A dynamic priority collaborative map construction and intelligent updating method and system

Abstract

The application discloses a dynamic priority cooperative map construction and intelligent updating method and system, relates to the field of robot environment mapping, and comprises the following steps: acquiring robot inspection data and constructing static feature data; extracting background features and key equipment features and constructing a hybrid cooperative map; acquiring a static target observed by a current robot, and calculating a distribution distance between an ellipsoid projection in the hybrid cooperative map and a corresponding two-dimensional projection of the static target; determining a multi-dimensional updating criterion according to running state information of the current robot and attribute information of each spatial region in the map; performing weighted fusion on the multi-dimensional updating criterion and the distribution distance to obtain a priority score corresponding to each spatial region; and matching the priority score with a preset adaptive threshold to dynamically update the hybrid cooperative map and improve the map consistency of the robot in a complex dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of robot environment mapping, and in particular to a method and system for dynamic priority collaborative map construction and intelligent updating. Background Technology

[0002] With the advancement of new power system construction, power distribution systems are characterized by diverse equipment, wide distribution, and complex environments. Traditional manual inspection methods suffer from low efficiency and high risk. In recent years, intelligent robot technology has been gradually applied in the field of power inspection, but currently, most operations still rely on single platforms, resulting in significant coverage blind spots and efficiency bottlenecks. To address this issue, collaborative operations using a cross-modal swarm of robots—comprising drones, unmanned vehicles, and quadruped robots—have become an important development direction.

[0003] Existing multi-robot collaborative mapping technologies have the following technical bottlenecks: traditional dense / sparse point cloud SLAM (such as OctoMap) / pure semantic SLAM often suffers from communication delays or loss of physical information due to massive geometric data, and existing local update mechanisms mostly rely on pixel-level comparison (such as IoU), which leads to frequent misjudgments and cannot support high-precision obstacle avoidance and trajectory planning.

[0004] This demonstrates how to effectively build maps through multi-robot collaboration and improve robots' performance in dynamic environments. Figure 1 Consistency has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a method and system for constructing and intelligently updating dynamic priority collaborative maps, in order to solve the problem of how to improve the collaborative reliability of robots in complex dynamic environments by constructing hybrid semantic collaborative maps.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for dynamic priority collaborative map construction and intelligent updating, comprising: Acquire robot inspection data to construct a static feature dataset; Background features and key equipment features are extracted from the static feature dataset to construct a hybrid collaborative map; Obtain the static target currently observed by the robot, and calculate the distribution distance between the corresponding ellipsoidal projection in the hybrid cooperative map and the corresponding two-dimensional projection of the static target; Based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map, determine the multidimensional update criteria corresponding to each spatial region; The multidimensional update criterion and the distribution distance are weighted and fused to quantify the priority score corresponding to each spatial region; The priority score is matched with a preset adaptive threshold, and the hybrid collaborative map is dynamically updated based on the matching result.

[0007] Furthermore, the acquisition of robot inspection data to construct a static feature dataset includes: The visual images and lidar data collected during robot inspection are acquired, and the timestamps of the visual images and lidar data are aligned to obtain aligned data. The alignment data is input into a lightweight edge detection network to identify dynamic objects and output the corresponding 2D bounding boxes. A dynamic object mask is generated by combining the 2D bounding box with the alignment data; Remove feature points within the dynamic object mask and output the static feature dataset.

[0008] Furthermore, the process of constructing the hybrid collaborative map includes: Using the aforementioned background features, a three-dimensional spatial voxel layer is constructed; In the dual space, multi-view 2D bounding boxes are extracted based on the key equipment features; Using the constraint relationships corresponding to the multi-view 2D bounding boxes, the multi-view 2D bounding boxes are jointly solved and reconstructed into a parameterized 3D ellipsoid layer that reflects the probability distribution characteristics. The three-dimensional spatial voxel layer is superimposed with the parameterized 3D ellipsoid layer to form the hybrid collaborative map.

[0009] Furthermore, calculating the distribution distance between the corresponding ellipsoidal projection in the hybrid collaborative map and the corresponding two-dimensional projection of the static target includes: The two-dimensional projection corresponding to the static target is taken as a two-dimensional Gaussian distribution; Extract the parameterized 3D ellipsoid layer from the hybrid collaborative map and establish the corresponding ellipsoid projection distribution; Calculate the Wasserstein distance between the two-dimensional Gaussian distribution and the ellipsoidal projection.

[0010] Furthermore, the step of determining the multidimensional update criterion corresponding to each spatial region based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map includes: The information timeliness criterion is calculated based on the time decay factor of the most recent update in each of the aforementioned spatial regions. The weights are calculated based on the Euclidean distance between the current robot's planned path and each of the aforementioned spatial regions, and are used as criteria for determining the criticality of the task. The perception credibility criterion is obtained by statistically calculating the cross-coverage rate of the robot's observation perspectives of each spatial region and the sensor confidence level.

[0011] Furthermore, the step of matching the priority score with a preset adaptive threshold and dynamically updating the hybrid collaborative map based on the matching result includes: When the priority score is lower than the first adaptive threshold, local maintenance tracking of the hybrid collaborative map is triggered. When the priority score is between the first adaptive threshold and the second adaptive threshold, a parameterized local update of the hybrid collaborative map is triggered. When the priority score exceeds the second adaptive threshold, a global forced update of the hybrid collaborative map is triggered.

[0012] Furthermore, after matching the priority score with a preset adaptive threshold and dynamically updating the hybrid collaborative map based on the matching result, the process further includes: When robot pose tracking loss is detected, the intersection-union ratio between the updated ellipsoidal projection and the two-dimensional projection in the dynamically updated hybrid cooperative map is calculated. The robot pose is relocalized based on the intersection-union comparison, and distributed conflict resolution among multiple agents is performed based on the relocalized robot pose.

[0013] Furthermore, after relocalizing the robot pose based on the intersection-union comparison, the process further includes: Real-time statistics are collected on the repositioning success rate and positioning error of each robot in different spatial regions. If the relocation success rate and positioning error of the current spatial region exceed the preset warning threshold, the gradient descent method is introduced to adjust the weights.

[0014] Another embodiment of the present invention provides a dynamic priority collaborative map construction and intelligent update system, comprising: The data acquisition module is used to acquire robot inspection data in order to build a static feature dataset; A map building module is used to extract background features and key equipment features from the static feature dataset to build a hybrid collaborative map; The projection distance analysis module is used to acquire the static target currently observed by the robot and calculate the distribution distance between the corresponding ellipsoidal projection in the hybrid collaborative map and the corresponding two-dimensional projection of the static target. The criterion determination module is used to determine the multidimensional update criterion corresponding to each spatial region based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map; The scoring module is used to perform weighted fusion of the multidimensional update criterion and the distribution distance to quantify the priority score corresponding to each spatial region; The map update module is used to match the priority score with a preset adaptive threshold and dynamically update the hybrid collaborative map based on the matching result.

[0015] Another embodiment of the present invention provides a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the dynamic priority collaborative map construction and intelligent update method as described above.

[0016] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: This invention first relies on inspection data collected by various robots to construct high-purity static feature data, thereby reducing the amount of invalid data processing at the source. Based on the background structure and key equipment, a cross-modal hybrid collaborative map is constructed, which greatly compresses the storage and transmission volume of the semantic map and provides a standardized mathematical foundation for subsequent dynamic updates based on probability distribution. By calculating the distribution distance between the current observation and the map ellipsoidal projection, precise decoupling of view artifacts and real physical changes is achieved, avoiding invalid reconstruction. Multi-dimensional criteria are determined based on robot state and map area attributes and weighted by distribution distance. Under a set threshold, hierarchical map updates are achieved, which can comprehensively evaluate the update needs of each area, reduce communication load, and enhance the reliability of multi-robot collaboration in dynamic environments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the dynamic priority collaborative map construction and intelligent update method in one embodiment of the present invention; Figure 2 This is a schematic diagram of a hierarchical communication update process in one embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a dynamic priority collaborative map construction and intelligent update system in one embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0020] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0022] One embodiment of the present invention provides a method for dynamic priority collaborative map construction and intelligent update. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown illustrates a flowchart of a dynamic priority collaborative map construction and intelligent update method according to one embodiment of the present invention, including the following steps: S1. Obtain robot inspection data to construct a static feature dataset.

[0023] This embodiment focuses on multi-modal, air-to-ground swarm robots / robots, using them as multiple agents to collaboratively build and update maps in dynamic environments. This step involves heterogeneous, multi-source data acquisition and feature analysis.

[0024] In some implementations of this embodiment, visual images and lidar data reflecting environmental depth are collected by relying on the RGB-D camera and lidar carried by the air-to-ground cross-modal swarm robot. The visual images and lidar data are then time-stamped and aligned based on the PTP protocol (such as the IEEE1588PTP protocol) to obtain aligned data.

[0025] To prevent moving objects such as people and vehicles in the inspection environment from interfering with the mapping, dynamic feature removal using depth masks is further performed. Specifically, when processing images at the edges, alignment data is input into a lightweight edge detection network to identify dynamic objects and output corresponding 2D bounding boxes. In this embodiment, the YOLO lightweight detection network is preferred for identification.

[0026] By combining 2D bounding boxes with alignment data, a dynamic object mask is generated. Feature points within the dynamic object mask are then removed, retaining only the feature data of the static background and key environmental equipment, forming a static feature dataset. This reduces the amount of invalid data processing at the source. In some implementations of this embodiment, let the first... i The depth value of the center of the dynamic target bounding box is To preserve any static background features that may exist within the bounding box, the depth front and back boundaries of the dynamic target are set: and Specifically, it is represented by the following formula: In the formula, The depth distance threshold is preset based on the scene (preferably 1.0 m in this embodiment).

[0027] Based on this, the first is defined i Mask function for a dynamic target bounding box Specifically, it is represented by the following formula: mask ( x , y )= In the formula, For pixels ( x , y The actual depth value. In this embodiment, it will be directly discarded. Visual and point cloud features within the region, only The static feature points participate in the subsequent mapping process, effectively eliminating dynamic trajectory drift from the source.

[0028] S2. Extract background features and key equipment features from the static feature dataset to construct a hybrid collaborative map.

[0029] To reduce the communication bandwidth load during multi-machine collaborative mapping, this real-time system abandons the traditional full point cloud transmission and designs a hybrid map architecture.

[0030] Specifically, for irregular background features such as walls and ground in static feature datasets, a three-dimensional spatial voxel layer is directly constructed to preserve the underlying geometric occupancy state.

[0031] For critical equipment such as distribution cabinets, valves, and transformers, this method first extracts 2D bounding boxes from multiple perspectives in the dual space. Then, based on camera projection geometry constraints, it constructs a homogeneous linear equation system of 2D observations and 3D spatial parameters. That is, the constraint relationship corresponding to the multi-view 2D bounding boxes refers to the projection geometry constraints between the equipment's 2D bounding box and the 3D ellipsoid matrix, specifically represented by a homogeneous linear equation system. By jointly solving and performing singular value decomposition on the multi-view 2D bounding boxes, the discrete 2D image features are reconstructed into a parameterized 3D ellipsoid layer with probability distribution characteristics, thereby achieving efficient semantic modeling of critical equipment. f In frame image observation, let the inscribed ellipse matrix of the 2D bounding box of the key equipment be... The 3D ellipsoidal matrix of the key device in dual space was reconstructed by solving a system of homogeneous linear equations. The homogeneous linear equation system is specifically represented as follows: In the formula, The coefficient matrix is ​​calculated from the camera projection matrix; vector Includes the target ellipsoid The 10 independent matrix parameters are obtained. By solving the above equation through singular value decomposition, the parameterized 3D ellipsoid matrix is ​​obtained, thereby reconstructing the parameterized 3D ellipsoid layer.

[0032] Finally, the 3D spatial voxel layer is overlaid with the parameterized 3D ellipsoid layer to form the required hybrid collaborative map. It is understandable that in some implementations of this embodiment, during cloud-edge communication, for a complex device, the edge only needs to send 10 floating-point parameters (including position, scale, and orientation) representing the 3D ellipsoid to the cloud, instead of tens of thousands of point cloud data points, thus compressing the map data volume by three orders of magnitude.

[0033] S3. Obtain the static target currently observed by the robot and calculate the distribution distance between the corresponding ellipsoidal projection in the hybrid cooperative map and the corresponding two-dimensional projection of the static target.

[0034] To further coordinate the real-time evaluation of various spatial regions in the hybrid collaborative map by the cloud-based perception fusion center and quantify the urgency of updates, this embodiment proposes a four-dimensional update criterion that couples physical spatiotemporal attributes, task orientation, and probabilistic manifolds in some implementations: information timeliness, task criticality, perception credibility, and asymmetric viewpoint-invariant geometric-semantic change rate.

[0035] This step aims to analyze the determination process of the key criterion "asymmetric viewpoint invariant geometric-semantic change rate", which is mainly used to eliminate "view artifacts" generated when heterogeneous multi-robot cross-view observation.

[0036] Specifically, the latest static target observed by each robot at the edge is first extracted, and then the two-dimensional projection corresponding to the static target is used as a two-dimensional Gaussian distribution. .

[0037] Next, the parameterized 3D ellipsoid layer in the hybrid collaborative map is invoked. Based on the real-time camera pose of the current agent, the ellipsoidal projection of the 3D ellipsoid onto the image plane is established, and similarly, it serves as the corresponding two-dimensional projection distribution. .

[0038] Finally, the Wasserstein distance between the two-dimensional Gaussian distribution and the ellipsoidal projection is calculated. Specifically, it is expressed as follows: = +tr In the formula, For asymmetric viewpoint-invariant geometric-semantic rate of change; These are the projection centers for the two-dimensional projection and the ellipsoidal projection, respectively. Let be the projection covariance matrix.

[0039] It should be understood that, due to the projection covariance matrix The camera's intrinsic parameter matrix has been analytically embedded in its construction. Camera rotation matrix And the shape parameters of the ellipsoid itself. Therefore, this covariance matrix is ​​a mathematical expression that is view-dependent but strictly bound to the physical properties of the device. Based on this mathematical characteristic, the Wasserstein distance can effectively distinguish between "view artifacts" and "real physical changes," specifically: When the robot's perspective rotates (For example, when a drone turns or a robot dog tilts its body) the projection center Covariance The change conforms to an equivalent conjugate transformation, and the calculated value at this time... If the system keeps the changes within a very low threshold, it identifies them as "not due to environmental changes and no update is needed"; however, when the geometry of the equipment itself changes due to factors such as the distribution cabinet door being opened or the transformer being moved, updates are not required. and This generates a nonlinear, irreversible deviation, making Exponential growth, thereby precisely triggering local updates.

[0040] S4~S5. Based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map, determine the multidimensional update criteria corresponding to each spatial region; perform weighted fusion of the multidimensional update criteria and distribution distance to quantify and obtain the priority score corresponding to each spatial region.

[0041] Next, this step aims to provide the process for determining the three sets of update criteria: information timeliness, task criticality, and perceived credibility.

[0042] In this embodiment, the information timeliness criterion is defined as the time decay factor of the most recent update in each spatial region. ,in, This represents the time interval since the last update for this region. is the time decay constant.

[0043] For the task criticality criterion, this embodiment uses the current robot's planned path P and each spatial region as the basis. The weights are calculated using the Euclidean distance d, defined as follows: This is used to characterize the dependency weight of the current collaborative planning path on the target within that spatial region. The specific calculation formula is as follows: In the formula, This is a modulating factor characterizing the range of influence. The model uses a Gaussian decay mechanism to ensure that the robot assigns higher weights to targets in areas adjacent to the planned path, thereby achieving dynamic optimization of resource allocation in collaborative tasks.

[0044] For the perception confidence criterion, this embodiment defines E as the cross-coverage rate of the robot's observation viewpoints for each spatial region and the sensor confidence level. trust .

[0045] Specifically, this embodiment evaluates perception quality by quantifying the redundancy of multi-view observations and the accuracy of sensor measurements. The specific calculation process is as follows: First, calculate the cross-coverage of the observation viewpoints. (The first term in the following formula) is obtained by statistically analyzing the target area from various perspectives. Overlapping coverage area First, the redundancy of the sensed information is characterized using an exponential decay function; second, the confidence level of the sensor is calculated. (The second term in the following equation) is the trace of the covariance matrix based on the observed data. Measuring measurement uncertainty. Ultimately, perceiving credibility. Defined as a weighted combination of the two, it is expressed as follows: ( i )= + In the formula, For preset weighting coefficients, This is the coverage gain factor. This criterion can effectively suppress perception bias caused by single-view occlusion or sensor noise, ensuring the robustness of collaborative planning.

[0046] Next, a comprehensive analysis , E trust as well as These four criteria are weighted and fused to quantify the results for each spatial region. k Corresponding priority score P update ( k Specifically, it is expressed as follows: P update ( k )= α 1E time + α 2E task + α 3E change + α 4(1−E trust ) In the formula, These are the normalized weighting coefficients.

[0047] Based on this, the fusion center is based on the full graph. The distribution determines the priority for dynamic updates of the hybrid collaborative map. Understandably, this embodiment employs a weighted optimization algorithm to quantify the "update urgency" of each spatial region in the map. Due to the introduction of a parameterized ellipsoidal difference metric, the system can accurately distinguish regions where the background remains unchanged but the target device state changes, generating a dynamically updated priority map mapped to the environmental space, thereby reflecting the real demand for computing power and communication resources at the current moment.

[0048] S6. Match the priority score with the preset adaptive threshold, and dynamically update the hybrid collaborative map based on the matching result.

[0049] Scoring based on priority P update Set a low-first adaptive threshold ( ), High-second adaptive threshold ( Two adaptive thresholds are used to perform hierarchical communication updates. For detailed update procedures, please refer to [reference needed]. Figure 2 As shown: When the priority score is lower than the first adaptive threshold, i.e. ( When this occurs, it is considered a low priority, triggering local maintenance tracking of the hybrid collaborative map. Each agent only maintains the odometer locally, and network communication is silent.

[0050] When the priority score is between the first adaptive threshold and the second adaptive threshold, i.e. ( When this occurs, it is considered a medium priority, triggering a parameterized local update of the hybrid collaborative map: for example, the agent only updates the "ellipsoidal matrix" of statically changed equipment (such as an opened power distribution cabinet door). The parameters (position, scale, attitude) are incrementally uploaded to the cloud, and the cloud distributes the matrix to other robots, achieving collaborative updates with extremely low bandwidth consumption.

[0051] When the priority score exceeds the second adaptive threshold, i.e. ( When a region experiences a sudden structural change (such as a landslide or the relocation of large equipment) and is the core area of ​​the current task, it is considered a high priority area; triggering a global forced update of the hybrid collaborative map: such as scheduling nearby drones in the cloud to perform a panoramic RGB-D point cloud dense rescan.

[0052] In some implementations of this embodiment, when pose tracking is lost due to strong electromagnetic interference or drastic changes in perspective (such as a quadruped robot suddenly falling and standing up), the agent invokes the updated 3D ellipsoid in the hybrid cooperative map for assisted relocalization. Specifically, in this embodiment, the P3P algorithm is preferably used to estimate candidate poses. First, the updated ellipsoid projection corresponding to the dynamically updated 3D ellipsoid is calculated. With two-dimensional projection Intersection over Union (IoU) score The robot's pose is relocalized, as shown below: Then, distributed conflict resolution among multiple agents is performed based on the relocalized robot pose. Understandably, after relocalization, each agent relies on a highly timely hybrid collaborative map and uses distributed model prediction (MPC) control to perform intentional interactions. That is, each agent uses a distributed model prediction control algorithm to iterate the cost function, resolving trajectory conflicts between them in narrow passageways (such as substation equipment rooms).

[0053] In some implementations of this embodiment, a feedback closed-loop mechanism is also included. Specifically, the cloud-based fusion center continuously monitors the relocalization success rate and localization error (RMSE) of each robot in different spatial regions. If the relocalization success rate and localization error for the current spatial region exceed a preset warning threshold, a gradient descent method is introduced to adjust the weights in the priority scoring formula. This increases the sensitivity to changes in the region, enabling long-term closed-loop self-learning and stable evolution of the robot collaborative mapping system.

[0054] To verify the effectiveness of this invention in air-to-ground cross-modal multi-robot collaborative operations, the following example is provided for detailed description. This example establishes a test platform including a quadruped robot, a drone, and an edge computing server. The sensor configuration consists of an RGB-D camera and a lightweight LiDAR, and communication tests are conducted under a limited bandwidth local area network. The test datasets used are: (1) the TUM RGB-D public dataset; and (2) a self-built substation multimodal collaborative inspection dataset (Substation-Multi).

[0055] The specific configuration of the self-built dataset Substation-Multi is as follows: 1) Data acquisition equipment: DJI M300 RTK drone, Unitree Go2 quadruped robot, Intel RealSense D435i depth camera, Livox Mid-360 LiDAR; 2) Data collection scenario: A 220kV substation in Hangzhou, containing typical power equipment such as distribution cabinets, transformers, circuit breakers, and overhead lines, with a site area of ​​approximately 5,000 square meters; 3) Data scale: 10 sequences, total duration 45 minutes, including extreme conditions such as dynamic personnel occlusion, sudden changes in viewpoint, and strong electromagnetic interference; 4) Annotation method: The 3D bounding boxes and semantic categories of static devices were manually annotated, with a total of 1200 target instances annotated.

[0056] 1. Comparative Experiment To verify the lightweight advantages of the "voxel-parametric ellipsoid" hybrid map architecture in this example, dense point cloud SLAM, OctoMap, and OA-SLAM were selected as baseline models for comparison. Please refer to Table 1 below.

[0057] Table 1 Performance comparison of different collaborative mapping methods in dynamic substation scenarios. As shown in Table 1, in highly dynamic scenarios, this example significantly reduces data redundancy in collaborative mapping by filtering out dynamic objects and converting static key equipment into parametric ellipsoids. Compared to traditional methods, the peak communication bandwidth of this invention is reduced to 1.85 Mbps, and the map memory usage is only 15.8 MB, achieving a high proportion of data compression. Simultaneously, the lightweight map architecture reduces the average keyframe processing latency to 28.5 ms and the absolute trajectory error (ATE) to 0.052 m, achieving an optimal balance between mapping accuracy and real-time performance in resource-constrained environments.

[0058] Furthermore, to verify the robustness of the relocation mechanism of this invention, a traditional relocation method based on the visual bag-of-words (BoW) was selected as a comparison, and tests were conducted under three typical extreme conditions, as shown in Table 2 below.

[0059] Table 2 shows that the traditional BoW method heavily relies on low-level pixel matching, and its success rate is only 15.8% when faced with sudden changes in viewpoint (condition B). This invention utilizes a static device ellipsoid model for object-level data association. Because the 3D ellipsoid possesses spatial rotation invariance and high-level semantic features, the repositioning success rate remains stable above 88% under various extreme conditions, with rotation errors controlled within 5°, effectively overcoming the pose loss problem in complex environments.

[0060] Table 2 Comparison of relocation performance under different severe disturbance conditions 2. Ablation experiment To verify the effectiveness of each criterion combination in the on-demand triggering mechanism, multi-dimensional ablation tests were conducted, as detailed in Table 3 below.

[0061] Table 3 Ablation experimental results of different criterion combinations in the on-demand triggering mechanism As shown in Table 3, traditional global uniform updates suffer from significant redundancy, with an end-to-end latency as high as 520 ms. By gradually introducing timeliness and task criticality, the number of redundant triggers significantly decreases. Furthermore, by introducing the "geometric-semantic change rate based on Wasserstein distance" of this invention, the system effectively filters pixel artifact interference, accurately identifies real physical changes, and reduces the number of redundant triggers to 21 times / min. Under the complete 4D criterion-driven approach, the system achieves accurate local on-demand updates, reducing communication latency to 68 ms, and the pose drift in the core task area is only 0.038 m.

[0062] Furthermore, to verify the impact of front-end dynamic feature removal on the quality of static device mapping, a comparison was made under different dynamic scenarios, as shown in Table 4 below.

[0063] Table 4. Impact of Dynamic Mask Removal Strategy on Reconstruction Quality of Key Equipment As shown in Table 4, in highly dynamic cross-operation scenarios, without dynamic removal, incorrectly registered dynamic features can lead to severe trajectory drift (ATE reaching 0.282 m) and ellipsoid reconstruction failure. This invention uses dynamic masking to filter out interfering features, ensuring consistency in pose estimation across multiple viewpoints. After removing dynamic features, the ATE in highly dynamic scenarios drops to 0.039 m, the number of static ellipsoid reconstructions increases from 1 to 6, and the average IoU reaches 0.67, ensuring the data reliability of the collaborative map.

[0064] In summary, the embodiments of the present invention construct a lightweight hybrid collaborative map of "voxels + ellipsoids" and introduce a hierarchical map update mechanism based on multi-dimensional criteria weighting and driving, which can greatly reduce the bandwidth and computing power overhead of multi-machine communication, significantly improve the positioning robustness in complex environments, and provide support for efficient collaboration and intelligent inspection of multiple agents.

[0065] One embodiment of the present invention provides a dynamic priority collaborative map construction and intelligent update system. For details, please refer to [link to documentation]. Figure 3 , Figure 3 The diagram shown illustrates the structure of a dynamic priority collaborative map construction and intelligent update system according to one embodiment of the present invention, including: The data acquisition module M1 is used to acquire robot inspection data to construct a static feature dataset; The map building module M2 is used to extract background features and key equipment features from the static feature dataset to build a hybrid collaborative map; The projection distance analysis module M3 is used to acquire the static target currently observed by the robot and calculate the distribution distance between the corresponding ellipsoidal projection in the hybrid collaborative map and the corresponding two-dimensional projection of the static target. The criterion determination module M4 is used to determine the multidimensional update criteria corresponding to each spatial region based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map. The scoring module M5 is used to weight and fuse the multidimensional update criterion and the distribution distance to quantify the priority score corresponding to each spatial region. The map update module M6 is used to match the priority score with a preset adaptive threshold and dynamically update the hybrid collaborative map based on the matching result.

[0066] In one embodiment of this application, a computer device is provided, which includes a memory and a processor. The memory stores machine-executable instructions that can be executed by the processor. When the processor executes a computer program, it implements the above steps. The computer device provided in this embodiment has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0067] The technical features and effects of the dynamic priority collaborative map construction and intelligent update system proposed in this embodiment of the invention are the same as those of the dynamic priority collaborative map construction and intelligent update method proposed in this embodiment of the invention, and will not be repeated here.

[0068] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for dynamic priority collaborative map construction and intelligent updating, characterized in that, include: Acquire robot inspection data to construct a static feature dataset; Background features and key equipment features are extracted from the static feature dataset to construct a hybrid collaborative map; Obtain the static target currently observed by the robot, and calculate the distribution distance between the corresponding ellipsoidal projection in the hybrid cooperative map and the corresponding two-dimensional projection of the static target; Based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map, determine the multidimensional update criteria corresponding to each spatial region; The multidimensional update criterion and the distribution distance are weighted and fused to quantify the priority score corresponding to each spatial region; The priority score is matched with a preset adaptive threshold, and the hybrid collaborative map is dynamically updated based on the matching result.

2. The dynamic priority collaborative map construction and intelligent update method as described in claim 1, characterized in that, The acquisition of robot inspection data to construct a static feature dataset includes: The visual images and lidar data collected during robot inspection are acquired, and the timestamps of the visual images and lidar data are aligned to obtain aligned data. The alignment data is input into a lightweight edge detection network to identify dynamic objects and output the corresponding 2D bounding boxes. A dynamic object mask is generated by combining the 2D bounding box with the alignment data; Remove feature points within the dynamic object mask and output the static feature dataset.

3. The dynamic priority collaborative map construction and intelligent update method as described in claim 1, characterized in that, The process of constructing the hybrid collaborative map includes: Using the aforementioned background features, a three-dimensional spatial voxel layer is constructed; In the dual space, multi-view 2D bounding boxes are extracted based on the key equipment features; Using the constraint relationships corresponding to the multi-view 2D bounding boxes, the multi-view 2D bounding boxes are jointly solved and reconstructed into a parameterized 3D ellipsoid layer that reflects the probability distribution characteristics. The three-dimensional spatial voxel layer is superimposed with the parameterized 3D ellipsoid layer to form the hybrid collaborative map.

4. The dynamic priority collaborative map construction and intelligent update method as described in claim 3, characterized in that, The calculation of the distribution distance between the corresponding ellipsoidal projection in the hybrid collaborative map and the corresponding two-dimensional projection of the static target includes: The two-dimensional projection corresponding to the static target is taken as a two-dimensional Gaussian distribution; Extract the parameterized 3D ellipsoid layer from the hybrid collaborative map and establish the corresponding ellipsoid projection distribution; Calculate the Wasserstein distance between the two-dimensional Gaussian distribution and the ellipsoidal projection.

5. The dynamic priority collaborative map construction and intelligent update method as described in claim 1, characterized in that, The step of determining the multidimensional update criteria corresponding to each spatial region based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map includes: The information timeliness criterion is calculated based on the time decay factor of the most recent update in each of the aforementioned spatial regions. The weights are calculated based on the Euclidean distance between the current robot's planned path and each of the aforementioned spatial regions, and are used as criteria for determining the criticality of the task. The perception credibility criterion is obtained by statistically calculating the cross-coverage rate of the robot's observation perspectives of each spatial region and the sensor confidence level.

6. The dynamic priority collaborative map construction and intelligent update method as described in claim 1, characterized in that, The step of matching the priority score with a preset adaptive threshold and dynamically updating the hybrid collaborative map based on the matching result includes: When the priority score is lower than the first adaptive threshold, local maintenance tracking of the hybrid collaborative map is triggered. When the priority score is between the first adaptive threshold and the second adaptive threshold, a parameterized local update of the hybrid collaborative map is triggered. When the priority score exceeds the second adaptive threshold, a global forced update of the hybrid collaborative map is triggered.

7. The dynamic priority collaborative map construction and intelligent update method as described in claim 1, characterized in that, After matching the priority score with a preset adaptive threshold and dynamically updating the hybrid collaborative map based on the matching result, the method further includes: When robot pose tracking loss is detected, the intersection-union ratio between the updated ellipsoidal projection and the two-dimensional projection in the dynamically updated hybrid cooperative map is calculated. The robot pose is relocalized based on the intersection-union comparison, and distributed conflict resolution among multiple agents is performed based on the relocalized robot pose.

8. The dynamic priority collaborative map construction and intelligent update method as described in claim 7, characterized in that, After relocalizing the robot pose based on the intersection-union ratio, the process further includes: Real-time statistics are collected on the repositioning success rate and positioning error of each robot in different spatial regions. If the relocation success rate and positioning error of the current spatial region exceed the preset warning threshold, the gradient descent method is introduced to adjust the weights.

9. A dynamic priority collaborative map construction and intelligent update system, characterized in that, include: The data acquisition module is used to acquire robot inspection data in order to build a static feature dataset; A map building module is used to extract background features and key equipment features from the static feature dataset to build a hybrid collaborative map; The projection distance analysis module is used to acquire the static target currently observed by the robot and calculate the distribution distance between the corresponding ellipsoidal projection in the hybrid collaborative map and the corresponding two-dimensional projection of the static target. The criterion determination module is used to determine the multidimensional update criterion corresponding to each spatial region based on the current robot's operating status information and the attribute information of each spatial region in the hybrid collaborative map; The scoring module is used to perform weighted fusion of the multidimensional update criterion and the distribution distance to quantify the priority score corresponding to each spatial region; The map update module is used to match the priority score with a preset adaptive threshold and dynamically update the hybrid collaborative map based on the matching result.

10. A computer device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the dynamic priority collaborative map construction and intelligent update method as described in any one of claims 1 to 8.

Images