Mobile robot positioning method and system based on laser radar scanning information

Abstract

The application discloses a mobile robot positioning method and system based on laser radar scanning information and belongs to the technical field of robot positioning, and aims to solve the technical problem of how to improve the shortcomings of robot particle filtering positioning algorithm in global positioning and following positioning in a large-scale map. The mobile robot is applied to a mobile robot provided with a perception sensor, the perception sensor comprises a mileage meter, an IMU and a laser radar, is used for collecting pose information and environment information, comprises a pre-processing module, a global initialization positioning module and a following positioning module, a scanning matching process based on a multi-resolution grid map and a high-confidence area is adopted, in the robot following positioning module, a high-confidence area and particle genetic mutation are introduced, in the resampling process, a particle is updated by applying a likelihood domain observation model and simultaneously adding a high-confidence area reference, so that the particle set converges to the vicinity of an execution track, the particle convergence effect is good, and the robot following precision is ensured.

Description

Technical Field

[0001] This invention relates to the field of robot positioning technology, specifically to a mobile robot positioning method and system based on lidar scanning information. Background Technology

[0002] In recent years, mobile robots have matured and are widely used in various fields such as industrial inspection, logistics distribution, and cleaning services, primarily to replace manual labor in performing simple, repetitive, and complex tasks. Considering the reliability and safety requirements of mobile robots, enterprises and research institutions often use lidar, which possesses excellent characteristics such as high measurement accuracy and high stability, as the sensing terminal for robot positioning and navigation research.

[0003] In data centers and industrial inspection, mobile robots primarily perform point-to-point operations using path tracking control algorithms. The robots then detect noise, temperature, humidity, meter readings, and other anomalies at designated fixed points. Therefore, the reliability of the end-effector detection module is inextricably linked to the robot's path-following accuracy and point-to-point precision. In logistics and distribution, mobile robots have become a crucial component of production lines, often collaborating with product lines and robotic arms to perform precise assembly and handling tasks. The robot's positioning performance directly impacts the entire collaborative process, thereby affecting production line efficiency. In cleaning services, mobile robots operate in dynamic and noisy environments such as shopping malls, living rooms, and hotels. The effective features of LiDAR-based mobile robots are drastically reduced, leading to erroneous positioning and motion planning. The efficiency of positioning recovery after feature restoration determines the robot's reliability in cleaning, food delivery, and other tasks. Clearly, improving the positioning performance of mobile robots is a catalyst for promoting industrial intelligence and efficiency, and an effective means for the efficient deployment of products across various application scenarios.

[0004] Currently, indoor mobile robot localization based on LiDAR scanning information mostly employs particle filtering algorithms, which offer excellent performance in nonlinear, non-Gaussian systems. While this algorithm is relatively mature and stable, it still has certain limitations. In the global localization process of a mobile robot, it can complete global localization relatively quickly in small-scale maps for subsequent follow-up localization. However, in larger-scale maps, fewer particles cannot efficiently iterate to find the optimal pose, while more particles result in higher time complexity, impacting the robot's real-time requirements. Furthermore, in the follow-up localization process, when the prior environment changes or robot aggroing occurs, particle degradation occurs, and the algorithm cannot effectively recover the robot's localization after resampling, leading to robot navigation task failure.

[0005] How to improve the shortcomings of the robot particle filter localization algorithm in global localization and following localization in large-scale maps is a technical problem that needs to be solved. Summary of the Invention

[0006] The technical objective of this invention is to address the above-mentioned shortcomings by providing a mobile robot localization method and system based on LiDAR scanning information, thereby solving the technical problem of how to improve the shortcomings of the robot particle filter localization algorithm in global localization and following localization in large-scale maps.

[0007] In a first aspect, the present invention provides a mobile robot positioning system based on lidar scanning information, applicable to a mobile robot equipped with sensing sensors, including an odometer, an IMU, and a lidar, for collecting pose information and environmental information. The positioning system includes a preprocessing module, a global initialization positioning module, and a following positioning module.

[0008] The preprocessing module is used to construct a prior grid map based on pose information and environmental information, generate a multi-resolution grid map based on the prior grid map, and divide the multi-resolution grid map into high-confidence and low-confidence regions according to the task information set by the user, thereby obtaining high-confidence regions and low-confidence regions. The task information is used to provide motion control and navigation planning for the mobile robot.

[0009] The global initialization positioning module is used to control the mobile robot to perform a 360-degree rotation in place. It is used to estimate the attitude of the mobile robot's rotation based on the point cloud information collected by the lidar and the pose information collected by the IMU, and to construct a global point cloud map based on the lidar point cloud information in the lidar coordinate system. It is used to perform branch delimitation scanning matching based on the multi-resolution raster map and high confidence region to obtain global positioning.

[0010] The following positioning module is used to perform motion updates based on the attitude information collected by the IMU and the position information collected by the odometry to obtain the predicted pose value. Based on the predicted pose value, the high confidence region, and particle genetic mutation, the LiDAR observation is updated to obtain the observed pose value. Based on the observed pose value, the pose value is resampled to obtain the pose.

[0011] Preferably, the preprocessing module interacts with the user through a preprocessing interface and performs the following:

[0012] Constructing a prior grid map: Adjust the mobile robot to scan the working environment, and construct a prior grid map of a specified resolution based on the environmental information collected by the perception sensors and graph optimization.

[0013] Constructing a multi-resolution raster map: Invalid points in the prior raster map are removed using drawing tools to obtain a preprocessed prior raster map. The length of the search box is calculated by using a sliding window on the prior raster map based on a multiple of its resolution to obtain the multi-resolution raster map.

[0014] Dividing regions into high and low confidence levels: When configuring high and low confidence level division, the user-defined task information is written into a multi-resolution raster map in raster form, and a likelihood domain space based on the trajectory corresponding to the task information is generated. The formula for calculating the likelihood domain space is as follows:

[0015] ,

[0016] in, For free space points in a multi-resolution raster map, These are the points in the trajectory corresponding to the task information. Set the width parameter for the likelihood domain. The distance from a blank grid cell in a multi-resolution raster map to a specified route will be... The likelihood domain space within the range is considered a high-confidence region, while any region exceeding the width of the likelihood domain is considered a low-confidence region with a score of -1.

[0017] Preferably, the global initialization positioning module is used to perform the following:

[0018] Constructing a cloud map of the entire scenic spot: Control the mobile robot to perform a 360-degree rotation in place, record the attitude information of the IMU at a predetermined frequency, and perform attitude estimation based on the timestamps of each point in the laser point cloud and the attitude information of the IMU under each timestamp of linear interpolation. Then, sequentially read the laser point cloud information under the laser radar coordinate system and stitch the laser point cloud information to obtain a cloud map of the entire scenic spot.

[0019] Global initial pose lookup: Global initial pose lookup is performed by scanning and matching based on multi-resolution raster maps and high-confidence regions, or by searching based on full-confidence regions.

[0020] The global initial pose lookup is performed using a scanning and matching method based on multi-resolution raster maps and high-confidence regions, including the following operations:

[0021] L1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0022] L2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. If the proportion of low-confidence areas in any one of the smaller regions exceeds 2 / 3, skip the scan matching calculation for that region and calculate the scan matching score using the following formula:

[0023] ,

[0024] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0025] L3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0026] If the optimal pose is not found through steps L1-L3, the global initialization localization module is used to perform a global initialization pose search by conducting a full confidence region search, including the following operations:

[0027] M1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0028] M2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. For each smaller region, calculate the scan matching score using the following formula:

[0029] ,

[0030] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0031] M3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0032] If the preprocessing module performs high and low confidence region segmentation, the global initialization localization module is used to perform global initial pose lookup based on scanning and matching of a multi-resolution raster map, including the following steps:

[0033] S1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0034] S2. Perform a branching operation on the target raster map. Divide the search space into four small regions by halving the x and y coordinates of the target raster map. For each small region, calculate the scan matching score using the following formula:

[0035] ,

[0036] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0037] S3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0038] Preferably, the following positioning module is used to perform the following:

[0039] Motion update: Based on the attitude information collected by the IMU and the position information collected by the odometry, the motion update of the mobile robot pose is performed based on the EKF algorithm to obtain the predicted value of each pose;

[0040] Observation Update: Taking the predicted pose value, the laser point cloud information collected by lidar, and the minimum resolution grid map in the multi-resolution grid map as input, the pose update value is calculated and output by calculating particle weights through the likelihood domain observation model. The pose update value is then observed and updated based on high confidence regions and particle genetic mutations to obtain the observation value for each pose. A particle set is then constructed based on the observation value for each pose.

[0041] Resampling: Select the pose with the largest observation value from the particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, resampling is performed based on a high-confidence region. The pose located in the high-confidence region is selected and added to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, global random sampling is performed based on high-confidence and low-confidence regions. Randomly select poses and add them to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, the particle with the highest weight in the particle set is selected for output to ensure real-time positioning.

[0042] Preferably, before resampling, the following positioning module performs the following: Particles are divided into two groups based on their weights; one-third of the smaller weighted particle set is selected for particle crossover and mutation to add new unknown particles, calculated using the following formula:

[0043] ,

[0044] ,

[0045] In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

[0046] In a second aspect, the present invention provides a mobile robot localization method based on lidar scanning information, applicable to a mobile robot equipped with sensing sensors, including an odometer, an IMU, and a lidar, for collecting pose information and environmental information. The method is used for localization via a mobile robot localization system based on lidar scanning information as described in any of the first aspects, and the method includes the following steps:

[0047] Preprocessing: A prior grid map is constructed based on pose information and environmental information. A multi-resolution grid map is generated based on the prior grid map. According to the task information set by the user, high and low confidence regions are divided in the multi-resolution grid map to obtain high confidence regions and low confidence regions. The task information is used to provide motion control and navigation planning for the mobile robot.

[0048] Global initialization localization: Control the mobile robot to perform a 360-degree rotation in place. Based on the point cloud information collected by the LiDAR and the pose information collected by the IMU, the attitude of the mobile robot's rotation is estimated. A global point cloud map is constructed based on the LiDAR coordinate system coordinate system point cloud information. Branch boundary scanning and matching are performed based on multi-resolution grid map and high confidence area to obtain global localization.

[0049] Follow-up localization: Motion updates are performed based on attitude information collected by IMU and position information collected by odometry to obtain the predicted pose value. Based on the predicted pose value, high confidence region and particle genetic mutation, lidar observation updates are performed to obtain the observed pose value. Based on the observed pose value, resampling is performed to obtain the pose.

[0050] As a preferred option, the preprocessing includes the following operations:

[0051] Constructing a prior grid map: The remote-controlled mobile robot scans the working environment and constructs a prior grid map of a specified resolution based on the environmental information collected by the perception sensors and graph optimization.

[0052] Constructing a multi-resolution raster map: Invalid points in the prior raster map are removed using drawing tools to obtain a preprocessed prior raster map. The length of the search box is calculated by using a sliding window on the prior raster map based on a multiple of its resolution to obtain the multi-resolution raster map.

[0053] Dividing regions into high and low confidence levels: When configuring high and low confidence level division, the user-defined task information is written into a multi-resolution raster map in raster form, and a likelihood domain space based on the trajectory corresponding to the task information is generated. The formula for calculating the likelihood domain space is as follows:

[0054] ,

[0055] in, For free space points in a multi-resolution raster map, These are the points in the trajectory corresponding to the task information. Set the width parameter for the likelihood domain. The distance from a blank grid cell in a multi-resolution raster map to a specified route will be... The likelihood domain space within the range is considered a high-confidence region, while any region exceeding the width of the likelihood domain is considered a low-confidence region with a score of -1.

[0056] As a preferred method, global initialization positioning includes the following operations:

[0057] Constructing a cloud map of the entire scenic spot: Control the mobile robot to perform a 360-degree rotation in place, record the attitude information of the IMU at a predetermined frequency, and perform attitude estimation based on the timestamps of each point in the laser point cloud and the attitude information of the IMU under each timestamp of linear interpolation. Then, sequentially read the laser point cloud information under the laser radar coordinate system and stitch the laser point cloud information to obtain a cloud map of the entire scenic spot.

[0058] Global initial pose lookup: Global initial pose lookup is performed by scanning and matching based on multi-resolution raster maps and high-confidence regions, or by searching based on full-confidence regions.

[0059] The global initial pose lookup is performed using a scanning and matching method based on multi-resolution raster maps and high-confidence regions, including the following operations:

[0060] L1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0061] L2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. If the proportion of low-confidence areas in any one of the smaller regions exceeds 2 / 3, skip the scan matching calculation for that region and calculate the scan matching score using the following formula:

[0062] ,

[0063] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0064] L3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0065] If the optimal pose is not found through steps L1-L3, a global initial pose search is performed using a full confidence region search, including the following operations:

[0066] M1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0067] M2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. For each smaller region, calculate the scan matching score using the following formula:

[0068] ,

[0069] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0070] M3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0071] If high and low confidence region segmentation was not performed during the initial preprocessing, then global initial pose lookup is performed based on scanning and matching of a multi-resolution raster map, including the following steps:

[0072] S1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0073] S2. Perform a branching operation on the target raster map. Divide the search space into four small regions by halving the x and y coordinates of the target raster map. For each small region, calculate the scan matching score using the following formula:

[0074] ,

[0075] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0076] S3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0077] As a preferred method, following positioning includes the following operations:

[0078] Motion update: Based on the attitude information collected by the IMU and the position information collected by the odometry, the motion update of the mobile robot pose is performed based on the EKF algorithm to obtain the predicted value of each pose;

[0079] Observation Update: Taking the predicted pose value, the laser point cloud information collected by lidar, and the minimum resolution grid map in the multi-resolution grid map as input, the pose update value is calculated and output by calculating particle weights through the likelihood domain observation model. The pose update value is then observed and updated based on high confidence regions and particle genetic mutations to obtain the observation value for each pose. A particle set is then constructed based on the observation value for each pose.

[0080] Resampling: Select the pose with the largest observation value from the particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, resampling is performed based on the high-confidence region. The pose located in the high-confidence region is selected and added to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, global random sampling is performed based on the high-confidence region and the low-confidence region. Randomly select poses and add them to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, the particle with the highest weight in the particle set is selected for output to ensure real-time positioning.

[0081] As a preferred method, before resampling, the following operation is performed: The particles are divided into two groups based on their weights. One-third of the smaller weighted particle set is selected for particle crossover and mutation to add new unknown particles. The calculation formula is as follows:

[0082] ,

[0083] ,

[0084] In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

[0085] The mobile robot localization method and system based on lidar scanning information of the present invention have the following advantages:

[0086] (1) In the global initialization pose localization module, it forms a constraint with the robot's following trajectory. In practical applications, the robot always performs tasks according to the given trajectory or task point. Similarly, the robot spends most of its time in the above-mentioned area. Therefore, these areas are considered to be high-confidence areas. When the robot performs global localization, it adopts a scanning and matching process based on multi-resolution grid map and high-confidence areas. This improves the problems of poor global initialization localization effect, low efficiency, and inability to solve robot kidnapping in traditional mobile robot localization methods in large-scale and complex environments. The particle convergence speed is enhanced, and the robot's scene adaptability is improved.

[0087] (2) In the robot following and localization module, the ideas of high confidence region and particle genetic mutation are introduced. During the resampling process, the high confidence region reference is added while the likelihood domain observation model is updated to update the particles, so that the particle set converges to the vicinity of the execution trajectory. The particle convergence effect is good and the robot following accuracy is also guaranteed.

[0088] (3) To enhance the robustness of the original algorithm, crossover and mutation of the small weighted particle set before resampling are performed to ensure the diversity of the particle set. Attached Figure Description

[0089] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0090] The invention will be further described below with reference to the accompanying drawings.

[0091] Figure 1 This is a flowchart illustrating the workflow of a mobile robot positioning system based on lidar scanning information, as described in Example 1.

[0092] Figure 2 This is a user task setting diagram based on a grid map in a mobile robot positioning system based on LiDAR scanning information, as shown in Example 1.

[0093] Figure 3 Example 1 shows a high and low confidence region division map based on a grid map in a mobile robot positioning system based on LiDAR scanning information. Detailed Implementation

[0094] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0095] This invention provides a mobile robot localization method and system based on lidar scanning information, which addresses the technical problem of how to improve the shortcomings of robot particle filter localization algorithm in global localization and following localization in large-scale maps.

[0096] Example 1:

[0097] This invention discloses a mobile robot positioning system based on lidar scanning information, applicable to mobile robots equipped with sensing sensors, including an odometer, an IMU, and a lidar, used to collect pose and environmental information. In this embodiment, the positioning system includes a preprocessing module, a global initialization positioning module, and a following positioning module.

[0098] The preprocessing module is used to construct a prior grid map based on pose information and environmental information, generate a multi-resolution grid map based on the prior grid map, and divide the multi-resolution grid map into high-confidence and low-confidence regions according to the task information set by the user, thereby obtaining high-confidence regions and low-confidence regions. The task information is used to provide motion control and navigation planning for the mobile robot.

[0099] The preprocessing module interacts with the user through a preprocessing interface and performs the following:

[0100] (1) Constructing a prior grid map: The remote-controlled mobile robot scans the working environment and constructs a prior grid map of a specified resolution based on the environmental information collected by the sensor and graph optimization.

[0101] (2) Constructing a multi-resolution raster map: Remove invalid points from the prior raster map using drawing tools to obtain a preprocessed prior raster map. Based on the multiple of the resolution of the prior raster map as the length of the search box, perform sliding window calculation on the prior raster map to obtain a multi-resolution raster map.

[0102] (3) Dividing high and low confidence regions: When configuring high and low confidence region division, the user-defined task information is written into a multi-resolution raster map in raster form, and a likelihood domain space based on the trajectory corresponding to the task information is generated. The formula for calculating the likelihood domain space is as follows:

[0103] ,

[0104] in, For free space points in a multi-resolution raster map, These are the points in the trajectory corresponding to the task information. Set the width parameter for the likelihood domain. The distance from a blank grid cell in a multi-resolution raster map to a specified route will be... The likelihood domain space within the range is considered a high-confidence region, while any region exceeding the width of the likelihood domain is considered a low-confidence region with a score of -1.

[0105] In this embodiment, the preprocessing module is a collaborative unit that interacts with the user. It fuses and processes visual information such as the given prior map and user-defined tasks to enrich the robot's localization reference information, thereby improving the robot's localization efficiency. When a mobile robot based on LiDAR scanning information performs operations, it needs to be controlled by a remote controller or manually pushed to collect attitude and environmental information from sensors such as odometers, IMUs, and LiDAR, thus completing the environmental map construction process, i.e., SLAM. To ensure that the robot stably completes path planning and fixed-point operations according to the intended route, the deployment personnel often add task settings and auxiliary information to the obtained prior map, including charging points for returning to the dock, robot pending work points, key work point task points, virtual walls, and set paths. The preprocessing module consists of two parts: multi-resolution raster map generation and high / low confidence region division. Generating the multi-resolution raster map is a computationally complex process with high time complexity. Therefore, it is necessary to calculate the current environment map before task execution to provide low-time complexity queries for subsequent global localization and follow-up localization, ensuring the real-time performance of the algorithm. The high / low confidence region division part divides the prior raster map into high-confidence regions and low-confidence regions based on the task settings and auxiliary information added by the user. High-confidence regions represent regions where the robot has a higher probability of being present, while low-confidence regions represent regions where the robot is less likely to appear. In practical applications, it has been found that the robot can form a high degree of constraint with the execution trajectory, task points, and auxiliary task points during normal operation. Even if the robot loses localization due to occlusion or sensor disconnection, it usually stops near the task setting information.

[0106] As a specific implementation of the preprocessing module, this module is used to perform the following operations:

[0107] Step A1: Constructing the Prior Environment Map

[0108] Due to the needs of positioning and navigation, environmental visualization, and user interaction, before performing the task, it is necessary to control a mobile robot equipped with sensors such as LiDAR to scan the working environment, thereby completing the construction of a prior environment map based on graph optimization and obtaining a grid map of a specified resolution. The method described in this paper selects a grid resolution of 5 cm / pixel.

[0109] Step A2: Multi-resolution raster map construction

[0110] To minimize the impact of the prior environment map on scanning matching, invalid points such as noise and specks in the raster map are first erased using drawing tools. Since the raster map is a reflection of the real environment, the higher its resolution and accuracy, the more accurate the environment description. However, scanning matching in this environment presents problems such as large search scale and difficulty in matching. To improve the efficiency and accuracy of scanning matching in subsequent modules, this step calculates sliding windows based on the original 5cm raster map obtained in step A1, with search box lengths of 10cm, 20cm, 40cm, 80cm, 160cm, and 320cm respectively. The window contains a specified number of 5cm grates, and all grates are covered by the maximum grayscale value within the window, thus obtaining 7 layers of raster map resolutions of 5cm, 10cm, 20cm, 40cm, 80cm, 160cm, and 320cm.

[0111] Step A3: Division of high and low confidence regions

[0112] In actual robot operation, it is often necessary to manually provide information such as the task start point, task end point, special points, and specified path. This step begins with the operator specifying the above information on the host computer interface, such as... Figure 2 As shown, the red dots represent the robot's designated task points in the entire environment, which are mostly fixed workstations used for assembly, inspection, and delivery on the production line. The yellow dots represent the robot's standby points, the return points after the robot has completed its tasks, so that it can wait for the next job. The burgundy dots represent the robot's charging points, used for robot status recovery and material replenishment. The green line represents the trajectory tracked by the robot.

[0113] If the parameters do not specify high and low confidence region division, the aforementioned multi-resolution raster map and the set task information will be sent to the global initialization positioning module and the follow positioning module. Otherwise, the module will generate high and low confidence regions based on the set task information. First, the set route will be written into the map in raster form, and a likelihood domain space based on the trajectory will be generated. The calculation formula is as follows:

[0114] ,

[0115] In the formula, For free space points in a multi-resolution raster map, Set the midpoint of the route for the grid. Set the width parameter for the likelihood domain. This refers to the distance from a blank grid cell in the raster map to a specified route. The closer the cell is to the specified route, the higher its score. The likelihood domain space within the set range is called the high-confidence region, while the score for values ​​exceeding the set width of the likelihood domain is -1, which is the low-confidence region. To minimize time complexity, the entire calculation process is offline; the robot only needs to perform the calculation by looking up a table during localization. Figure 3 The shaded area shown is the high-confidence region of the entire map, while the remaining space is the low-confidence region.

[0116] The global initialization localization module is used to control the mobile robot to perform a 360-degree rotation in place. It is used to estimate the attitude of the mobile robot's rotation based on the point cloud information collected by the LiDAR and the pose information collected by the IMU, and to construct a global point cloud map based on the laser point cloud information in the LiDAR coordinate system. It is used to perform branch delimitation scanning and matching based on the multi-resolution raster map and high confidence area to obtain global localization.

[0117] The global initialization positioning module is used to perform the following:

[0118] Constructing a cloud map of the entire scenic spot: Control the mobile robot to perform a 360-degree rotation in place, record the attitude information of the IMU at a predetermined frequency, and perform attitude estimation based on the timestamps of each point in the laser point cloud and the attitude information of the IMU under each timestamp of linear interpolation. Then, sequentially read the laser point cloud information under the laser radar coordinate system and stitch the laser point cloud information to obtain a cloud map of the entire scenic spot.

[0119] Global initial pose lookup: Global initial pose lookup is performed by scanning and matching based on multi-resolution raster maps and high-confidence regions, or by searching based on full-confidence regions.

[0120] The global initial pose lookup is performed using a scanning and matching method based on multi-resolution raster maps and high-confidence regions, including the following operations:

[0121] L1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0122] L2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. If the proportion of low-confidence areas in any one of the smaller regions exceeds 2 / 3, skip the scan matching calculation for that region and calculate the scan matching score using the following formula:

[0123] ,

[0124] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0125] L3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0126] If the optimal pose is not found through steps L1-L3, the global initialization localization module is used to perform a global initialization pose search by conducting a full confidence region search, including the following operations:

[0127] M1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0128] M2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. For each smaller region, calculate the scan matching score using the following formula:

[0129] ,

[0130] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0131] M3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0132] If the preprocessing module performs high and low confidence region segmentation, the global initialization localization module is used to perform global initial pose lookup based on scanning and matching of a multi-resolution raster map, including the following steps:

[0133] S1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0134] S2. Perform a branching operation on the target raster map. Divide the search space into four small regions by halving the x and y coordinates of the target raster map. For each small region, calculate the scan matching score using the following formula:

[0135] ,

[0136] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0137] S3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0138] In this embodiment, the global initialization localization module is the process of acquiring the global pose in the map coordinate system after observation by the robot's own perception module. Typically, due to factors such as map size and perception range, the robot obtains its global positioning from locations with prior information, such as charging stations or waiting points, to provide accurate prior poses for follow-up localization. However, when the mobile robot is kidnapped or located in an environment lacking prior features, the robot may experience low localization efficiency or even be unable to locate itself. The global initialization localization module optimizes existing localization methods by using a panoramic cloud map with richer environmental features as perception observation information. To improve the efficiency and accuracy of global localization in large-scale and complex environments, it performs scanning matching based on the multi-resolution map and high-confidence areas from the preprocessing module to obtain the global initial pose. However, this matching process still has interference such as measurement noise. Therefore, this module generates a Gaussian pose sequence based on the obtained estimated pose and uses it as the initial input particle for particle filtering or as random particles in the resampling after particle degradation. This changes the particle set from a uniform distribution to a Gaussian distribution, enhancing the algorithm's robustness.

[0139] As a specific implementation of the global initialization positioning module, this module is used to perform the following operations:

[0140] Step B1: Building a Cloud Map of the Entire Scenic Spot by Robot

[0141] In complex, chaotic, and relatively open scenes, single-frame LiDAR data provides limited environmental feature descriptions, making it impossible for the robot to calculate its current position. To maximize point cloud information, this step constructs a panoramic point cloud map for the robot. The robot performs a self-rotation motion at a fixed angular velocity. Since the LiDAR has inherent measurement distortion, motion distortion removal must be performed simultaneously with the LiDAR measurement. Considering the robot's constant rotational angular velocity, an IMU (Induction Mutual Amplifier) ​​is used for distortion correction. This step requires recording the IMU's attitude angle data at a high frequency and linearly interpolating the IMU's attitude information at each timestamp of the LiDAR point cloud to complete the correct attitude estimation, thereby improving the accuracy of the environmental description. After the rotation motion is completed, the LiDAR point cloud information based on the LiDAR coordinate system is sequentially read to complete the stitching motion, constructing a panoramic point cloud map for the robot.

[0142] Step B2: Scanning and matching process based on multi-resolution raster maps and high-confidence regions

[0143] If the aforementioned preprocessing module completes the high and low confidence region division configuration, this step first performs a branching operation on the maximum resolution raster map generated in the preprocessing module, halving the map's x and y coordinates to divide the entire search space into four smallest regions. If the proportion of low confidence regions in any one of these small regions exceeds 2 / 3, the scan matching calculation for this region is skipped. The scan matching score calculation formula is as follows:

[0144] ,

[0145] In the formula This is a bicubic interpolation function used to smooth grayscale information in raster maps. For attitude adjustment variables, This provides information about the laser point measurements. The entire function calculates the maximum score for the cloud covering the entire viewpoint within the occupied grid cell.

[0146] Next, this step will find the search space with the highest score in the raster map of this layer resolution, and then complete a new round of branching operations within the search space range. The above operations are repeated with the raster map of the lower layer resolution until the optimal pose is found.

[0147] If the above steps do not find the optimal pose, a full-confidence region search is performed again to find the global initial pose. If the above preprocessing module did not perform high- and low-confidence region segmentation, this step only performs a scan matching process based on a multi-resolution raster map to calculate the global initial pose.

[0148] The follow-up positioning module is used to update the motion based on the attitude information collected by the IMU and the position information collected by the odometry to obtain the predicted pose value. Based on the predicted pose value, the high confidence region and particle genetic mutation, the lidar observation is updated to obtain the observed pose value. Based on the observed pose value, the pose value is resampled to obtain the pose.

[0149] The follow-up positioning module is used to perform the following:

[0150] Motion update: Based on the attitude information collected by the IMU and the position information collected by the odometry, the motion update of the mobile robot pose is performed based on the EKF algorithm to obtain the predicted value of each pose;

[0151] Observation Update: Taking the predicted pose value, the laser point cloud information collected by lidar, and the minimum resolution grid map in the multi-resolution grid map as input, the pose update value is calculated and output by calculating particle weights through the likelihood domain observation model. The pose update value is then observed and updated based on high confidence regions and particle genetic mutations to obtain the observation value for each pose. A particle set is then constructed based on the observation value for each pose.

[0152] Resampling: Select the pose with the largest observation value from the particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, resampling is performed based on the high-confidence region. The pose located in the high-confidence region is selected and added to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, global random sampling is performed based on the high-confidence region and the low-confidence region. Randomly select poses and add them to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, the particle with the highest weight in the particle set is selected for output to ensure real-time positioning.

[0153] Before resampling, the following localization module performs the following: Particles are divided into two groups based on their weights. One-third of the smaller weighted particle set is selected for particle crossover and mutation to add new unknown particles. The calculation formula is as follows:

[0154] ,

[0155] ,

[0156] In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

[0157] In this embodiment, the following positioning module is the process of calculating the robot's real-time pose in the map coordinate system by continuously moving, predicting, observing and updating the initial pose given by the global initialization positioning module. This module mainly consists of three units: mobile robot pose and motion update based on EKF, lidar observation update based on high-confidence regions and particle genetic mutation, and resampling based on high-confidence regions. The pose and motion update unit uses the robust and high-precision EKF algorithm to calculate pose integration, providing more reliable initial information for subsequent observation update units. The lidar observation update unit based on high-confidence regions and particle genetic mutation uses the high-confidence region formed by the charging point, undetermined work site, and set path from the preprocessing module as a necessary condition for particle weight observation update, improving both robot positioning accuracy and computational efficiency. This unit also introduces a genetic mutation algorithm to solve the particle degradation problem inherent in the particle filtering algorithm, further improving robot positioning robustness. When particle degradation occurs, the particle filtering algorithm involves a full-map random resampling process. When the map scale is large or the number of particles is small, the robot positioning calculation efficiency and accuracy cannot be guaranteed. To address this issue, a resampling unit based on high-confidence regions is introduced into the follow-up positioning module.

[0158] As a specific implementation of the follow-up positioning module, this module is used to perform the following operations:

[0159] Step C1: EKF-based pose and motion update of the mobile robot

[0160] When a mobile robot navigates with a known global initial pose, the IMU attitude and Odom position information are used to calculate the robot's trajectory using the EKF algorithm. This reduces the input error of the localization algorithm and enhances the robustness of the trajectory calculation method. Then, the absolute pose prediction value of the trajectory calculation, the current frame LiDAR point cloud, and the prior obstacle grid map are input into the particle filter to expand and calculate the value near the true pose.

[0161] Step C2: LiDAR observation update based on high-confidence region and particle genetic mutation

[0162] This module changes the original particle filtering algorithm's method of calculating particle weights through a likelihood domain observation model. It simultaneously applies a high-confidence region to modify particle weights. After resampling, more particles converge to the tracking trajectory, creating a strong constraint between the robot and the tracking trajectory. This ensures the robot's positioning accuracy and rapid particle convergence during path tracking. Similarly, to mitigate particle degradation and enhance the robustness of particle filtering, this module applies the concept of genetic mutation to increase diversity while maintaining particle consistency. Before resampling, particles are divided into two groups based on their weights. One-third of the smaller-weighted particle set is selected for particle crossover and mutation to add new unknown particles. The calculation formula is as follows:

[0163] ,

[0164] ,

[0165] In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

[0166] Example 2:

[0167] This invention discloses a mobile robot localization method based on lidar scanning information, applicable to a mobile robot equipped with sensing sensors, including an odometer, an IMU, and a lidar, used to collect pose information and environmental information. The method in this embodiment is used for localization via the localization system disclosed in Embodiment 1, and includes the following steps:

[0168] S100 Pre-processing: Construct a prior grid map based on pose information and environmental information, generate a multi-resolution grid map based on the prior grid map, and divide the multi-resolution grid map into high-confidence and low-confidence regions according to the task information set by the user to obtain high-confidence regions and low-confidence regions. The task information is used to provide motion control and navigation planning for the mobile robot.

[0169] S200 Global Initialization Localization: Control the mobile robot to perform a 360-degree rotation in place. Based on the point cloud information collected by the LiDAR and the pose information collected by the IMU, the attitude of the mobile robot's rotation is estimated. A global point cloud map is constructed based on the LiDAR point cloud information in the LiDAR coordinate system. Branch boundary scanning and matching are performed based on the multi-resolution grid map and high confidence area to obtain global localization.

[0170] S300 Follow Positioning: Based on the attitude information collected by the IMU and the position information collected by the odometry, motion updates are performed to obtain the predicted pose value. Based on the predicted pose value, high confidence region, and particle genetic mutation, lidar observation updates are performed to obtain the observed pose value. Based on the observed pose value, resampling is performed to obtain the pose.

[0171] The preprocessing step S100 includes the following operations:

[0172] (1) Constructing a prior grid map: Adjust the scanning operation environment of the mobile robot, and construct a prior grid map of a specified resolution based on the environmental information collected by the sensor and graph optimization.

[0173] (2) Constructing a multi-resolution raster map: Remove invalid points from the prior raster map using drawing tools to obtain a preprocessed multi-resolution raster map. Based on the multiple of the resolution of the prior raster map as the length of the search box, perform sliding window calculation on the prior raster map to obtain the multi-resolution raster map.

[0174] (3) Dividing high and low confidence regions: When configuring high and low confidence region division, the user-defined task information is written into a multi-resolution raster map in raster form, and a likelihood domain space based on the trajectory corresponding to the task information is generated. The formula for calculating the likelihood domain space is as follows:

[0175] ,

[0176] in, For free space points in a multi-resolution raster map, These are the points in the trajectory corresponding to the task information. Set the width parameter for the likelihood domain. The distance from a blank grid cell in a multi-resolution raster map to a specified route will be... The likelihood domain space within the range is considered a high-confidence region, while any region exceeding the width of the likelihood domain is considered a low-confidence region with a score of -1.

[0177] Step S200 global initialization positioning includes the following operations:

[0178] (1) Construct a cloud map of the entire scenic spot: Control the mobile robot to perform a 360-degree rotation in place, record the attitude information of the IMU at a predetermined frequency, and estimate the attitude based on the timestamps of each point in the laser point cloud and the attitude information of the IMU under each timestamp of the linear interpolation. Then, read the laser point cloud information under the laser radar coordinate system in sequence and stitch the laser point cloud information to obtain a cloud map of the entire scenic spot.

[0179] (2) Global initial pose lookup: Global initial pose lookup is performed by scanning and matching based on multi-resolution grid map and high confidence region, or by searching the full confidence region.

[0180] The global initial pose lookup is performed using a scanning and matching method based on multi-resolution raster maps and high-confidence regions, including the following operations:

[0181] L1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0182] L2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. If the proportion of low-confidence areas in any one of the smaller regions exceeds 2 / 3, skip the scan matching calculation for that region and calculate the scan matching score using the following formula:

[0183] ,

[0184] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0185] L3. Select the search space with the highest scan matching score from the four small regions, and perform the next round of branching operation on the selected search space. Repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0186] If the optimal pose is not found through steps L1-L3, a global initial pose search is performed using a full confidence region search, including the following operations:

[0187] M1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0188] M2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. For each smaller region, calculate the scan matching score using the following formula:

[0189] ,

[0190] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0191] M3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0192] If high and low confidence regions were not divided during preprocessing, global initial pose lookup is performed based on scanning and matching of multi-resolution raster maps, including the following steps:

[0193] S1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map;

[0194] S2. Perform a branching operation on the target raster map. Divide the search space into four small regions by halving the x and y coordinates of the target raster map. For each small region, calculate the scan matching score using the following formula:

[0195] ,

[0196] in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point;

[0197] S3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

[0198] Step S300 follow-up positioning includes the following operations:

[0199] (1) Motion update: Based on the attitude information collected by the IMU and the position information collected by the odometry, the motion update of the mobile robot pose is performed based on the EKF algorithm to obtain the predicted value of each pose;

[0200] (2) Observation update: The predicted pose value, the laser point cloud information collected by the lidar and the minimum resolution grid map in the multi-resolution grid map are used as inputs. The particle weight is calculated and output by the likelihood domain observation model. The pose update value is updated by observation based on the high confidence region and particle genetic mutation to obtain the observation value of each pose. A particle set is constructed based on the observation value of each pose.

[0201] (3) Resampling: Select the pose with the largest observation value from the particle set. If the observation value of the pose is greater than the threshold THR, output the pose. If the observation value of the pose is less than or equal to the threshold THR, resample based on the high confidence region and select the pose located in the high confidence region to add to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observation value of the pose is greater than the threshold THR, output the pose. If the observation value of the pose is less than or equal to the threshold THR, perform global random sampling based on the high confidence region and the low confidence region, randomly select the pose to add to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observation value of the pose is greater than the threshold THR, output the pose. If the observation value of the pose is less than or equal to the threshold THR, select the particle with the highest weight in the particle set for output to ensure real-time positioning.

[0202] Step (3) Before resampling, perform the following operations: Divide the particles into two groups according to their weights, and select one-third of the smaller weighted particle set for particle crossover and mutation to add new unknown particles. The calculation formula is as follows:

[0203] ,

[0204] ,

[0205] In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

[0206] The present invention has been shown and described in detail above with reference to the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above embodiments, those skilled in the art will know that more embodiments of the present invention can be obtained by combining the code review methods in the different embodiments. These embodiments are also within the protection scope of the present invention.

Claims

1. A mobile robot positioning system based on lidar scanning information, characterized in that, This system is applied to mobile robots equipped with sensing sensors, including odometry, IMU, and lidar, for collecting pose and environmental information. The positioning system includes a preprocessing module, a global initialization positioning module, and a following positioning module. The preprocessing module is used to construct a prior grid map based on pose information and environmental information, generate a multi-resolution grid map based on the prior grid map, and divide the multi-resolution grid map into high-confidence and low-confidence regions according to the task information set by the user, thereby obtaining high-confidence regions and low-confidence regions. The task information is used to provide motion control and navigation planning for the mobile robot. The global initialization positioning module is used to control the mobile robot to perform a 360-degree rotation in place. It is used to estimate the attitude of the mobile robot's rotation based on the point cloud information collected by the lidar and the pose information collected by the IMU, and to construct a global point cloud map based on the lidar point cloud information in the lidar coordinate system. It is used to perform branch delimitation scanning matching based on the multi-resolution raster map and high confidence region to obtain global positioning. The following positioning module is used to perform motion updates based on the attitude information collected by the IMU and the position information collected by the odometry to obtain the predicted pose value. Based on the predicted pose value, the likelihood domain observation model, the high confidence region, and particle genetic mutation, the LiDAR observation is updated to obtain the observed pose value. Based on the observed pose value, the pose is resampled to obtain the updated pose.

2. The mobile robot positioning system based on lidar scanning information according to claim 1, characterized in that, The preprocessing module interacts with the user through a preprocessing interface and performs the following: Constructing a prior grid map: The remote-controlled mobile robot scans the working environment and constructs a prior grid map of a specified resolution based on the environmental information collected by the perception sensors and graph optimization. Constructing a multi-resolution raster map: Invalid points in the prior raster map are removed using drawing tools to obtain a preprocessed prior raster map. The length of the search box is calculated by using a sliding window on the prior raster map based on a multiple of its resolution to obtain the multi-resolution raster map. Dividing regions into high and low confidence levels: When configuring high and low confidence level division, the user-defined task information is written into a multi-resolution raster map in raster form, and a likelihood domain space based on the trajectory corresponding to the task information is generated. The formula for calculating the likelihood domain space is as follows: ， in, For free space points in a multi-resolution raster map, These are the points in the trajectory corresponding to the task information. Set the width parameter for the likelihood domain. The distance from a blank grid cell in a multi-resolution raster map to a specified route will be... The likelihood domain space within the range is considered a high-confidence region, while any region exceeding the width of the likelihood domain is considered a low-confidence region with a score of -1.

3. The mobile robot positioning system based on lidar scanning information according to claim 1, characterized in that, The global initialization positioning module is used to perform the following: Constructing a cloud map of the entire scenic spot: Control the mobile robot to perform a 360-degree rotation in place, record the attitude information of the IMU at a predetermined frequency, and perform attitude estimation based on the timestamps of each point in the laser point cloud and the attitude information of the IMU under each timestamp of linear interpolation. Then, sequentially read the laser point cloud information under the laser radar coordinate system and stitch the laser point cloud information to obtain a cloud map of the entire scenic spot. Global initial pose lookup: Global initial pose lookup is performed by scanning and matching based on multi-resolution raster maps and high-confidence regions, or by searching based on full-confidence regions. The global initial pose lookup is performed using a scanning and matching method based on multi-resolution raster maps and high-confidence regions, including the following operations: L1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map; L2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. If the proportion of low-confidence areas in any one of the smaller regions exceeds 2 / 3, skip the scan matching calculation for that region and calculate the scan matching score using the following formula: ， in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point; L3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found. If the optimal pose is not found through steps L1-L3, the global initialization localization module is used to perform a global initialization pose search by conducting a full confidence region search, including the following operations: M1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map; M2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. For each smaller region, calculate the scan matching score using the following formula: ， in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point; M3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found. If the preprocessing module does not perform high and low confidence region segmentation, the global initialization localization module is used to perform global initial pose lookup based on scanning and matching of multi-resolution raster maps, including the following steps: S1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map; S2. Perform a branching operation on the target raster map. Divide the search space into four small regions by halving the x and y coordinates of the target raster map. For each small region, calculate the scan matching score using the following formula: ， in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point; S3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

4. The mobile robot positioning system based on lidar scanning information according to any one of claims 1-3, characterized in that, The follow-up positioning module is used to perform the following: Motion update: Based on the attitude information collected by the IMU and the position information collected by the odometry, the motion update of the mobile robot pose is performed based on the EKF algorithm to obtain the predicted value of each pose; Observation Update: Taking the predicted pose value, the laser point cloud information collected by lidar, and the minimum resolution grid map in the multi-resolution grid map as input, the pose update value is calculated and output by calculating particle weights through the likelihood domain observation model. The pose update value is then observed and updated based on high confidence regions and particle genetic mutations to obtain the observation value for each pose. A particle set is then constructed based on the observation value for each pose. Resampling: Select the pose with the largest observation value from the particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, resampling is performed based on the high-confidence region. The pose located in the high-confidence region is selected and added to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, global random sampling is performed based on the high-confidence region and the low-confidence region. Randomly select poses and add them to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, the particle with the highest weight in the particle set is selected for output to ensure real-time positioning.

5. The mobile robot positioning system based on lidar scanning information according to claim 4, characterized in that, Before resampling, the following localization module performs the following: Particles are divided into two groups based on their weights. One-third of the smaller weighted particle set is selected for particle crossover and mutation to add new unknown particles. The calculation formula is as follows: ， ， In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

6. A method for locating a mobile robot based on lidar scanning information, characterized in that, An application to a mobile robot equipped with sensing sensors, including an odometry system, an IMU, and a lidar, for collecting pose and environmental information, the method is used for localization via a mobile robot localization system based on lidar scanning information as described in any one of claims 1-5, the method comprising the following steps: Preprocessing: A prior grid map is constructed based on pose information and environmental information. A multi-resolution grid map is generated based on the prior grid map. According to the task information set by the user, high and low confidence regions are divided in the multi-resolution grid map to obtain high confidence regions and low confidence regions. The task information is used to provide motion control and navigation planning for the mobile robot. Global initialization localization: Control the mobile robot to perform a 360-degree rotation in place. Based on the point cloud information collected by the LiDAR and the pose information collected by the IMU, the attitude of the mobile robot's rotation is estimated. A global point cloud map is constructed based on the LiDAR coordinate system coordinate system point cloud information. Branch boundary scanning and matching are performed based on multi-resolution grid map and high confidence area to obtain global localization. Follow-up localization: Motion updates are performed based on attitude information collected by IMU and position information collected by odometry to obtain the predicted pose value. Based on the predicted pose value, the likelihood domain observation model, the high confidence region, and particle genetic mutation, lidar observation updates are performed to obtain the observed pose value. Based on the observed pose value, resampling is performed to obtain the pose.

7. The mobile robot positioning method based on lidar scanning information according to claim 6, characterized in that, Preprocessing includes the following steps: Constructing a prior grid map: The remote-controlled mobile robot scans the working environment and constructs a prior grid map of a specified resolution based on the environmental information collected by the perception sensors and graph optimization. Constructing a multi-resolution raster map: Invalid points in the prior raster map are removed using drawing tools to obtain a preprocessed prior raster map. The length of the search box is calculated by using a sliding window on the prior raster map based on a multiple of its resolution to obtain the multi-resolution raster map. Dividing regions into high and low confidence levels: When configuring high and low confidence level division, the user-defined task information is written into a multi-resolution raster map in raster form, and a likelihood domain space based on the trajectory corresponding to the task information is generated. The formula for calculating the likelihood domain space is as follows: ， in, For free space points in a multi-resolution raster map, These are the points in the trajectory corresponding to the task information. Set the width parameter for the likelihood domain. The distance from a blank grid cell in a multi-resolution raster map to a specified route will be... The likelihood domain space within the range is considered a high-confidence region, while any region exceeding the width of the likelihood domain is considered a low-confidence region with a score of -1.

8. The mobile robot positioning method based on lidar scanning information according to claim 6, characterized in that, Global initialization positioning includes the following operations: Constructing a cloud map of the entire scenic spot: Control the mobile robot to perform a 360-degree rotation in place, record the attitude information of the IMU at a predetermined frequency, and perform attitude estimation based on the timestamps of each point in the laser point cloud and the attitude information of the IMU under each timestamp of linear interpolation. Then, sequentially read the laser point cloud information under the laser radar coordinate system and stitch the laser point cloud information to obtain a cloud map of the entire scenic spot. Global initial pose lookup: Global initial pose lookup is performed by scanning and matching based on multi-resolution raster maps and high-confidence regions, or by searching based on full-confidence regions. The global initial pose lookup is performed using a scanning and matching method based on multi-resolution raster maps and high-confidence regions, including the following operations: L1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map; L2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. If the proportion of low-confidence areas in any one of the smaller regions exceeds 2 / 3, skip the scan matching calculation for that region and calculate the scan matching score using the following formula: ， in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point; L3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found. If the optimal pose is not found through steps L1-L3, the global initialization localization module is used to perform a global initialization pose search by conducting a full confidence region search, including the following operations: M1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map; M2. Perform a branching operation on the target raster map. Divide the search space into four smaller regions by halving the x and y coordinates of the target raster map. For each smaller region, calculate the scan matching score using the following formula: ， in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point; M3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found. If the preprocessing module does not perform high and low confidence region segmentation, the global initialization localization module is used to perform global initial pose lookup based on scanning and matching of multi-resolution raster maps, including the following steps: S1. Select the highest resolution raster map from the multi-resolution raster maps as the target raster map; S2. Perform a branching operation on the target raster map. Divide the search space into four small regions by halving the x and y coordinates of the target raster map. For each small region, calculate the scan matching score using the following formula: ， in, This represents the bicubic interpolation function, used to smooth the grayscale information of a target raster map. Indicates attitude adjustment variables, For the first Measurement information for each laser point; S3. Select the search space with the highest scan matching score from the four small regions, perform the next round of branching operation on the selected search space, and repeat step L2 with a lower resolution raster map until the optimal pose is found.

9. The mobile robot positioning method based on lidar scanning information according to any one of claims 6-8, characterized in that, Follow positioning includes the following operations: Motion update: Based on the attitude information collected by the IMU and the position information collected by the odometry, the motion update of the mobile robot pose is performed based on the EKF algorithm to obtain the predicted value of each pose; Observation Update: Taking the predicted pose value, the laser point cloud information collected by lidar, and the minimum resolution grid map in the multi-resolution grid map as input, the pose update value is calculated and output by calculating particle weights through the likelihood domain observation model. The pose update value is then observed and updated based on high confidence regions and particle genetic mutations to obtain the observation value for each pose. A particle set is then constructed based on the observation value for each pose. Resampling: Select the pose with the largest observation value from the particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, resampling is performed based on a high-confidence region. The pose located in the high-confidence region is selected and added to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, global random sampling is performed based on high-confidence and low-confidence regions. Randomly select poses and add them to the particle set to obtain a new particle set. Select the pose with the largest observation value from the new particle set. If the observed value of the pose is greater than the threshold THR, output the pose. If the observed value of the pose is less than or equal to the threshold THR, the particle with the highest weight in the particle set is selected for output to ensure real-time positioning.

10. The mobile robot positioning method based on lidar scanning information according to claim 9, characterized in that, Before resampling, the following operation is performed: Particles are divided into two groups based on their weights. One-third of the smaller weighted particle set is selected for particle crossover and mutation to add new unknown particles. The calculation formula is as follows: ， ， In the formula, , , , These are the small-weight particle set, the large-weight particle set, the cross-particle set, and the mutation particle set, respectively.

Images