Construction site intelligent navigation and task allocation method based on BIM

Abstract

The invention provides a BIM-based construction site intelligent navigation and task allocation method, and belongs to the technical field of building informatization and robot automatic construction. According to the method, semantic information such as a wall material, a construction period and room information is extracted from a BIM model, construction key nodes such as doors and windows, inflection points and wall positions are detected from an image in combination with a neural network model, and feature vectors are extracted and fused with the semantic information. Then, through multi-dimensional layering and dynamic weight distribution, a layered topological map based on materials, rooms and construction periods is constructed, and clustering of construction nodes and selection of center nodes are achieved; the topological map is combined with the element controllers and the sub-controllers, efficient path planning and obstacle avoidance control are achieved, and construction efficiency and safety are improved. The construction sequence is obviously optimized, the navigation cost and the energy consumption are reduced, and the method is effectively suitable for a complex construction environment.

Description

Technical Field

[0001] This invention relates to the fields of building information modeling (BIM) and robotic automated construction technology, specifically a BIM-based intelligent navigation and task allocation method for construction sites. Background Technology

[0002] With the informatization and digitalization of the construction industry, Building Information Modeling (BIM) technology has been widely applied in my country. BIM technology provides rich information support for design, construction, and operation by constructing three-dimensional building models. However, how to efficiently utilize the information in the BIM model to enable indoor construction robots to efficiently plan for complex indoor construction problems remains a problem that urgently needs to be solved.

[0003] Currently, navigation and task allocation at construction sites suffer from several problems: unscientific construction path planning, often based on experience, makes it difficult to select the optimal path, leading to high energy consumption and low efficiency during construction. Furthermore, chaotic task scheduling, lacking optimization, easily results in unreasonable task execution sequences, impacting construction progress.

[0004] To address these issues, researchers have proposed various solutions, such as construction resource optimization allocation methods based on genetic algorithms and construction path planning methods based on ant colony algorithms. However, these methods still suffer from limitations in practical applications, including high model complexity, slow computation speed, and over-reliance on empirical parameters, resulting in poor generalizability.

[0005] Technical comparison with Chinese patent CN120147573A, "A method for extracting key features and constructing topology maps of 3D maps of substation construction sites":

[0006] The data sources and semantic information processing methods differ. This application (Intelligent Navigation and Task Allocation Method for Construction Sites Based on BIM) extracts information from Building Information Modeling (BIM), which itself is a structured data source containing rich semantics (such as wall material, construction period, and room function). In contrast, Chinese Patent CN120147573A is based on 3D point cloud and image data collected on-site. These are raw, unstructured data, and one of the core methods of this patent is to use the Mamba algorithm to extract key features and semantic information from this raw data. In short, this application utilizes existing design semantic information, while Chinese Patent CN120147573A focuses on inferring semantics from post-collected geometric data.

[0007] The core technologies and algorithms are fundamentally different. This application uses a method based on convolutional neural networks (ResNet-50 and CBAM) to identify building structure nodes such as doors, windows, and walls from BIM-converted images. Chinese patent CN120147573A, on the other hand, uses the Mamba algorithm (a state-space model) to process point cloud data to extract key information such as power equipment and transmission lines. The two are completely different in the deep learning models used for feature extraction.

[0008] The concepts and focuses of map construction differ. This application constructs a "layered topology map" to optimize the construction process. It layers, clusters, and dynamically weights nodes based on multiple dimensions such as construction period, material, and space, aiming to optimize the construction sequence and robot task allocation. In contrast, Chinese patent CN120147573A focuses on accurately and reliably reflecting the spatial geometric relationships of the site environment, and designs steps for reliability assessment of connection relationships and map optimization adjustments to ensure map accuracy.

[0009] The ultimate application goals differ. This application aims to provide a complete "intelligent navigation and task allocation system," whose constructed map serves as an intermediate tool for achieving efficient robot path planning and task scheduling. It also designs a meta-controller and sub-controllers to realize this closed-loop system. The core output of Chinese Patent CN120147573A is the "topology map construction method" itself. This map provides more reliable model support for subsequent applications such as path planning and target localization, but its implementation mainly focuses on the mapping stage. Summary of the Invention

[0010] To address the aforementioned technical challenges, this invention proposes a BIM-based intelligent navigation and task allocation method for construction sites. When processing BIM models, it addresses how to effectively extract semantic information of rooms and information on key construction nodes to facilitate subsequent construction planning and management. In complex construction environments, it addresses how to rationally organize construction nodes, achieving multi-dimensional layering and dynamic weight allocation based on nodes to optimize construction sequence, reduce construction costs, and improve construction efficiency. Furthermore, it addresses how to construct a layered topology map based on different node attributes, giving it the advantage of considering factors such as wall material, room type, and project urgency, thereby supporting efficient construction management and decision-making by the robot. This invention aims to provide a navigation framework for efficient navigation of building interior maintenance robots based on BIM information.

[0011] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0012] The BIM-based intelligent navigation and task allocation method for construction sites comprises the following steps:

[0013] Step 1: Process the BIM model to obtain room semantic information and key construction node information, and combine the nodes with semantic information to extract indoor building information;

[0014] The process begins by exporting IFC information from the BIM model, reading the IFC file using an IFC library, loading the model data, and extracting attributes from relevant entities via scripts. Extracted information includes wall material, construction period, and semantic information about rooms. Next, the BIM model is converted into images using an automated script combined with a 3D rendering engine to transform the model into overall building and room images. These images are then saved after camera view settings and batch rendering. Subsequently, ResNet-50 and CBAM attention mechanisms are used to detect key nodes corresponding to door and window locations, room corners, and wall positions from the images, extracting feature vectors. An attention mechanism is then used to emphasize important parts of the input features through a weight matrix. Finally, the identified nodes are combined with their location and semantic information.

[0015] Step 2: Construct a hierarchical topology map by multi-dimensional layering of nodes, clustering and dynamic weight allocation, combined with construction information;

[0016] In the process of constructing the layered topology map, the node set was extracted from the BIM model. Considering that walls of different materials need to be constructed separately, and that it is more convenient to construct nodes in the same room, as well as the principle of tight schedule, the construction sequence was arranged. Then, the nodes of each group were agglomerated into dense clusters through clustering algorithm, and the central node of each group was calculated. Then, using dynamic weight information, these central nodes were merged into the main nodes of the topology map to ensure its advantages in terms of materials, rooms and schedule. By verifying the reachability relationship, other nodes were connected to the main node as secondary nodes, while eliminating connections that cross obstacles. Finally, the accurate construction of the layered topology map was completed.

[0017] Step 3: Establish the robot's kinematic model, design a hierarchical controller, the meta-controller selects the main node and secondary node in the hierarchical topology map, and the sub-controller performs obstacle avoidance to ensure collision-free arrival at the sub-target and complete the navigation objective;

[0018] During the process of establishing the robot's kinematic model, the meta-controller prioritizes nearby construction nodes through regional compactness analysis and selects secondary nodes that are close to the main node and have tight deadlines by adopting a neighborhood node priority strategy. Following the principle of unified construction procedures, nodes with consistent materials and task types are given priority. Finally, the shortest path is evaluated and selected. The sub-controller is responsible for obstacle avoidance to achieve collision-free arrival at the selected target node. Based on the current state and the selected node, the sub-controller perceives obstacles in the environment in real time, generates a collision-free trajectory, and controls the robot's movement based on the robot's kinematic model to safely navigate to the target node.

[0019] As a further improvement to the present invention, the specific process of step one is as follows:

[0020] Given an input image Feature maps are extracted using ResNet-50. , denoted as:

[0021] ;

[0022] in This represents a ResNet-50 convolutional neural network used for image feature extraction. It contains 50 layers and uses residual blocks to address the vanishing gradient problem in deep networks. One residual block contains three convolutional layers. Assume the input is... The output of the residual block for:

[0023] ;

[0024] in It is a residual function, and its specific structure is as follows:

[0025]

[0026] Here These are the weights of the three convolutional layers. It is the corresponding bias term;

[0027] Feature enhancement is performed using CBAM to obtain the enhanced feature map. :

[0028] ;

[0029] in The attention module represents a system that comprises two modules: a channel attention module and a spatial attention module. for:

[0030] ;

[0031] in It is the Sigmoid function. It is a multilayer perceptron. and These are global average pooling and global max pooling, respectively.

[0032] Spatial Attention Module F s calculate:

[0033] ;

[0034] in, It's a convolution operation. It's a splicing operation. and These are average pooling and max pooling, respectively.

[0035] The final weighted feature map F' is:

[0036] ;

[0037] in This represents element-wise multiplication;

[0038] Based on the obtained feature vector of the target, the target is identified and its two-dimensional position coordinates in the image are determined. Assuming the detected node is... Its position coordinates are By fusing location information with semantic information to create a comprehensive node description, this process can be represented as follows: For each node... Its comprehensive description Represented as a tuple containing position coordinates and related semantic attributes: 1

[0039] ;

[0040] in, It is a node Position coordinates in the image, It is a collection containing semantic information;

[0041] ;

[0042] Combine the comprehensive descriptions of all nodes to form a complete model description. :

[0043] ;

[0044] in This will increase the total number of detected nodes.

[0045] As a further improvement to the present invention, step two is specifically as follows:

[0046] First, all nodes are divided into multiple levels based on their different attributes. These attributes include the wall material, the room the node belongs to, and the urgency of the construction period. Each subset contains nodes with the same attributes. The set of all nodes. Represented as a tuple, For having the same semantics If the set of nodes is a subset, then:

[0047] ;

[0048] in, These are the room type, wall material, and construction period attributes;

[0049] A subset of nodes for each semantic attribute Within, clustering algorithms are applied to divide the nodes into several clusters. Let... Grouping The number of clusters, For the first Each cluster:

[0050] ;

[0051] in, It is a node The index of the assigned cluster;

[0052] For each cluster Calculate its central node The central node is the average of the spatial location attribute values ​​of all nodes within the cluster, representing the characteristics of the cluster. Let... For the first The central node of each cluster, then:

[0053] ;

[0054] in, Cluster The number of nodes in Represents a node The coordinates;

[0055] To calculate the master node of the topology map based on the central node of each cluster, a dynamic weight needs to be assigned to each central node. This weight is based on multiple factors, including the urgency of the construction period, room type, and wall material. The total number of weighting factors. For the first The weight vector of each weight factor. As the central node In the The score on each weighted factor, then the center node Overall weight for:

[0056] ;

[0057] An optimization method is used to calculate the master nodes of the topology map based on the hierarchical center nodes. The goal is to find the locations that minimize the weighted distances between nodes in the entire topology map. and They are the central nodes. and The weight, As the central node and The distance between them is then used to optimize the problem as follows:

[0058] ;

[0059] The reachability between primary nodes is verified using a path search algorithm from graph theory to ensure that all primary nodes are interconnected. Then, primary nodes are connected to secondary nodes to form a topology map. Represents a topological map. Represents the set of master nodes. Let the set of edges be:

[0060] ;

[0061] Ultimately, a hierarchical topology is constructed that takes into account multiple attributes, ensuring that the task faces advantages under constraints in terms of materials, rooms, and schedules.

[0062] As a further improvement to the present invention, in step two, during the process of constructing the hierarchical topology map, starting from the node set extracted from the BIM model, considering that walls of different materials need to be constructed separately, that it is more convenient to construct nodes in the same room, and the principle of urgency in the construction period, the construction sequence is arranged as follows:

[0063] The nodes are divided into several layers: spatial layering, grouping nodes according to room information to facilitate efficient construction of nodes in the same room; material layering, dividing nodes into several groups according to different wall materials to ensure consistency in treatment methods and materials used during construction; and schedule layering, sorting and grouping nodes according to the urgency of the schedule to ensure reasonable allocation of construction progress.

[0064] As a further improvement to the present invention, the step of the ternary controller selecting the primary node and secondary node in the hierarchical topology map is as follows:

[0065] First, a regional compactness analysis is performed, prioritizing nodes with concentrated construction tasks and close spatial distances, and a compactness function is defined. To evaluate the compactness of nodes within a certain region, let the set of principal nodes within that region be used. ,node and The distance between them is , For nodes The amount of construction work. Then:

[0066] ;

[0067] in, The smaller the value, the more likely it is to be a node. The more concentrated the construction tasks, the closer the spatial distance, and the more compact the node distribution;

[0068] Then, the meta controller according to The value of the master node Choice:

[0069] ;

[0070] in Refers to a node in set N;

[0071] Next, a neighborhood node priority strategy is adopted, prioritizing secondary nodes that are close to the primary node and have tight deadlines to reduce navigation costs. For nodes The urgency of the construction period The distance between the primary node and the secondary node determines the priority of the secondary node. for:

[0072] ;

[0073] The meta controller determines the priority weight of each secondary node. Sort the nodes and select the set of secondary nodes with higher priority weights. :

[0074] ;

[0075] Following the principle of unified construction procedures, priority is given to processing nodes with the same materials and consistent task types to reduce the need for switching robot working modes. Task types are defined accordingly. and wall material Representing nodes respectively Based on the construction information, construct a process similarity function between nodes. :

[0076] ;

[0077] in, For indicator functions, The larger the value, the more consistent the node's material and task type are with the main node. Adjust the selection of secondary nodes based on the results. ,in Represents a node If the processes are similar, then:

[0078] ;

[0079] Finally, the total travel path is evaluated, assessing the travel paths from the current node to the primary and secondary nodes, selecting the shortest path to reduce energy consumption, and calculating the total travel path. :

[0080] ;

[0081] in Given the distance between two nodes, optimize the combination of secondary nodes based on the result. :

[0082] ;

[0083] Ultimately, the meta-controller selects the optimal master node. and secondary node set .

[0084] As a further improvement of the present invention, in step three, the sub-controller performs obstacle avoidance to ensure collision-free arrival at the sub-target. The specific process is as follows:

[0085] The sub-controller is responsible for generating collision-free trajectories and controlling the robot to safely reach the sub-target. First, based on the sub-target selected by the meta controller. Position coordinates Convert it to the target position in the local obstacle avoidance coordinate system. The specific conversion formula is as follows:

[0086] ;

[0087] in, This is the robot's current orientation angle;

[0088] Next, based on the target position in the local obstacle avoidance coordinate system The target direction that the computational robot should face. :

[0089] ;

[0090] Target direction This refers to the direction the robot should move in unobstructed conditions.

[0091] During obstacle avoidance, the robot needs to dynamically adjust its angular velocity based on the distribution of obstacles. Assuming the robot uses differential drive, its linear velocity... Angular velocity is a constant. The calculation process is as follows:

[0092] Select robot orientation In the direction of the obstacle avoidance coordinate system, a forward-looking point is selected, which is 1 unit away from the origin. At this point, we obtain:

[0093] ;

[0094] Where R is the radius of the arc formed by the current coordinates and the look-ahead point, which is the turning radius of the differential drive robot. Therefore:

[0095] ;

[0096] Based on the turning radius of the differential drive robot and linear velocity Relationship Calculate angular velocity :

[0097] ;

[0098] in, The target angle that the robot should turn to;

[0099] To prevent excessive fluctuations in the robot's angular velocity during obstacle avoidance, a first-order IIR filter is used to control the angular velocity. After smoothing, the difference equation of the filter is:

[0100] ;

[0101] in, The angular velocity at the previous moment. The new angular velocity calculated at the current moment. The filtered angular velocity. This is the weighting constant.

[0102] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0103] This invention proposes a BIM-based intelligent navigation and task allocation method for construction sites. The system dynamically allocates tasks to robots using intelligent algorithms based on the construction schedule, resource distribution, and work priorities within the BIM model. This ensures that robots execute tasks in the optimal work sequence, reducing conflicts and waiting times during operations and improving overall work efficiency on the construction site. Through the construction of a hierarchical topology map, the construction sequence can be optimized based on attributes such as wall material, room type, and project urgency, achieving efficient node clustering and dynamic weight allocation, thus meeting the needs of complex construction tasks. Attached Figure Description

[0104] Figure 1 This is a flowchart of the method of the present invention;

[0105] Figure 2 This is a schematic diagram of the robot's motion control model.

[0106] Figure 3 This is a system framework diagram of the present invention;

[0107] Figure 4 This is the expected working path of the meta-control results in the embodiments of the present invention;

[0108] Figure 5 This is the actual execution path of the sub-controller in this embodiment of the invention. Detailed Implementation

[0109] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings:

[0110] like Figure 1 As shown, this invention provides a BIM-based intelligent navigation and task allocation system and method for construction sites. It employs a technical solution combining the construction of a hierarchical topology map containing semantic information based on a BIM model and the design of a hierarchical controller. Specifically, it includes the following steps:

[0111] Step 1: Process the BIM model to obtain room semantic information and key construction node information, and combine the nodes with semantic information to extract indoor building information;

[0112] The process begins by exporting IFC information from the BIM model. An IFC library is used to read the IFC file, load the model data, and a script is used to find relevant entities and extract their attributes. For example, material information is obtained from the IfcMaterial attribute associated with IfcWall. Extracted information includes semantic information such as wall material, construction period, and room details. Next, the BIM model is converted into images. An automated script combined with a 3D rendering engine converts the model into overall building and room images. These images are then processed and batch rendered and saved. Subsequently, ResNet-50 and CBAM attention mechanisms are used to detect key nodes such as doors, windows, room corners, and wall positions from the images. Feature vectors are extracted, and an attention mechanism is used to emphasize important parts of the input features through a weight matrix. Finally, the identified nodes are combined with their location and semantic information.

[0113] This process can be represented as: given an input image Feature maps are extracted using ResNet-50. , denoted as:

[0114]

[0115] in This represents the ResNet-50 convolutional neural network. It's used for image feature extraction, contains 50 layers, and uses residual blocks to address the vanishing gradient problem in deep networks. Each residual block contains three convolutional layers. Assume the input is... The output of the residual block is:

[0116]

[0117] in It is a residual function, and its specific structure is as follows:

[0118]

[0119] Here These are the weights of the three convolutional layers. This is the corresponding bias term.

[0120] Feature enhancement is performed using CBAM (Convolutional Block Attention Module) to obtain the enhanced feature map. :

[0121]

[0122] in The attention module consists of two parts: a channel attention module and a spatial attention module. The channel attention module is as follows:

[0123]

[0124] in It is the Sigmoid function. It is a multilayer perceptron. and These are global average pooling and global max pooling, respectively.

[0125] Spatial attention module calculation:

[0126]

[0127] in, It's a convolution operation. It's a splicing operation. and These are average pooling and max pooling, respectively.

[0128] The final weighted feature map is as follows:

[0129]

[0130] in This indicates element-wise multiplication.

[0131] Based on the obtained feature vector of the target, the target is identified and its two-dimensional position coordinates in the image are determined. Assume the detected node is... Its position coordinates are By fusing location information with semantic information, a comprehensive node description is created. This process can be represented as: for each node... Its comprehensive description It can be represented as a tuple containing the location coordinates and related semantic attributes:

[0132]

[0133] in, It is a node Position coordinates in the image. It is a collection containing semantic information.

[0134]

[0135] Combine the comprehensive descriptions of all nodes to form a complete model description. :

[0136]

[0137] in This will increase the total number of detected nodes.

[0138] Step 2: Construct a hierarchical topology map by multi-dimensional layering of nodes, clustering and dynamic weight allocation, combined with construction information;

[0139] In constructing the layered topology map, the process begins with extracting a set of nodes from the BIM model. Considering the need for separate construction of walls with different materials, the convenience of constructing nodes within the same room, and the urgency of the construction schedule, the construction sequence is arranged accordingly. Nodes are divided into several layers: spatial layering (grouping nodes according to room information to facilitate efficient construction of nodes within the same room); material layering (grouping nodes based on wall materials to ensure consistency in treatment methods and materials used during construction); and schedule layering (grouping nodes according to their urgency to ensure a reasonable allocation of construction time). Next, a clustering algorithm is used to agglomerate the nodes into dense clusters, and the central node of each group is calculated. Then, using dynamic weight information, these central nodes are merged into the main nodes of the topology map, ensuring their advantages in terms of materials, rooms, and schedule. By verifying reachability and connecting other nodes as secondary nodes to the main nodes, while excluding connections that traverse obstacles, the accurate construction of the layered topology map is finally completed. The specific process is as follows:

[0140] First, all nodes are divided into multiple levels based on their different attributes. These attributes include the node's wall material, the room it belongs to, and its construction urgency. Each subset contains nodes with the same attributes. Let... The set of all nodes. For having the same semantics If the set of nodes is a subset, then:

[0141]

[0142] in, These can be attributes such as room type, wall material, and construction period.

[0143] In each attribute subset Within the cluster, a clustering algorithm is applied to divide the nodes into several clusters. This ensures that nodes within the same cluster that are close together can be processed together during construction, improving construction efficiency. Let... Grouping The number of clusters, For the first Each cluster:

[0144]

[0145] in, It is a node The index of the assigned cluster.

[0146] For each cluster Calculate its central node The central node is the average of the spatial location attribute values ​​of all nodes within the cluster; it represents the characteristics of the cluster. Let... For the first The central node of each cluster, then:

[0147]

[0148] in, Cluster The number of nodes in Represents a node The coordinates.

[0149] To calculate the master nodes of the topology map based on the central nodes of each cluster, a dynamic weight needs to be assigned to each central node. This weight is based on multiple factors, such as project urgency, room type, and wall material. Let... The total number of weighting factors. For the first The weight vector of each weight factor. As the central node In the The score on each weighted factor, then the center node Overall weight for:

[0150]

[0151] An optimization method is used to calculate the master nodes of the topology map based on the hierarchical center nodes. The goal is to find the locations that minimize the weighted distances between nodes in the entire topology map; this problem can be understood as an optimization problem. Let... and They are the central nodes. and The weight, As the central node and Given the distance between them, the optimization problem can be expressed as:

[0152]

[0153] To ensure the validity of the topology map, the reachability between primary nodes needs to be verified. This is done using a pathfinding algorithm from graph theory to ensure that all primary nodes are interconnected. Then, the primary and secondary nodes are connected to form the topology map. Represents a topological map. Represents the set of master nodes. Let the set of edges be:

[0154]

[0155] Ultimately, a hierarchical topology is constructed that takes into account multiple attributes, ensuring that the task faces advantages under constraints in terms of materials, rooms, and schedules.

[0156] Step 3: Establish the robot's kinematic model, design a hierarchical controller, the meta-controller selects the main node and secondary node in the hierarchical topology map, and the sub-controller performs obstacle avoidance to ensure collision-free arrival at the sub-target and complete the navigation objective;

[0157] The process of establishing a kinematic model of a robot is as follows: Figure 2 As shown, the specific process is as follows:

[0158] The robot operates in a differential motion mode, and its state can be represented by a vector. It means that, among them and These represent the robot's x and y coordinates, respectively. This indicates the robot's orientation angle.

[0159] In order to track the desired trajectory, an error dynamics model needs to be established to describe the difference between the robot's actual state and the desired state, as shown in the following equation:

[0160]

[0161] in These represent the deviations between the robot's actual position and its desired position on the x-axis and y-axis, respectively. This indicates the deviation between the robot's actual orientation and its desired orientation.

[0162] To track the desired trajectory, the tracking error must be eliminated by calculating the robot's linear velocity. and angular velocity The calculation method is as follows:

[0163]

[0164] in These represent instantaneous linear velocity and angular velocity, respectively. It is the control input, which can be obtained according to a nonlinear control law, so as to make the robot move along the desired trajectory. The calculation is as follows:

[0165]

[0166] According to the linear control law The calculation is as follows:

[0167]

[0168] in This is the assumed gain.

[0169] The robot is differentially driven, therefore its instantaneous turning radius is... Determined by the following formula:

[0170]

[0171] The meta-controller prioritizes nearby construction nodes through area compactness analysis, avoiding frequent switching between multiple rooms. It also employs a neighborhood node priority strategy to select secondary nodes with tight deadlines and close to the main node, reducing repetitive navigation costs. Furthermore, adhering to the principle of unified construction procedures, it prioritizes nodes with consistent materials and task types, minimizing robot mode switching. Finally, it evaluates and selects the shortest path, reducing energy consumption and improving construction efficiency. The sub-controller, on the other hand, is responsible for obstacle avoidance to achieve collision-free arrival at the selected target node. Based on the current state and the selected node, the sub-controller perceives obstacles in the environment in real time, generates a collision-free trajectory, and controls the robot's movement based on robot kinematic modeling to safely navigate to the target node.

[0172] First, a regional compactness analysis is performed, prioritizing nodes with concentrated construction tasks and close spatial distances to reduce room-to-room switching. A compactness function is then defined. To evaluate the compactness of nodes within a certain region, let the set of principal nodes within that region be used. ,node and The distance between them is , For nodes The amount of construction work. Then:

[0173]

[0174] in, The smaller the value, the more likely it is to be a node. The more concentrated the construction tasks are, the closer the spatial distance is, and the more compact the node distribution is.

[0175] Then, the meta controller according to The value of the master node Choice:

[0176]

[0177] Next, a neighborhood node priority strategy is adopted, prioritizing secondary nodes that are close to the main node and have urgent deadlines to reduce navigation costs. Let... For nodes The urgency of the construction period The distance between the primary node and the secondary node determines the priority of the secondary node. for:

[0178]

[0179] The meta controller determines the priority weight of each secondary node. Sort the nodes and select the set of secondary nodes with higher priority weights. :

[0180]

[0181] Following the principle of unified construction procedures, priority is given to processing nodes with the same materials and consistent task types to reduce the need for switching robot working modes. Define task types. and wall material Represents a node Based on the construction information, construct a process similarity function between nodes. :

[0182]

[0183] in, For indicator functions, The larger the value, the more consistent the node's material and task type are with the main node. Adjust the selection of secondary nodes based on the results. :

[0184]

[0185] Finally, the total travel path is evaluated, assessing the travel paths from the current node to the primary and secondary nodes, selecting the shortest path to reduce energy consumption, and calculating the total travel path. :

[0186]

[0187] in This represents the distance between the two nodes. Optimize the combination of secondary nodes based on the results. :

[0188]

[0189] Ultimately, the meta-controller selects the optimal master node. and secondary node set This is to optimize the construction path and improve efficiency.

[0190] The sub-controller is responsible for generating collision-free trajectories and controlling the robot to safely reach the sub-target. First, based on the sub-target selected by the meta-controller. Position coordinates Convert it to the target position in the local obstacle avoidance coordinate system. The specific conversion formula is as follows:

[0191]

[0192] in, This represents the robot's current orientation angle.

[0193] Next, based on the target position in the local obstacle avoidance coordinate system The target direction that the computational robot should face. :

[0194]

[0195] This direction is the direction the robot should move in when there are no obstacles.

[0196] During obstacle avoidance, the robot needs to dynamically adjust its angular velocity based on the distribution of obstacles. Assuming the robot uses differential drive, its linear velocity... Angular velocity is a constant. The calculation process is as follows:

[0197] Select robot orientation In the direction of the obstacle avoidance coordinate system, a forward-looking point is selected, which is 1 unit away from the origin. Then, we can obtain:

[0198]

[0199] Where R is the radius of the arc formed by the current coordinates and the look-ahead point, which is the radius of the robot's turn. Therefore:

[0200]

[0201] Based on the turning radius of the differential drive robot and linear velocity Relationship Calculate angular velocity :

[0202]

[0203] in, The target angle that the robot should turn to.

[0204] To prevent excessive fluctuations in the robot's angular velocity during obstacle avoidance, a first-order IIR filter is used to control the angular velocity. After smoothing, the difference equation of the filter is:

[0205]

[0206] in, The angular velocity at the previous moment. The new angular velocity calculated at the current moment. The filtered angular velocity. This is the weighting constant.

[0207] like Figure 3 As shown, the overall framework of the method disclosed in this invention can be divided into several functional modules. These modules work together to generate and optimize the final execution path from raw BIM data to robot task allocation.

[0208] This method specifically includes the following modules:

[0209] The BIM data processing module serves as the initial input and preprocessing stage of the system, and includes the following two sub-modules that execute in parallel:

[0210] The IFC file export and parsing section exports building model data conforming to the Industry Foundation Classes (IFC) standard format from the original BIM model and performs semantic parsing to extract the geometric shape, spatial topology, and semantic attribute information of building components (such as walls, doors, passages, etc.).

[0211] The rendering, image generation, and node detection section renders the BIM model to generate a two-dimensional structural image. Then, it processes the image using an image recognition algorithm to detect key visual nodes in the building space (such as corners, doorways, and passageway centers) and extract their corresponding geometric and texture features.

[0212] The node semantic integration module, as the core module of information fusion, integrates the IFC parsing results and image detection results. By establishing a mapping relationship between visual nodes and BIM semantic nodes, the system assigns clear semantic labels to each node, such as "entrance / exit," "corner," and "connection area," forming a semantically rich integrated node set to provide input for subsequent topology construction.

[0213] The hierarchical topology graph construction module is the core processing part of the system, used to construct a navigation topology graph with a hierarchical structure, and includes the following sub-modules:

[0214] The multi-dimensional node layering module performs preliminary layering of semantic nodes based on attributes such as node floor affiliation and functional area division.

[0215] The node clustering and center node calculation module performs cluster analysis on the nodes in each layer and determines the center node of each cluster as a candidate main path node.

[0216] The dynamic weight allocation and master node selection module dynamically assigns weights to nodes based on parameters such as node importance, connectivity, and spatial location, and selects the master node as the global navigation skeleton.

[0217] The secondary node identification and connection module determines whether the remaining nodes can serve as secondary auxiliary nodes and establishes their connectivity with the primary node. The output of this module can be fed back to the primary node selection module, forming a closed-loop optimization mechanism that continuously adjusts weights and connections to ensure the stability and rationality of the topology.

[0218] The reachability verification and topology generation module is responsible for verifying the reachability of connections between all primary and secondary nodes after the initial topology structure is established. It uses a path search algorithm to detect the traversability of each path segment in the graph, eliminating impassable or unreasonable connections, and ultimately generating a navigation topology graph that is executable in the physical world.

[0219] The hierarchical control and path planning module is the application-oriented output phase of the system, responsible for specific path planning and execution control. It includes the following modules:

[0220] The meta controller module, as a high-level controller, performs preliminary path planning based on the global topology graph, generating the optimal path sequence from the starting point to the target point.

[0221] During path execution, the path optimization and update module dynamically optimizes or replans the path based on real-time environmental data or task constraints to improve path efficiency or save energy.

[0222] The sub-controller and obstacle avoidance module serve as low-level controllers. This module refines the planned path into specific motion commands for the robot and combines sensor inputs to perform dynamic obstacle avoidance and safety control.

[0223] The robot execution and feedback module is responsible for sending control commands to the robot platform to achieve path tracking and action execution, and collecting execution status in real time (such as current position, whether it has deviated, whether it has encountered obstacles, etc.), and feeding it back to the upper controller for path updates.

[0224] The embodiments of this invention demonstrate the working principle of the core technology of this patent, the "layered controller," clearly distinguishing between the two levels of the primary controller and the sub-controller. Figure 4 The meta-controller is responsible for strategic planning. Through algorithms such as "regional compactness analysis," it selected three areas—"masterbedroom," "bedroom2," and "kitchen"—as the work targets. The path it planned (red arrow) is a macro-level, conceptual sequence that determines which room to visit first and how to move within the room, aiming to optimize the overall construction sequence and efficiency while reducing energy consumption. Figure 5 The sub-controller dynamically avoids obstacles based on the robot's kinematic model and real-time sensor information, generating a collision-free trajectory. The orange "robot trajectory" line in the figure records in detail how the robot starts from the starting point (green dot), smoothly turns, passes through narrow doorways, moves within the designated work area, and finally reaches the ending point (red dot).

[0225] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A BIM-based intelligent navigation and task allocation method for construction sites, characterized in that, The specific steps are as follows: Step 1: Process the BIM model to obtain room semantic information and key construction node information, and combine the nodes with semantic information to extract indoor building information; The process begins by exporting IFC information from the BIM model, reading the IFC file using an IFC library, loading the model data, and extracting attributes from relevant entities via scripts. Extracted information includes wall material, construction period, and semantic information about rooms. Next, the BIM model is converted into images using an automated script combined with a 3D rendering engine to transform the model into overall building and room images. These images are then processed through camera view settings and batch rendering. Subsequently, ResNet-50 and CBAM attention mechanisms are used to detect key nodes corresponding to door and window locations, room corners, and wall positions from the images, extracting feature vectors. An attention mechanism is then used to emphasize important parts of the input features through a weight matrix. Finally, the identified nodes are combined with their location and semantic information. Step 2: Construct a hierarchical topology map by multi-dimensional layering of nodes, clustering and dynamic weight allocation, combined with construction information; In the process of constructing the layered topology map, the node set was extracted from the BIM model. Considering that walls of different materials need to be constructed separately, and that it is more convenient to construct nodes in the same room, as well as the principle of tight schedule, the construction sequence was arranged. Then, the nodes of each group were agglomerated into dense clusters through clustering algorithm, and the central node of each group was calculated. Then, using dynamic weight information, these central nodes were merged into the main nodes of the topology map to ensure its advantages in terms of materials, rooms and schedule. By verifying the reachability relationship, other nodes were connected to the main node as secondary nodes, while eliminating connections that cross obstacles. Finally, the accurate construction of the layered topology map was completed. Step 3: Establish the robot's kinematic model, design a hierarchical controller, the meta-controller selects the main node and secondary node in the hierarchical topology map, and the sub-controller performs obstacle avoidance to ensure collision-free arrival at the sub-target and complete the navigation objective; During the process of establishing the robot's kinematic model, the meta-controller prioritizes nearby construction nodes through regional compactness analysis and selects secondary nodes that are close to the main node and have tight deadlines by adopting a neighborhood node priority strategy. Following the principle of unified construction procedures, nodes with consistent materials and task types are given priority. Finally, the shortest path is evaluated and selected. The sub-controller is responsible for obstacle avoidance to achieve collision-free arrival at the selected target node. Based on the current state and the selected node, the sub-controller perceives obstacles in the environment in real time, generates a collision-free trajectory, and controls the robot's movement based on the robot's kinematic model to safely navigate to the target node.

2. The BIM-based intelligent navigation and task allocation method for construction sites according to claim 1, characterized in that: The specific process of step one is as follows: Given an input image Feature maps are extracted using ResNet-50. , denoted as: ； in This represents a ResNet-50 convolutional neural network used for image feature extraction. It contains 50 layers and uses residual blocks to address the vanishing gradient problem in deep networks. One residual block contains three convolutional layers. Assume the input is... The output of the residual block for: ； in It is a residual function, and its specific structure is as follows: ； Here These are the weights of the three convolutional layers. It is the corresponding bias term; Feature enhancement is performed using CBAM to obtain the enhanced feature map. : ； in The attention module represents a system that comprises two modules: a channel attention module and a spatial attention module. for: ； in It is the Sigmoid function. It is a multilayer perceptron. and These are global average pooling and global max pooling, respectively. Spatial Attention Module F s calculate: ； in, It's a convolution operation. It's a splicing operation. and These are average pooling and max pooling, respectively. The final weighted feature map F' is: ； in This represents element-wise multiplication; Based on the obtained feature vector of the target, the target is identified and its two-dimensional position coordinates in the image are determined. Assuming the detected node is... Its position coordinates are By fusing location information with semantic information to create a comprehensive node description, this process can be represented as follows: For each node... Its comprehensive description Represented as a tuple containing the position coordinates and related semantic attributes: 1; ； in, It is a node Position coordinates in the image, It is a collection containing semantic information; ； Combine the comprehensive descriptions of all nodes to form a complete model description. : ； in This will increase the total number of detected nodes.

3. The BIM-based intelligent navigation and task allocation method for construction sites according to claim 1, characterized in that: Step two is described in detail below: First, all nodes are divided into multiple levels based on their different attributes. These attributes include the wall material, the room the node belongs to, and the urgency of the construction period. Each subset contains nodes with the same attributes. The set of all nodes. Represented as a tuple, For having the same semantics If the set of nodes is a subset, then: ； in, These are the room type, wall material, and construction period attributes; A subset of nodes for each semantic attribute Within, clustering algorithms are applied to divide the nodes into several clusters. Let... Grouping The number of clusters, For the first Each cluster: ； in, It is a node The index of the assigned cluster; For each cluster Calculate its central node The central node is the average of the spatial location attribute values ​​of all nodes within the cluster, representing the characteristics of the cluster. Let... For the first The central node of each cluster, then: ； in, Cluster The number of nodes in Represents a node The coordinates; To calculate the master node of the topology map based on the central node of each cluster, a dynamic weight needs to be assigned to each central node. This weight is based on multiple factors, including the urgency of the construction period, room type, and wall material. The total number of weighting factors. For the first The weight vector of each weight factor. As the central node In the The score on each weighted factor, then the center node Overall weight for: ； An optimization method is used to calculate the master nodes of the topology map based on the hierarchical center nodes. The goal is to find the locations that minimize the weighted distances between nodes in the entire topology map. and They are the central nodes. and The weight, As the central node and The distance between them is then used to optimize the problem as follows: ； The reachability between primary nodes is verified using a path search algorithm from graph theory to ensure that all primary nodes are interconnected. Then, primary nodes are connected to secondary nodes to form a topology map. Represents a topological map. Represents the set of master nodes. Let the set of edges be: ； Ultimately, a hierarchical topology is constructed that takes into account multiple attributes, ensuring that the task faces advantages under constraints in terms of materials, rooms, and schedules.

4. The BIM-based intelligent navigation and task allocation method for construction sites according to claim 3, characterized in that: In step two, during the construction of the layered topology map, starting from the node set extracted from the BIM model, and considering that walls of different materials need to be constructed separately, that it is more convenient to construct nodes in the same room, and the principle of urgency, the construction sequence is arranged as follows: The nodes are divided into several layers: spatial layering, grouping nodes according to room information to facilitate efficient construction of nodes in the same room; material layering, dividing nodes into several groups according to different wall materials to ensure consistency in treatment methods and materials used during construction; and schedule layering, sorting and grouping nodes according to the urgency of the schedule to ensure reasonable allocation of construction progress.

5. The BIM-based intelligent navigation and task allocation method for construction sites according to claim 3, characterized in that: The step of the ternary controller selecting the primary and secondary nodes in the hierarchical topology map is as follows: First, a regional compactness analysis is performed, prioritizing nodes with concentrated construction tasks and close spatial distances, and a compactness function is defined. To evaluate the compactness of nodes within a certain region, let the set of principal nodes within that region be used. ,node and The distance between them is , For nodes The amount of construction work. Then: ； in, The smaller the value, the more likely it is to be a node. The more concentrated the construction tasks, the closer the spatial distance, and the more compact the node distribution; Then, the meta controller according to The value of the master node Choice: ； in Refers to a node in set N; Next, a neighborhood node priority strategy is adopted, prioritizing secondary nodes that are close to the primary node and have tight deadlines to reduce navigation costs. For nodes The urgency of the construction period The distance between the primary node and the secondary node determines the priority of the secondary node. for: ； The meta controller determines the priority weight of each secondary node. Sort the nodes and select the set of secondary nodes with higher priority weights. : ； Following the principle of unified construction procedures, priority is given to processing nodes with the same materials and consistent task types to reduce the need for switching robot working modes. Task types are defined accordingly. and wall material Representing nodes respectively Based on the construction information, construct a process similarity function between nodes. : ； in, For indicator functions, The larger the value, the more consistent the node's material and task type are with the main node. Adjust the selection of secondary nodes based on the results. ,in Represents a node If the processes are similar, then: ； Finally, the total travel path is evaluated, assessing the travel paths from the current node to the primary and secondary nodes, selecting the shortest path to reduce energy consumption, and calculating the total travel path. : ； in Given the distance between two nodes, optimize the combination of secondary nodes based on the result. : ； Ultimately, the meta-controller selects the optimal master node. and secondary node set .

6. The BIM-based intelligent navigation and task allocation method for construction sites according to claim 3, characterized in that: In step three, the sub-controller performs obstacle avoidance to ensure collision-free arrival at the sub-target. The specific process is as follows: The sub-controller is responsible for generating collision-free trajectories and controlling the robot to safely reach the sub-target. First, based on the sub-target selected by the meta controller. Position coordinates Convert it to the target position in the local obstacle avoidance coordinate system. The specific conversion formula is as follows: ； in, This is the robot's current orientation angle; Next, based on the target position in the local obstacle avoidance coordinate system The target direction that the computational robot should face. : ； Target direction This refers to the direction the robot should move in unobstructed conditions. During obstacle avoidance, the robot needs to dynamically adjust its angular velocity based on the distribution of obstacles. Assuming the robot uses differential drive, its linear velocity... Angular velocity is a constant. The calculation process is as follows: Select the robot's current orientation angle In the direction of the obstacle avoidance coordinate system, a forward-looking point is selected, which is 1 unit away from the origin. At this point, we obtain: ； Where R is the radius of the arc formed by the current coordinates and the look-ahead point, which is the turning radius of the differential drive robot. Therefore: ； Based on the turning radius of the differential drive robot and linear velocity Relationship Calculate angular velocity : ； in, The target angle that the robot should turn to; To prevent excessive fluctuations in the robot's angular velocity during obstacle avoidance, a first-order IIR filter is used to control the angular velocity. After smoothing, the difference equation of the filter is: ； in, The angular velocity at the previous moment. The new angular velocity calculated at the current moment. The filtered angular velocity. This is the weighting constant.

Images