Unmanned system cluster remote cooperative localization and global map construction method and system

Abstract

The invention provides an unmanned system cluster remote co-location and global map construction method and system. Firstly, a remote server determines an initial relative pose of a node according to point cloud data of each node and performs initial global point cloud map splicing; each node obtains a speedometer pose and transmits the speedometer pose to the remote server in real time, and then the global pose P1 of each node is updated in a global point cloud map of the remote server; each node calculates self displacement change according to the speedometer pose, and when the displacement change is greater than a set distance threshold value, a local point cloud d1 is extracted and transmitted to the remote server; the remote server locally extracts a local point cloud D1; and performing registration on the d1 and the D1 to realize updating of the global map by the local point cloud. By using the method, the matching efficiency and the collaborative mapping efficiency can be improved; and meanwhile, the single-node resource consumption is reduced, and the single-node cruising ability is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned system navigation, and particularly relates to a method and system for remote cooperative positioning and global map construction of an unmanned system cluster. Background Art

[0002] In the research field of cooperative positioning and mapping of unmanned systems, the existing technologies mainly focus on improving the cooperation efficiency and accuracy among multiple robots or unmanned aerial vehicles. These technologies enable unmanned systems to achieve precise positioning and efficient map construction in complex environments by using advanced sensor fusion, machine learning algorithms, and communication protocols. For example, some patents describe methods for simultaneous localization and mapping (SLAM) using sensor data such as lidar, cameras, and IMUs (inertial measurement units), as well as mechanisms for data sharing and cooperative work through wireless communication networks. The development of these technologies not only promotes the application of unmanned systems in fields such as search and rescue, environmental monitoring, and logistics distribution, but also lays a foundation for the further development of future automation and intelligence.

[0003] A method and system for cross-domain cooperative positioning and mapping based on three-dimensional point clouds (Publication No.: CN113470089A) proposes a method and system for cross-domain cooperative positioning and mapping based on three-dimensional point clouds. This method uses multiple aerial unmanned vehicles and multiple ground unmanned vehicles, combined with a lidar odometry system, a machine-side air-ground cross-domain global positioning and mapping system, and a control-side system on a central server, to achieve continuous point cloud registration, local point cloud fragment map construction, feature extraction, global point cloud map matching and association, pose trajectory global optimization, pose update, and precise update of the current pose. Three-dimensional point cloud feature extraction is performed through a deep learning model, and precise positioning is achieved using the maximum clique algorithm and the dual ICP registration method.

[0004] A method for cooperative positioning and mapping based on multiple robots (Publication No.: CN118362112A) realizes a method for cooperative positioning and mapping based on multiple robots by grading the mapping area according to the environmental complexity and accordingly allocating a corresponding number of robots to perform specific tasks. In the specific implementation process, the first task robot uses lidar to obtain and denoise point cloud data, the second task robot determines the robot's position through a GPS locator, the third task robot is responsible for collecting environmental images, and the fourth task robot receives this data and uses the SLAM algorithm to complete environmental mapping. In addition, this method also involves using the PointNet++ model and 3D reconstruction algorithm to process point cloud data, as well as achieving information sharing and cooperative work among robots through a wireless local area network, and finally applying the mapping result to the positioning and route planning of autonomous driving vehicles to improve the accuracy and efficiency of mapping.

[0005] In the prior art, although there are servers other than unmanned systems for processing complex calculations to achieve multi-robot collaborative positioning and mapping, the functions of the servers have not been fully utilized. For example, the advantages of using a remote server to improve the relocalization effect of a single robot have not been considered. That is, using a remote server to perform relocalization for multiple single-robot unmanned systems can not only separate the computational loads of the front and rear ends of the positioning and mapping algorithms, improve the endurance of the unmanned system, but also adapt to the characteristics of low relocalization frequency requirements and large computational amounts. Another example is that when the number of nodes in the unmanned system cluster increases, a large amount of observation data will be introduced, and existing collaborative positioning and mapping methods rarely pay attention to how to improve the problem of poor real-time performance further caused by the increased computational amount. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for remote collaborative positioning and global map construction of an unmanned system cluster. By introducing a remote server into the unmanned system cluster architecture, the computational amount of server endpoint cloud matching is reduced, the matching efficiency is improved, and thus the collaborative mapping efficiency can be improved; at the same time, the resource consumption of a single node is reduced, and the endurance of a single point is improved.

[0007] To solve the above technical problems, the present invention is implemented as follows.

[0008] A method for remote collaborative positioning and global map construction of an unmanned system cluster includes:

[0009] Step 1: The remote server receives the point cloud data of each node in the unmanned system cluster in the initial stationary state, performs point cloud matching, obtains the initial relative pose of the nodes, and performs initial global point cloud map stitching;

[0010] Step 2: Each node obtains the odometry pose and transmits it to the remote server in real time, and then updates the global pose P1 of each node in the global point cloud map of the remote server; the initial value of registration is obtained through coordinate transformation according to the real-time updated global pose P1;

[0011] Each node also calculates its own displacement change according to the odometry pose. When the displacement change is greater than the set distance threshold d 0 , the node transmits the first local point cloud d1 scanned by the node from the moment when the distance threshold was last triggered to the current moment t k to the remote server to trigger map stitching; the remote server selects the second local point cloud D1 in the global point cloud map within a spherical region centered at the global position of the node at the current moment t k and with a radius of multiple d 0 ; the first local point cloud d1 and the second local point cloud D1 are registered to update the local point cloud to the global map.

[0012] Step 3: Each node of the unmanned system cluster transmits single-frame lidar point cloud data to the remote server at a set frequency, and the remote server assists in realizing remote relocalization.

[0013] Preferably, the threshold d 0 is determined according to the size of the environmental scene and hardware resource limitations; the narrower the scene, the greater the change in the scene during movement, and a smaller threshold d needs to be set 0 ; and the smaller the threshold d 0 is, the higher the frequency of map stitching and updating, and the greater the occupation of network and computing resources.

[0014] Preferably, the method further includes Step 4: The remote server eliminates redundant global point cloud map points in the global point cloud map.

[0015] Preferably, Step 4 includes:

[0016] Step a: The remote server uses Voxel Grid filtering to divide the point cloud space into regular three-dimensional voxels, and calculates a representative point for the points within each voxel to represent all the points within a voxel in the original point cloud, thereby realizing point cloud sparsification;

[0017] Step b: Use the density-based clustering algorithm DBSCAN to eliminate redundant information in the point cloud data, and use the centroid of each cluster obtained by the DBSCAN algorithm as the representative point to realize compressed point cloud.

[0018] Preferably, calculating a representative point for the points within each voxel in Step a is: for each voxel, select the mean value of all the points within the voxel as the representative point.

[0019] The present invention also provides an unmanned system cluster remote collaborative positioning and global map construction system, including a node-side processing module and a server-side processing module; the node-side processing module is set in each node of the unmanned system cluster; the server-side processing module is set in the remote server;

[0020] The node-side processing module includes a first initialization sub-module, a monitoring sub-module, and a local point cloud extraction sub-module; the server-side processing module includes a second initialization sub-module, an update sub-module, a local point cloud extraction and update sub-module, and a relocalization sub-module;

[0021] The first initialization sub-module is used to collect point cloud data in the initial stationary state during initialization and send it to the remote server;

[0022] The monitoring sub-module is used to obtain the odometer pose during collaborative positioning and transmit it to the remote server in real time; at the same time, calculate its own displacement change according to the odometer pose, and when the displacement change is greater than the set threshold d 0, trigger the local point cloud extraction sub-module;

[0023] The local point cloud extraction sub-module is used to extract the first local point cloud d1 scanned by this node from the previous trigger threshold moment to the current moment t k and transmit it to the remote server to trigger map stitching;

[0024] The second initialization sub-module is used to receive the point cloud data of each node in the initial static state during initialization, perform point cloud matching, obtain the relative pose of the initialized node, and perform initial global point cloud map stitching;

[0025] The update sub-module is used to update the global pose P1 of each node in the global point cloud map according to the real-time odometry pose sent by the node during collaborative positioning; obtain the initial value of registration through coordinate transformation based on the real-time updated global pose P1;

[0026] The local point cloud extraction and update sub-module is used to, after receiving the first local point cloud d1 from the node, within a spherical region centered on the global position of the node at the current moment t k with multiple times of d 0 as the radius, select the second local point cloud D1 in the global point cloud map on the server side; perform registration using the first local point cloud d1 and the second local point cloud D1 to realize the update of the local point cloud to the global map;

[0027] The repositioning sub-module is used to receive the single-frame lidar point cloud data transmitted by each node at a set frequency to assist in realizing remote repositioning.

[0028] Preferably, the threshold d used by the monitoring sub-module 0 is determined according to the size of the environmental scene and the hardware resource limitations; the narrower the scene, the greater the change in the scene during the movement process, and a smaller threshold d needs to be set 0 ; and the smaller the threshold d 0 is, the higher the frequency of map stitching update, and the greater the occupation of network and computing resources.

[0029] Preferably, the server-side processing module further includes a redundancy elimination sub-module for eliminating redundant global point cloud map points in the global point cloud map.

[0030] Preferably, the redundancy elimination sub-module includes a Voxel Grid filtering module and a DBSCAN algorithm module;

[0031] The Voxel Grid filtering module is used to divide the point cloud space into regular three-dimensional voxels using Voxel Grid filtering, and calculate a representative point for the points within each voxel to represent all the points within a voxel in the original point cloud, thereby realizing point cloud sparsification;

[0032] The DBSCAN algorithm module is used to eliminate redundant information in the point cloud data using the DBSCAN algorithm, and take the centroid of each cluster obtained by the DBSCAN algorithm as a representative point to achieve point cloud compression.

[0033] Beneficial effects:

[0034] (1) In the present invention, the displacement length is calculated by the cluster single-node odometer and the threshold is set to control the local map stitching frequency, which can timely trigger the update of the global map when the moving distance is sufficient and the environment changes significantly. Moreover, the displacement interval can be used as a set, and only the point cloud map points generated within the displacement interval are transmitted for global map construction, improving the collaborative mapping efficiency.

[0035] When the single-node point cloud is transmitted to the remote server for matching, the local point cloud within the spherical radius neighborhood centered on the single-node global position in the remote server is used as the target for matching, reducing the point cloud matching calculation amount and improving the matching efficiency.

[0036] (2) The present invention makes full use of the characteristics of relocalization, which has low real-time requirements and large computational complexity. The relocalization process with high computational complexity is handed over to the remote server. While the remote server realizes global map construction and positioning, it also completes the relocalization process of the cluster single-node and returns the corrected positioning information. The single node only needs to transmit a single-frame point cloud information to the server at a low frequency for matching with the global map, and the restored pose after matching is then transmitted by the server to the single node for pose correction. Separating the relocalization process from the single unmanned system node can reduce the resource consumption of the single node and improve the battery life.

[0037] (3) To address the problems of large point cloud data volume and poor real-time performance caused by the increase in cluster nodes and long-term operation, after the remote server realizes global map construction, redundant map points are removed from the global map to compress the point cloud data, achieving improved system real-time performance at the remote server end, reducing the map volume and hardware resource occupancy. Description of the drawings

[0038] Figure 1 It is a flowchart for the initialization of the unmanned system cluster.

[0039] Figure 2 It is a flowchart for collaborative global map construction and update.

[0040] Figure 3 It is a flowchart for the remote server to eliminate redundant global point cloud map points.

[0041] Figure 4 It is a flowchart for remote relocalization of the remote server.

[0042] Figure 5Overall flowchart of the preferred embodiment.

[0043] Figure 6 Structural diagram of the unmanned system cluster remote collaborative positioning and global map construction system of the present invention.

[0044] Figure 7 is Figure 6 Schematic diagram of the composition of the redundant sub-module elimination in Detailed implementation manners

[0045] The present invention provides a method for unmanned system cluster remote collaborative positioning and global map construction. The core idea is that each node in the unmanned system cluster uses the odometer pose to calculate its own displacement change. When the displacement change is greater than the set threshold d 0 , the node will transmit the first local point cloud d1 scanned by this node from the moment when the threshold was last triggered to the current moment t k to the remote server to trigger map stitching; the remote server selects the second local point cloud D1 in the global point cloud map within a spherical area centered on the global position of the node at the current moment t k and with a radius of multiple d 0 ; the first local point cloud d1 and the second local point cloud D1 are registered to update the local point cloud to the global map.

[0046] It can be seen that the present invention uses the displacement change amount as the threshold, which is different from the prior art that uses the time frequency as the basis. Using the fixed time frequency as the basis will attempt to perform map update operations even when the speed is very small or stopped, wasting computing resources, and may cause too much scene change to affect the stitching effect when the speed is relatively large. The present invention sets a distance threshold to control the local map stitching frequency, which can trigger the update of the global map in time when the moving distance is sufficient and the environment changes significantly. It can also use the displacement interval as a set and only transmit the point cloud map points generated within the displacement interval for global map construction, improving the collaborative mapping efficiency. At the same time, when the remote server performs matching, it uses the local point cloud within the spherical radius neighborhood centered on the global position of a single node in the remote server as the target for matching, reducing the point cloud matching calculation amount and improving the matching efficiency.

[0047] In addition, the present invention also makes full use of the characteristics of relocalization that has low real-time requirements and large computational complexity, transfers the relocalization processing with high computational complexity to the remote server, and adds an operation to eliminate redundant global point cloud map points in the remote server, improving the system real-time performance from the perspective of reducing the node burden, reducing the hardware resource occupation of the server, and improving the node endurance.

[0048] The following combines the accompanying drawings and gives examples to describe the present invention in detail.

[0049] Step 1: Initialization of the unmanned system cluster: Taking advantage of the characteristic that there is a short-distance common visual area in the initial state of the unmanned system cluster, the remote server receives the point cloud data of each node in the unmanned system cluster in the initial static state, performs point cloud matching, initializes the relative poses of the nodes, and performs initial global point cloud map stitching.

[0050] Specifically, the remote server connects to each node of the cluster through the network, assigns a unique identifier to the successfully connected single machine to distinguish different single unmanned node systems. The single unmanned system node in the initial state remains stationary to collect the point cloud information of the surrounding environment. The point cloud is transmitted to the remote server through the network, point cloud matching is performed, the relative poses of each node are obtained, and initial global point cloud map stitching is carried out.

[0051] In this step, the process for the server to achieve point cloud matching is as follows:

[0052] First, use the sampling consistency initial registration algorithm to achieve rough point cloud registration. Select n sampling points from the point cloud P to be registered. To ensure that the sampled points have different FPFH features as much as possible, the distance between the sampling points should satisfy being greater than the pre-given minimum distance threshold d. Search for one or more points in the target point cloud Q that have similar FPFH features to the sampling points in the point cloud P, and randomly select one point from these similar points as the corresponding point of the point cloud P in the target point cloud Q. Finally, calculate the rigid body transformation matrix between the corresponding points.

[0053] Take the two point clouds P′ (the source point cloud after coordinate transformation) and Q after initial registration as the initial point sets for fine registration. For each point p in the source point cloud P′ i , find the corresponding point q with the closest distance in the target point cloud Q i , as the corresponding point of this point in the target point cloud, and form the initial corresponding point pairs.

[0054] Use SVD decomposition to calculate the rotation matrix R and the translation vector T to minimize, that is, to minimize the mean square error between the corresponding point sets. Set a certain threshold ε = d k -d k-1 and the maximum number of iterations N max , and apply the rigid body transformation obtained in the previous step to the source point cloud P′ to obtain the new point cloud P″.

[0055] For each point p' in the local point cloud P′ i =(x i ,y i ,z i ), its position p'i′ in the global coordinate system can be calculated by the following formula: p' i ' = Rp' i +t; where R is the rotation matrix and t is the translation vector.

[0056] Calculate the distance error between P″ and Q. If the error between two iterations is less than the threshold ε or the current iteration number is greater than N max , the iteration ends. Otherwise, update the initially registered point set to P″ and Q, and continue to repeat the above steps until the convergence condition is met.

[0057] If the point cloud registration is successful, perform initial global map stitching. Register the initially registered point cloud of the next node with the initially global map stitched by the successfully initialized unmanned system node. If the registration is successful, the remote server continues to incrementally update the initially global map, and the initialization is completed after all nodes are processed. If the point cloud registration of a node in the cluster fails, the node rotates in place by a preset angle to move the point cloud scanning area to find a co-visible area, and then performs initial registration at the new position after rotation to obtain the initial relative pose.

[0058] As Figure 1 shown, after the remote server completes the initialization, the initial relative pose and the initial global point cloud map are obtained. At this time, notify each node to start running its own odometry calculation method. Jump to Figure 2 , and each node uses the odometry information to calculate its own odometry pose information in real time and construct its own local point cloud map.

[0059] Step 2: After each node of the unmanned system cluster triggers the distance threshold, extract the local map according to the set rules and transmit it to the remote server for processing. The remote server extracts the local local map according to the set rules and registers it with the local map from the node to update the local point cloud to the global map.

[0060] This step specifically includes the following sub-steps:

[0061] Step 2.1: The odometry poses of each node of the unmanned system cluster are transmitted to the remote server in real time, and the coordinate system alignment is performed through pose transformation in the remote server, and then the global poses P1 of each node are updated in the global map of the server. According to the globally updated pose P1, the initial value of the point cloud registration pose is obtained through coordinate transformation, which is equivalent to rough alignment here. See Figure 2 for a branch.

[0062] Step 2.2: Each node of the unmanned system cluster calculates its own displacement change according to the odometry information. Set a threshold d 0 for the displacement change amount, which is used as the basis for the map stitching frequency. The narrower the scene, the greater the scene change during the movement, and a smaller d 0 needs to be set. The smaller the displacement threshold, the higher the frequency of map stitching update, and the greater the occupation of network and computing resources. Therefore, the specific size of d 0 needs to be determined according to the actual size of the environmental scene and the hardware resource limitations.

[0063] When the displacement changes calculated by each node of the unmanned system cluster exceed the pre-set distance threshold d 0 , at this time, each single node transmits the local point cloud scanned by itself from the last update time (triggering the distance threshold) to the current time t k (denoted as d1) to the remote server to trigger map stitching. See Figure 2 another branch in

[0064] Step 2.3: The remote server selects the point cloud data (denoted as D1) in the global point cloud map within a spherical region centered at the global position of the node at time t k and with a radius of multiple d 0 for point cloud matching. The local point cloud d1 is registered with the local point cloud D1 to update the local point cloud to the global map.

[0065] Step 3: The remote server eliminates redundant points in the global point cloud map.

[0066] See Figure 3 , in this embodiment, eliminating redundancy includes the following sub-steps:

[0067] Step 3.1: The remote server uses Voxel Grid filtering to divide the point cloud space into regular three-dimensional grids (voxels) and calculates a representative point for the points within each voxel.

[0068] In this step, first determine the size of the voxel, which is pre-set and determines the granularity of the filtering. Then map each point to the center of the nearest voxel; for each voxel, the mean value of all points within the voxel can be selected as the representative point. A sparser point cloud is obtained through Voxel Grid filtering, where each point represents all points within a voxel in the original point cloud.

[0069] Step 3.2: Use the density-based clustering algorithm (DBSCAN) to further eliminate redundant information in the point cloud data, and use the centroid of each cluster obtained by the DBSCAN algorithm as the representative point to compress the point cloud.

[0070] In this step, set two parameters: ε (neighborhood radius) and MinPts (the minimum number of neighbors required to become a core point). For each point p, find all its neighbors within the ε distance; if the number of neighbors of p is not less than MinPts, then mark p as a core point. For each core point, if its neighbors are also core points, then divide them into the same cluster. Border points are neighbors of core points but have fewer neighbors than MinPts, and they are assigned to the nearest cluster. Points that do not belong to any cluster are regarded as noise and eliminated.

[0071] Step 4: Each node of the unmanned system cluster transmits single-frame lidar point cloud data to the remote server at a low frequency, and the remote server assists in realizing remote relocalization.

[0072] The point cloud matching in this step is similar to that in Step 2, except that only the single-frame real-time point cloud data at a certain moment is transmitted, that is, the data of a complete scan / circle of the lidar, and there is no need to transmit the local map for matching.

[0073] See Figure 4 , this Step 4 includes the following sub-steps:

[0074] Step 4.1: Set the frequency of relocalization according to the scene size and hardware resource limitations.

[0075] Step 4.2: When relocalization is triggered, a single node of the unmanned system cluster transmits a frame of point cloud data to the remote server.

[0076] Step 4.3: Perform 3D-3D point cloud registration and iteratively obtain pose estimation information.

[0077] Step 4.4: Transform the estimated pose of the previous step to the local coordinate system of the single node to obtain the corrected pose.

[0078] Step 4.5: Transmit the corrected pose of the previous step to the corresponding node to update the single-node pose and realize relocalization.

[0079] Based on the above method, the present invention also provides an unmanned system cluster remote collaborative localization and global map construction system, as Figure 6 shown. The system includes a node-side processing module and a server-side processing module; the node-side processing module is set in each node of the unmanned system cluster; the server-side processing module is set in the remote server;

[0080] The node-side processing module includes a first initialization sub-module, a monitoring sub-module, and a local point cloud extraction sub-module; the server-side processing module includes a second initialization sub-module, an update sub-module, a local point cloud extraction and update sub-module, and a relocalization sub-module;

[0081] The first initialization sub-module is used to collect point cloud data in the initial stationary state during initialization and send it to the remote server;

[0082] The monitoring sub-module is used to obtain the odometry pose during collaborative localization and transmit it to the remote server in real time; at the same time, calculate its own displacement change according to the odometry pose. When the displacement change is greater than the set threshold d 0 , trigger the local point cloud extraction sub-module;

[0083] The local point cloud extraction sub-module is used to extract the point cloud from the moment of the previous trigger threshold to the current moment t when triggeredk Between the node scans the first local point cloud d1 and transmits it to the remote server to trigger map stitching;

[0084] The second initialization sub-module is used to receive the point cloud data of each node in the initial static state during initialization, perform point cloud matching, obtain the relative pose of the initialized node, and perform initial global point cloud map stitching;

[0085] The update sub-module is used to update the global pose P1 of each node in the global point cloud map according to the real-time odometry pose sent by the node during collaborative positioning; obtain the initial value of registration through coordinate transformation according to the globally updated pose P1 in real time;

[0086] The local point cloud extraction and update sub-module is used to, after receiving the first local point cloud d1 from the node, at the current moment t k take the global position of the node as the center of the circle and multiple d 0 as the radius, select the second local point cloud D1 in the global point cloud map on the server side; register the first local point cloud d1 with the second local point cloud D1 to realize the update of the local point cloud to the global map;

[0087] The relocalization sub-module is used to receive the single-frame lidar point cloud data transmitted by each node at a set frequency to assist in realizing remote relocalization.

[0088] In a preferred embodiment, the threshold d used by the monitoring sub-module 0 is determined according to the size of the environmental scene and the hardware resource limit; the narrower the scene, the greater the change in the scene during movement, and a smaller threshold d needs to be set 0 ; and the smaller the threshold d 0 is, the higher the frequency of map stitching update, and the greater the network and computing resources occupied.

[0089] In a preferred embodiment, the server-side processing module further includes a redundancy elimination sub-module for eliminating redundant global point cloud map points in the global point cloud map.

[0090] In a preferred embodiment, as Figure 7 shown, the redundancy elimination sub-module includes a Voxel Grid filtering module and a DBSCAN algorithm module;

[0091] The Voxel Grid filtering module is used to divide the point cloud space into regular three-dimensional voxels by using Voxel Grid filtering, and calculate a representative point for the points in each voxel to represent all the points in a voxel in the original point cloud, so as to realize point cloud sparsification;

[0092] The DBSCAN algorithm module is used to eliminate redundant information in the point cloud data using the DBSCAN algorithm, and take the centroid of each cluster obtained by the DBSCAN algorithm as the representative point to achieve compression of the point cloud.

[0093] The above specific embodiments only describe the design principle of the present invention. The shapes and names of the components in this description can be different and are not limited. Therefore, those skilled in the art of the present invention can modify or equivalently replace the technical solutions recorded in the foregoing embodiments; and these modifications and replacements do not depart from the gist and technical solutions of the present invention, and shall all fall within the protection scope of the present invention.

Claims

1. A method for remote collaborative positioning and global map construction of unmanned system clusters, characterized in that: include: Step 1: The remote server receives the point cloud data of each node in the unmanned system cluster in the initial static state, performs point cloud matching, obtains the initial relative position of the node, and performs initial global point cloud map stitching; Step 2: Each node obtains the odometer pose and transmits it to the remote server in real time, and then updates the global pose P1 of each node in the global point cloud map of the remote server; the initial value of the registration is obtained through coordinate transformation based on the real-time updated global pose P1; Each node also calculates its own displacement change based on the odometer posture. When the displacement change is greater than the set distance threshold d0, the node will k The node scans the first local point cloud d1 and transmits it to the remote server to trigger map stitching; the remote server uses the current time t k In a spherical area with the global position of the node as the center and multiple d0 as the radius, a second local point cloud D1 in the global point cloud map is selected; the first local point cloud d1 is used to align with the second local point cloud D1 to realize the update of the global map by the local point cloud; Step 3: Each node in the unmanned system cluster transmits single-frame lidar point cloud data to the remote server at a set frequency, and the remote server assists in remote repositioning.

2. The method according to claim 1, characterized in that The threshold d0 is determined based on the size of the environment scene and the hardware resource limitations; the narrower the scene, the greater the scene changes during movement, and a smaller threshold d0 needs to be set; and the smaller the threshold d0, the higher the frequency of map splicing updates, and the more network and computing resources are occupied.

3. The method according to claim 1, characterized in that The method further includes step 4: the remote server eliminates redundant global point cloud map points in the global point cloud map.

4. The method according to claim 3, characterized in that The step 4 comprises: Step a: The remote server uses Voxel Grid filtering to divide the point cloud space into regular three-dimensional voxels, and calculates a representative point for each point in a voxel to represent all points in a voxel in the original point cloud, thereby achieving point cloud sparseness. Step b: Use the density-based clustering algorithm DBSCAN to eliminate redundant information in the point cloud data, and use the centroid of each cluster obtained by the DBSCAN algorithm as the representative point to achieve compressed point cloud.

5. The method according to claim 4, characterized in that The step a is to calculate a representative point for each point in the voxel as follows: for each voxel, the mean of all points in the voxel is selected as the representative point.

6. An unmanned system cluster remote collaborative positioning and global map building system, characterized in that: It includes a node-side processing module and a server-side processing module; the node-side processing module is set in each node in the unmanned system cluster; the server-side processing module is set in the remote server; The node-side processing module includes a first initialization submodule, a monitoring submodule, and a local point cloud extraction submodule; the server-side processing module includes a second initialization submodule, an update submodule, a local point cloud extraction and update submodule, and a relocation submodule; The first initialization submodule is used to collect point cloud data in an initial static state during initialization and send it to a remote server; The monitoring submodule is used to obtain the odometer posture during collaborative positioning and transmit it to the remote server in real time; at the same time, it calculates its own displacement change according to the odometer posture, and when the displacement change is greater than the set threshold d0, it triggers the local point cloud extraction submodule; The local point cloud extraction submodule is used to extract the time from the last trigger threshold to the current time t when it is triggered. k The node scans the first local point cloud d1 and transmits it to the remote server to trigger map stitching; The second initialization submodule is used to receive the point cloud data of each node in an initial static state during initialization, perform point cloud matching, obtain the relative position of the initialized nodes, and perform initial global point cloud map splicing; The updating submodule is used to update the global pose P1 of each node in the global point cloud map according to the real-time odometer pose sent by the node during collaborative positioning; and obtain the initial value of the registration through coordinate transformation according to the real-time updated global pose P1; The local point cloud extraction and update submodule is used to extract and update the local point cloud at the current time t after receiving the first local point cloud d1 from the node. k In a spherical area with the global position of the node as the center and multiple d0 as the radius, a second local point cloud D1 in the global point cloud map on the server side is selected; the first local point cloud d1 is used to align with the second local point cloud D1 to realize the update of the global map by the local point cloud; The repositioning submodule is used to receive single-frame laser radar point cloud data transmitted by each node at a set frequency to assist in remote repositioning.

7. The system according to claim 6, characterized in that The threshold d0 used by the monitoring submodule is determined based on the size of the environmental scene and the hardware resource limitations; the narrower the scene, the greater the scene changes during movement, and a smaller threshold d0 needs to be set; and the smaller the threshold d0, the higher the frequency of map splicing updates, and the greater the network and computing resources occupied.

8. The system according to claim 6, characterized in that The server-side processing module further includes a redundancy elimination submodule for eliminating redundant global point cloud map points in the global point cloud map.

9. The system according to claim 8, characterized in that The redundancy elimination submodule includes a Voxel Grid filter module and a DBSCAN algorithm module; The Voxel Grid filtering module is used to divide the point cloud space into regular three-dimensional voxels using Voxel Grid filtering, and calculate a representative point for each point in a voxel to represent all points in a voxel in the original point cloud, thereby achieving point cloud sparseness; The DBSCAN algorithm module is used to eliminate redundant information in point cloud data using the DBSCAN algorithm, and use the centroid of each cluster obtained by the DBSCAN algorithm as a representative point to achieve compressed point cloud.

Images