. The covariance matrix of the observation factors corresponding to the scanned point cloud data in the weight adjustment result. Represented as based on right Find the square of the Mahalanobis distance. The residual of the observation factor corresponding to the image data. Let be the image data of the j-th local region at time t. The covariance matrix of the observation factors corresponding to the image data in the weight adjustment result. The first nominal covariance matrix is ​​preset. The preset sensitivity coefficient, The second nominal covariance matrix is ​​preset. This is the preset maximum magnification. The steepness of the preset change curve.

2. The mobile robot positioning and navigation method according to claim 1, characterized in that, The specific expression for the intensity feature is as follows: ； ； ； in, Let be the intensity anomaly degree corresponding to the intensity feature of the j-th local region at time t. Let be the number of point clouds in the j-th local region at time t. Let be the reflection intensity at the i-th point in the j-th local region at time t. Let be the average reflection intensity of the j-th local region at time t. Let be the standard deviation of the reflection intensity of the j-th local region at time t. Indicated as about Indicator functions, Let t be the metal intensity threshold of the j-th local region at time t.

3. The mobile robot positioning and navigation method according to claim 1, characterized in that, The specific expression for the geometric feature is as follows: ； ； in, Let be the flatness anomaly degree corresponding to the geometric feature of the j-th local region at time t. Let be the number of effective point clouds in the j-th local region at time t. Let be the direction of the maximum variance of the i-th point in the j-th local region at time t. Let be the direction of the intermediate variance of the i-th point in the j-th local region at time t. Let be the direction of minimum variance of the i-th point in the j-th local region at time t. This is the preset minimum value.

4. The mobile robot positioning and navigation method according to claim 1, characterized in that, The specific expression for the density feature is as follows: ； ； ； in, Let be the point density anomaly degree corresponding to the density feature of the j-th local region at time t. Indicated as to Take the maximum value. Let be the average point density of the j-th local region at time t. Let be the global average point density at time t. Let be the number of point clouds in the j-th local region at time t. Let be the volume of the local bounding box of the j-th local region at time t. Let be the total number of points in the global point cloud at time t. Let be the bounding box volume of the global point cloud at time t.

5. The mobile robot positioning and navigation method according to claim 1, characterized in that, The specific expression for the stability feature is: ； in, The temporal stability corresponding to the stability feature of the j-th local region at time t. The number of valid past frames at time t. For the pose transformation of the mobile robot from the nk-th valid past frame to the n-th valid past frame, Let i be the coordinates of the i-th point in the j-th local region of the nk-th valid past frame. The coordinates are the coordinates of the i-th point in the j-th local region of the n-th valid past frame.

6. The mobile robot positioning and navigation method according to claim 1, characterized in that, The specific steps in step S3 include: S31. Calculate the metal interference factor according to the following formula: ； ； in, Let be the metal interference factor of the j-th local region at time t. Represented as transpose, To and The weight vector corresponding to the dimension Let be the feature vector of the j-th local region at time t. The preset bias parameters, Indicated as to transpose, Let be the intensity anomaly degree corresponding to the intensity feature of the j-th local region at time t. Let be the flatness anomaly degree corresponding to the geometric feature of the j-th local region at time t. Let be the point density anomaly degree corresponding to the density feature of the j-th local region at time t. Let be the temporal stability corresponding to the stability feature of the j-th local region at time t.

7. A mobile robot positioning and navigation device based on the mobile robot positioning and navigation method as described in any one of claims 1-6, characterized in that, include: The first acquisition module is used to acquire multi-source sensor data of the mobile robot in the current environment; the multi-source sensor data includes scanned point cloud data, image data and inertial measurement data; The extraction module is used to divide the scanned point cloud data into several local regions and extract multi-dimensional features of each local region; The generation module is used to fuse the multi-dimensional features to generate a metal interference factor for quantitatively characterizing the degree of environmental interference in each local area. The configuration module is used to dynamically configure the weights of the observation factors corresponding to different sensors deployed on the mobile robot based on the metal interference factor, so as to obtain the weight adjustment result for the metal interference factor. The dynamic weight configuration includes: adjusting the covariance matrix of the observation factor corresponding to the scanned point cloud data and the observation factor corresponding to the image data in real time according to the value of the metal interference factor, so as to reduce the weight of the data in the interfered area. The second acquisition module is used to acquire the observation factor corresponding to the inertial measurement data; The solution module is used to input the weight adjustment results and the observation factors corresponding to the inertial measurement data into a preset factor graph model, and to optimize and solve the factor graph model to obtain the optimal pose estimation result of the mobile robot. The control module is used to control the mobile robot to perform positioning and navigation based on the optimal pose estimation result.

8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the mobile robot positioning and navigation method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the steps in the mobile robot positioning and navigation method as described in any one of claims 1-6.

Images