A construction robot cooperative work system

Abstract

The application relates to the technical field of building engineering automation, and discloses a building robot cooperative operation system, a BIM modeling module is used for establishing a three-dimensional model of building engineering and dividing the three-dimensional model into multiple operation areas; an information collection module is used for collecting state information and operation environment information of the robot; a central control module is used for task distribution and operation planning by using an ant colony algorithm; an information sharing module is used for real-time communication between the robots through wireless communication, sharing of operation progress, state information and environment information during the operation process; and an operation evaluation module is used for real-time monitoring of operation progress and state of each robot, evaluation of operation effect of the robot, timely adjustment of task distribution and planning strategy according to the evaluation result, and optimization of the cooperative operation process of the robot. The application solves the problems of unreasonable task distribution, low communication efficiency and difficult operation coordination in the existing building robot cooperative operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building engineering automation, and particularly relates to a building robot cooperative working system. BACKGROUND

[0002] In the field of building engineering, with the development of automation technology, the application of building robots is becoming more and more widespread. However, at present, when building robots work cooperatively, there are problems such as unreasonable task allocation, low communication efficiency, and difficult work coordination, which lead to low work efficiency and cannot fully exert the advantages of building robots. For example, when multiple robots cooperatively complete a building task, some robots may have too heavy tasks, while some robots are idle. At the same time, communication between robots may be delayed or incorrect, affecting the smooth progress of work. SUMMARY

[0003] The present application aims to solve the above problems and designs a building robot cooperative working system.

[0004] The present application provides a building robot cooperative working system, which comprises:

[0005] A BIM modeling module is configured to establish a three-dimensional model of a building engineering by using BIM technology, and divide the BIM model into multiple working areas;

[0006] An information acquisition module is configured to acquire real-time state information and working environment information of each robot through sensors equipped on each robot;

[0007] A central control module is configured to receive state information and working environment information sent by each robot, combine working area division and working requirements in the BIM model, and use an ant colony algorithm to allocate tasks and plan work;

[0008] An information sharing module is configured to enable each robot to move to a corresponding working area for work according to the allocated tasks and work plan. During the work process, the robots communicate in real time through wireless communication and share work progress, state information, and environment information;

[0009] A work evaluation module is configured to monitor the work progress and state of each robot in real time, evaluate the work effect of the robots, and timely adjust the task allocation and planning strategy according to the evaluation results to optimize the cooperative working process of the robots.

[0010] Optionally, in the first implementation manner of the present application, the BIM modeling module comprises:

[0011] The semantic segmentation submodule is configured to acquire three-dimensional point cloud data of the BIM model, perform semantic segmentation on the three-dimensional point cloud data, extract geometric features and semantic features, and construct a multi-dimensional feature vector containing spatial positions and structural semantics.

[0012] The learning submodule is configured to abstract the three-dimensional point cloud data into nodes in a graph structure based on the multi-dimensional feature vector output by the semantic segmentation submodule, learn spatial dependency relationships between the nodes through a graph neural network, and generate a graph representation containing structural logic.

[0013] The division submodule is configured to perform multi-scale feature extraction on the BIM model based on the graph representation output by the learning submodule, introduce a hierarchical clustering algorithm, and perform multi-scale spatial region hierarchical division to divide the BIM model into a plurality of work regions.

[0014] Optionally, in the second implementation manner of the present application, the division submodule comprises:

[0015] The three-dimensional convolutional neural network comprises a plurality of convolutional layers and pooling layers, local detail features are extracted through shallow convolutional layers, global structural features are extracted through deep convolutional layers, and multi-scale feature representations are generated through a pyramid pooling module to integrate feature maps of different scales.

[0016] Based on the extracted multi-scale features, each small region is taken as an initial clustering unit, the feature similarity between each small region is calculated by using cosine similarity, the clustering units with the highest similarity are merged in a bottom-up manner to form larger regions, and a region division result is obtained.

[0017] Optionally, in the third implementation manner of the present application, the information acquisition module comprises:

[0018] The establishment submodule is configured to perform timestamp alignment on the original data of the sensors, establish a conversion relationship between the coordinate systems of the sensors and the coordinate system of the robot body, and generate a spatio-temporal alignment dataset.

[0019] The processing submodule is configured to construct a spatio-temporal attention network, process the multi-modal data in the spatio-temporal alignment dataset, extract spatial features of visual images through a convolutional neural network, extract three-dimensional spatial features of lidar data through a point cloud processing algorithm, capture dependency relationships between data of different sensors at different times by using a self-attention mechanism of a Transformer encoder, and output a fusion feature vector containing spatio-temporal context information.

[0020] The analysis submodule is configured to analyze the fusion feature vector to obtain state information of the robot and work environment information.

[0021] Optionally, in the fourth implementation manner of the present application, the analysis submodule comprises:

[0022] The feature distribution model in the normal working state is constructed by using a self-encoder, the fusion feature vector is input into the fusion feature vector, and an abnormal event is detected by calculating a reconstruction error;

[0023] The fusion feature vector is subjected to semantic segmentation by using a ViT-BERT model, the category and position of an object are identified, and an environment map with semantic labels is generated;

[0024] The outputs of the feature distribution model and the ViT-BERT model are integrated to obtain the state information of the robot and the working environment information.

[0025] Optionally, in the fifth implementation manner of the present application, the central control module comprises:

[0026] An initialization submodule is configured to decompose the construction engineering task into a plurality of subtasks, each subtask corresponding to a working area, and initialize ant colony parameters;

[0027] A selection submodule is configured to select a next subtask to be executed according to the pheromone concentration and heuristic information of the current subtasks, each ant representing a robot, and the ants follow a roulette selection strategy when selecting a task;

[0028] A distribution submodule is configured to update the pheromone concentration between the subtasks according to the results of the task distribution after all ants complete a task distribution;

[0029] An iteration submodule is configured to obtain an optimal task distribution scheme after a plurality of iterations, distribute the subtasks and working areas to each robot, and plan the working order and path of the robot.

[0030] Optionally, in the sixth implementation manner of the present application, the ant colony parameters at least include the number of ants, the initial concentration of pheromones, a heuristic factor, and a pheromone evaporation factor.

[0031] Optionally, in the seventh implementation manner of the present application, the information sharing module comprises:

[0032] An updating submodule is configured to update the neighbor node list in real time by signal strength detection when the robot moves, dynamically adjust the communication link by using a Q-learning algorithm, and form an adaptive Mesh network topology;

[0033] A calculation submodule is configured to calculate a conflict resolution weight according to the task priority, the current load of the robot, and the resource occupation duration when two robots request the same resource, and put the robot with a low priority into a waiting queue;

[0034] The implementation submodule is configured to obtain an optimal yielding strategy through a Nash equilibrium algorithm if a path planning conflict occurs, so as to realize real-time communication and cooperative control of the multiple robots.

[0035] Optionally, in an eighth implementation manner of the present application, the implementation submodule comprises:

[0036] Each robot predicts a trajectory point sequence in a future time window based on a current position, a speed and a target path, and generates a space-time occupancy grid map using a Kalman filter;

[0037] The space-time occupancy grid maps of all the robots are spatially superimposed to identify an area with space-time overlap, and a conflict state vector is constructed;

[0038] Based on the conflict state vector, a set of optional strategies is defined for each robot, and for each strategy combination, a benefit of the robot is calculated to construct a multi-dimensional benefit matrix;

[0039] A Nash equilibrium point is solved by traversing the multi-dimensional benefit matrix to determine a final executed Nash equilibrium strategy combination, so as to obtain an optimal yielding strategy.

[0040] Optionally, in a ninth implementation manner of the present application, a method for implementing a cooperative working system of a building robot is provided, and the method comprises the following steps:

[0041] A three-dimensional model of a building project is established using BIM technology, and the BIM model is divided into multiple working areas;

[0042] State information and working environment information of each robot are collected in real time through sensors equipped on the robot;

[0043] The state information and working environment information sent by each robot are received, and task allocation and working planning are performed using an ant colony algorithm in combination with working area division and working requirements in the BIM model;

[0044] Each robot moves to a corresponding working area to perform work according to the allocated task and working planning, and the robots communicate in real time through wireless communication to share working progress, state information and environment information during the working process;

[0045] The working progress and state of each robot are monitored in real time, the working effect of the robot is evaluated, and the task allocation and planning strategy are adjusted in a timely manner according to the evaluation result to optimize the cooperative working process of the robot.

[0046] The technical scheme provided by the application comprises the following steps: a three-dimensional model of a building project is established by using a BIM technology, and the BIM model is divided into a plurality of work areas; state information and work environment information of each robot are collected in real time by sensors equipped on the robot; the state information and the work environment information sent by each robot are received, and the BIM model is combined with work area division and work requirements to perform task allocation and work planning by using an ant colony algorithm; each robot moves to a corresponding work area to perform work according to the allocated task and work planning, and the robots communicate in real time by wireless communication to share work progress, state information and environment information during the work; work progress and state of each robot are monitored in real time, work effect of the robot is evaluated, task allocation and planning strategies are adjusted in time according to the evaluation result, and the collaborative work process of the robot is optimized; the application provides accurate basic data for task allocation by establishing the BIM model, improves rationality and efficiency of task allocation by using the ant colony algorithm for task allocation and planning, realizes efficient collaborative work of the robot by real-time communication and collaborative control algorithm between the robots, meanwhile, work monitoring and evaluation can adjust work strategies in time to ensure work quality and efficiency, and the application solves problems of unreasonable task allocation, low communication efficiency and difficult work coordination in the existing building robot collaborative work, and has a wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0047] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not considered a limitation of the application.

[0048] Figure 1 A structural schematic diagram of a building robot collaborative work system provided by an embodiment of the application is shown in FIG. 1.

[0049] Figure 2 A structural schematic diagram of a BIM modeling module provided by an embodiment of the application is shown in FIG. 2.

[0050] Figure 3 A structural schematic diagram of an information collection module provided by an embodiment of the application is shown in FIG. 3. DETAILED DESCRIPTION

[0051] The terms "first", "second", "third", "fourth" and the like in the description and claims of the present application and above-mentioned drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of the terms so construed herein is merely for convenience and only to aid in understanding the application and is in no way a limitation on its broader scope. Furthermore, the terms "comprise", "comprising", "include", "including", and the like are to be construed open-ended, i.e., to mean including but not limited to, to indicate the presence of what follows, inclusive or non-exclusive.

[0052] For the convenience of understanding, the specific flow of the embodiment of the present application is described below. Please refer to Figure 1 The building robot cooperative working system provided by the embodiment of the present application has the structure diagram as shown in the figure, and the system comprises:

[0053] A BIM modeling module is configured to establish a three-dimensional model of a building project by using a BIM technology, and divide the BIM model into a plurality of working areas.

[0054] An information acquisition module is configured to acquire state information and working environment information of the robots in real time through sensors equipped on each robot.

[0055] A central control module is configured to receive the state information and working environment information sent by the robots, combine working area division and working requirements in the BIM model, and perform task allocation and working planning by using an ant colony algorithm.

[0056] An information sharing module is configured to move the robots to corresponding working areas for working according to the allocated tasks and working planning, and the robots share working progress, state information and environment information in real time through wireless communication between the robots during the working process.

[0057] A working evaluation module is configured to monitor working progress and state of each robot in real time, evaluate working effect of the robots, and timely adjust task allocation and planning strategy according to the evaluation result, so as to optimize the cooperative working process of the robots.

[0058] In the embodiment, please refer to Figure 2 The BIM modeling module comprises:

[0059] A semantic segmentation sub-module is configured to acquire three-dimensional point cloud data of the BIM model, perform semantic segmentation on the three-dimensional point cloud data, extract geometric features and semantic features, and construct a multi-dimensional feature vector containing spatial position and structural semantics.

[0060] The learning submodule is configured to abstract the three-dimensional point cloud data into nodes in a graph structure based on the multi-dimensional feature vector output by the semantic segmentation submodule, learn spatial dependency between the nodes by using a graph neural network, and generate a graph representation containing structural logic;

[0061] The division submodule is configured to perform multi-scale feature extraction on the BIM model by using a three-dimensional convolutional neural network based on the graph representation output by the learning submodule, and perform multi-scale spatial region hierarchical division by using a hierarchical clustering algorithm to divide the BIM model into a plurality of work regions.

[0062] In this embodiment, the division submodule includes:

[0063] The three-dimensional convolutional neural network includes a plurality of convolutional layers and pooling layers, and is configured to extract local detail features by using the shallow convolutional layers, extract global structural features by using the deep convolutional layers, integrate feature maps of different scales by using a pyramid pooling module, and generate a multi-scale feature representation.

[0064] Based on the extracted multi-scale features, each small region is taken as an initial clustering unit, a cosine similarity is calculated between each small region, and the clustering units with the highest similarity are merged in a bottom-up manner to form larger regions, and a region division result is obtained.

[0065] In this embodiment, refer to Figure 3 The information acquisition module includes:

[0066] The establishment submodule is configured to perform timestamp alignment on the raw data of the sensors, establish a conversion relationship between the coordinate systems of the sensors and the coordinate system of the robot body, and generate a spatio-temporal alignment dataset.

[0067] The processing submodule is configured to construct a spatio-temporal attention network, process the multi-modal data in the spatio-temporal alignment dataset, extract spatial features of the visual images by using a convolutional neural network, extract three-dimensional spatial features of the laser radar data by using a point cloud processing algorithm, capture dependency relationships between data of different sensors at different times by using a self-attention mechanism of a Transformer encoder, and output a fusion feature vector containing spatio-temporal context information.

[0068] The analysis submodule is configured to analyze the fusion feature vector to obtain state information of the robot and work environment information.

[0069] In this embodiment, the analysis submodule includes:

[0070] The autoencoder is configured to construct a feature distribution model in a normal work state, input the fusion feature vector into the fusion feature vector, and detect abnormal events by calculating a reconstruction error.

[0071] The ViT-BERT model is used for semantic segmentation of the fusion feature vector to identify the category and position of the object and generate an environment map with semantic labels.

[0072] The output of the feature distribution model and the ViT-BERT model is integrated to obtain the state information of the robot and the working environment information.

[0073] In this embodiment, the central control module comprises:

[0074] The initialization submodule is configured to decompose the construction engineering task into a plurality of subtasks, each subtask corresponding to a work area, and initialize ant colony parameters.

[0075] The selection submodule is configured to select the next subtask to be executed according to the pheromone concentration and heuristic information of each subtask, and each ant follows a roulette selection strategy when selecting a task.

[0076] The assignment submodule is configured to update the pheromone concentration between each subtask according to the results of task assignment after all ants complete a task assignment.

[0077] The iteration submodule is configured to obtain an optimal task assignment scheme after multiple iterations, assign each robot with a subtask and a work area, and plan the work order and path of the robot.

[0078] In this embodiment, the ant colony parameters at least include the number of ants, the initial concentration of pheromone, the heuristic factor, and the pheromone evaporation factor.

[0079] In this embodiment, the information sharing module comprises:

[0080] The update submodule is configured to update the neighbor node list in real time through signal strength detection when the robot moves, dynamically adjust the communication link using the Q-learning algorithm, and form an adaptive Mesh network topology.

[0081] The calculation submodule is configured to calculate the conflict resolution weight according to the task priority, the current load of the robot, and the resource occupation time when two robots request the same resource, and put the robot with low priority into a waiting queue.

[0082] The implementation submodule is configured to obtain an optimal compromise strategy through the Nash equilibrium algorithm if a path planning conflict occurs, and realize real-time communication and collaborative control of multiple robots.

[0083] In this embodiment, the implementation submodule comprises:

[0084] Each robot uses a Kalman filter to predict a sequence of trajectory points within a future time window based on the current position, speed, and target path, and generates a space-time occupancy grid.

[0085] Superimpose the space-time occupation grid maps of all robots, identify the areas with space-time overlap, and construct a conflict state vector;

[0086] Based on the conflict state vector, define a set of optional strategies for each robot, and for each strategy combination, calculate the robot's revenue to construct a multi-dimensional revenue matrix;

[0087] Iterate through the multi-dimensional revenue matrix to solve the Nash equilibrium point, determine the final executed Nash equilibrium strategy combination, and obtain the optimal concession strategy.

[0088] In this embodiment, the specific extension of the BIM modeling module includes:

[0089] Downsample the three-dimensional point cloud data exported from the BIM model, filter the data redundancy through voxel grid, and at the same time retain the key geometric features; remove outliers from the downsampled point cloud, use statistical outlier filtering algorithm to calculate the average distance of each point to its neighborhood points, and regard the points deviating from the statistical distribution as noise points and remove them; finally, convert the processed point cloud data into a regular voxel grid representation, providing a unified input format for subsequent convolution operations;

[0090] Use the PointNet++ network architecture to perform semantic segmentation on the voxelized point cloud. This network captures local geometric features at different scales through multiple layers of feature extractors. First, use the Set Abstraction module to downsample the input point cloud and extract features of local regions. Then, use the Feature Propagation module to upsample the low-resolution features to the original point cloud density, achieving point-by-point classification. During the training process, use the cross-entropy loss function combined with data augmentation techniques (random rotation, scaling) to improve the model's generalization ability, and finally assign semantic labels (walls, floors, beams, columns, etc.) to each point.

[0091] Extract the geometric features of the point cloud, including the normal vector, curvature, local surface roughness, etc., by calculating the covariance matrix of each point neighborhood and performing feature decomposition; at the same time, extract semantic features such as class probability distribution and semantic boundary information from the semantic segmentation results; concatenate the geometric features and semantic features to construct a high-dimensional feature vector; to reduce feature redundancy, use principal component analysis to reduce the dimension of the feature vector, retain the principal components with the largest variance contribution, and generate a multi-dimensional feature vector containing spatial position and structural semantics;

[0092] The point cloud data after semantic segmentation and feature extraction is abstracted as nodes in a graph structure, each node representing a semantic region (a wall, a column); the node attributes include the geometric center coordinates, size parameters (length, width, height), semantic category labels, and statistical features (average normal vector, curvature distribution) extracted from the point cloud; at the same time, each node is assigned construction process constraint attributes, pouring order requirements, load-bearing dependency relationships, etc., which come from the design parameters of the BIM model and construction specifications;

[0093] Edges are constructed between nodes to represent spatial dependency relationships, and the weights of the edges reflect the correlation strength between nodes; first, an initial graph structure is constructed based on spatial proximity, and if the minimum distance between two nodes is less than a certain threshold, an edge is established between them; then, the weights of the edges are calculated, considering geometric correlation (coplanarity, adjacent area ratio), semantic correlation (supporting relationship between walls and beams and columns), and construction sequence constraints (the main structure construction must be completed after the foundation construction); for nodes with direct functional dependency (doors and windows and walls), higher edge weights are assigned;

[0094] The graph convolution network (GCN) is used to learn the spatial dependency relationships between nodes, and the node feature representation is updated through multi-layer graph convolution operations; in each convolution layer, the node updates its own features by aggregating the feature information of its neighbor nodes, and the aggregation function considers the edge weights to distinguish the importance of different neighbors; during training, self-supervised learning tasks (node classification, link prediction) are used to optimize network parameters, so that the model can capture the potential dependency relationships in the building structure; by predicting the semantic category of the node, the model is forced to learn the correlation between spatial layout and structural function; finally, the graph representation containing spatial dependency relationships is output, providing structural constraints for subsequent work area division;

[0095] A three-dimensional convolutional neural network is used to extract multi-scale features from the BIM model; the network contains multiple convolutional and pooling layers, which capture spatial features of different scales through convolution kernels of different sizes; shallow convolutional layers extract local detail features (component edges, holes), and deep convolutional layers extract global structural features (floor layout, functional partition); after each convolution layer, batch normalization and ReLU activation functions are applied to enhance model stability and non-linear representation ability; pyramid pooling modules are used to integrate feature maps of different scales to generate multi-scale feature representations;

[0096] Based on the extracted multi-scale features, a hierarchical clustering algorithm is used for spatial region division. First, each voxel or small region is taken as the initial clustering unit, and the feature similarity (cosine similarity) between them is calculated. Then, the clustering units with the highest similarity are merged in a bottom-up manner to form larger regions. During the merging process, the Ward method is used to minimize the within-class variance to ensure the consistency of features within the same region. At the same time, spatial continuity constraints are introduced to preferentially merge adjacent regions and avoid generating fragmented region division results. By setting different clustering stop conditions, multi-level region division results (floor, room, functional area, etc.) are obtained.

[0097] The initial division regions are optimized in combination with the constraints of building construction technology. First, according to the construction sequence requirements (main body first, decoration second), the region boundaries are adjusted to ensure that the tasks within the same region can be executed continuously in sequence. Then, considering the limitations of robot operation capability, the regions are decomposed or merged to ensure that the size and complexity of each region are within the range of robot operability. Large-area walls are divided into multiple sub-regions, with the height and width of each sub-region not exceeding the maximum operating radius of the robotic arm. Finally, based on the spatial dependency relationships learned by the graph neural network, it is checked whether the region division results meet the structural construction logic, and the regions that violate the dependency relationships are adjusted to ensure the construction feasibility of the divided work regions.

[0098] In this embodiment, the specific expansion of the information collection module includes:

[0099] The raw data of sensors such as cameras, lidar, IMU, etc. are time-stamped aligned, and the conversion relationship between the sensor coordinate systems and the robot body coordinate system is established through the hand-eye calibration algorithm. For asynchronously collected sensors, linear interpolation is used to synchronize the data in time, ensuring the consistency of environment perception data in the spatial coordinate system at the same time, generating a spatio-temporal aligned dataset containing position, attitude, environment point cloud, and visual image.

[0100] A spatio-temporal attention network is constructed. First, the spatial features (obstacle outline, material position) of the visual image are extracted through a convolutional neural network, and the three-dimensional spatial features of the lidar data are extracted through a point cloud processing algorithm. Then, multiple frames of features in the time sequence are input into the Transformer encoder, and the self-attention mechanism is used to capture the dependency between different time and different sensor data (dynamic obstacle motion trajectory), outputting a fusion feature vector containing spatio-temporal context information, which is used to describe the dynamic changes of the environment around the robot.

[0101] The self-encoder is used to construct a feature distribution model in a normal working state. Real-time fusion features are input into the model to detect abnormal events (device failure, personnel intrusion) by calculating the reconstruction error. Meanwhile, a pre-trained ViT-BERT model is used to perform semantic segmentation on the fused scene image to identify the class and location of key objects (scaffolding, steel bars), and generate an environment map with semantic labels to provide interactive semantic information for task planning.

[0102] The extended Kalman filter is used to fuse data from the robot's body sensors (encoders, force sensors) to estimate the robot's position, attitude, joint torque, and other state parameters in real time. Combined with historical failure data, a device reliability model based on a hidden Markov model is established to predict parameters such as battery endurance and mechanical arm wear, and output a comprehensive perception report containing the robot's health status and environmental risk level.

[0103] In this embodiment, the specific expansion of the central control module includes:

[0104] The improved DBSCAN algorithm is used to cluster all micro-task units, with the task area center coordinates as the spatial dimension and the construction time window as the time dimension. The spatio-temporal distance between tasks is calculated. The density peak clustering is introduced to optimize the initial clustering center, and high correlation task clusters (wall building and concrete pouring tasks on the same floor) are identified. The sub-tasks within each task cluster have spatial proximity and process dependency, serving as the basic unit for macro-level task allocation;

[0105] The building structure tree is used as the macro-level planning framework to construct a task-level dependency graph. When searching in the structure tree, the ant colony prefers to select task clusters on the critical path (main structure construction). The pheromone update rule combines construction progress delay risk (weather influence probability) and resource occupation cost (crane usage time). The ant colony iteratively generates a task cluster execution order that conforms to the building construction logic, ensuring that tasks across floors and functional areas are sorted according to the "first up, then down, first main structure, then decoration" principle;

[0106] For each sub-task within a task cluster, a robot-task matching matrix is constructed, with elements including the time cost of the robot completing the task (based on a historical data fitting operation efficiency model), and the ability matching degree (welding robot's professional adaptation to steel structure tasks). When searching at the micro-level, each ant represents a robot, and selects sub-tasks according to the roulette strategy. The selection probability combines pheromone concentration and heuristic function (task urgency / robot current load). Local search optimization is used to avoid single robot task overload and ensure load balancing in task allocation;

[0107] The robots are divided into sub-populations such as carrying, masonry, and spraying according to function types, and the ant colonies of different sub-populations realize cooperation by sharing "cross-task type cooperation pheromone". After the carrying robot completes material transportation, it releases high-concentration pheromone to the masonry robot sub-population to guide it to preferentially select the corresponding masonry task. When a robot failure or environmental mutation is detected, local re-planning is triggered: high-priority sub-tasks are selected from the tasks not completed by the failed robot, and the pheromone matrix is updated to reflect the latest task status through rapid reallocation by the elite ant colony.

[0108] In this embodiment, the specific expansion of the information sharing module includes:

[0109] Each robot locally maintains a knowledge graph containing task progress, device status, and environmental semantics, and periodically aggregates global knowledge through a federated learning framework. In the parameter aggregation stage, the local knowledge vector is encrypted using secure multi-party computation. The global knowledge graph feeds back to each robot, updating its local construction knowledge base (standard process parameters, safety specifications);

[0110] A multi-attribute decision-making model is established to handle task execution conflicts: when two robots request the same resource (material storage area), the conflict resolution weight is calculated based on task priority (urgency x duration weight), current robot load, and resource occupation time, and the robot with lower priority enters the waiting queue. If a path planning conflict occurs, the Nash equilibrium algorithm in game theory is used to solve the optimal compromise strategy, such as adjusting the travel speed or taking a detour, to minimize the collaboration cost (time delay + increased energy consumption);

[0111] Multi-agent reinforcement learning is used to train the collaboration strategy of the robot team. Each robot observes the local environment (its own position, the position of nearby robots, and obstacle distribution) and shares key state information through an attention mechanism. In the trajectory planning stage, an initial path is generated based on the artificial potential field method, and then optimized through the deep deterministic policy gradient (DDPG) algorithm. A collision prediction module (spatiotemporal occupancy grid method) is introduced to detect the intersection of motion trajectories in real time, dynamically adjust the speed and direction, and ensure collision-free collaboration of multiple robots in a shared work area.

[0112] In this embodiment, the Nash equilibrium point solution includes:

[0113] Pure strategy Nash equilibrium detection: all strategy combinations are traversed to check if there is a combination that makes all participants unable to obtain higher benefits by changing their strategies unilaterally;

[0114] Mixed strategy Nash equilibrium calculation: if there is no pure strategy equilibrium, the optimal probability distribution of each participant choosing different strategies is calculated through linear programming method;

[0115] Equilibrium point screening: If there are multiple Nash equilibrium points, select the Pareto optimal equilibrium point (i.e. no other equilibrium point can make all participants' benefits improve at the same time).

[0116] In this embodiment, the specific extension of the job evaluation module includes:

[0117] A digital twin model of the construction robot and the construction environment is established, real-time collection of robot position, joint angle, energy consumption data, and environmental parameters such as temperature and humidity, dust concentration of the work area is performed through an Internet of Things interface; a space mapping algorithm is used to convert physical coordinates into virtual coordinates of the digital twin, interpolation is performed on missing data through a state estimation model, ensuring that the state synchronization accuracy of the virtual space and the physical world reaches centimeter level, and multi-dimensional construction performance evaluation is performed:

[0118] Quality dimension: construction defects (wall perpendicularity deviation, welding missing) are identified through computer vision algorithms, and quality deviation rate is calculated in combination with design parameters of the BIM model;

[0119] Efficiency dimension: a construction process model is established based on queuing theory, material transportation bottlenecks (elevator use conflicts) and robot waiting time are analyzed, and efficiency bottleneck points on the critical path are identified;

[0120] Safety dimension: the collision probability of the robot with personnel and obstacles is predicted using a space-time collision detection algorithm, a safety risk assessment matrix is constructed in combination with historical accident data, and a real-time safety warning level is output;

[0121] A construction process causal diagram is constructed, the causal relationship between task allocation strategy (robot load balancing degree), environmental parameters and construction efficiency / quality is analyzed; when an abnormal performance indicator is detected, the counterfactual results of different intervention measures (reassigning tasks, adjusting robot motion parameters) are simulated through a causal inference algorithm, the optimization scheme that maximizes performance improvement is selected, and control instructions including parameter adjustment amplitude and execution time are generated.

[0122] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only preferred embodiments of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A construction robot cooperative work system characterized by comprising: The system comprises: a BIM modeling module for establishing a three-dimensional model of a construction project using BIM technology and dividing the BIM model into a plurality of work areas; an information collection module for collecting state information and work environment information of the robots in real time through sensors equipped on each robot; a central control module for receiving state information and work environment information sent by each robot, combining work area division and work requirements in the BIM model, and using an ant colony algorithm to allocate tasks and plan work; an information sharing module for each robot to move to the corresponding work area for work according to the allocated tasks and work plan, and for the robots to communicate in real time through wireless communication to share work progress, state information and environment information during the work; a work evaluation module for monitoring the work progress and state of each robot in real time, evaluating the work effect of the robots, and adjusting the task allocation and planning strategy in a timely manner to optimize the collaborative work process of the robots; The information collection module comprises: an establishment submodule for timestamp alignment of raw data of the sensors, establishment of a conversion relationship between each sensor coordinate system and the robot body coordinate system, and generation of a time-space aligned data set; a processing submodule for constructing a space-time attention network, processing multi-modal data in the time-space aligned data set, extracting spatial features of visual images through a convolutional neural network, extracting three-dimensional spatial features of laser radar data through a point cloud processing algorithm, capturing the dependency relationship between different time and different sensor data using the self-attention mechanism of the Transformer encoder, and outputting a fusion feature vector containing space-time context information; an analysis submodule for constructing a feature distribution model under normal work state using an autoencoder, inputting real-time fusion features into the model, detecting abnormal events by calculating reconstruction error, identifying the category and position of objects using a ViT-BERT model to generate an environment map with semantic labels, and integrating the outputs of the feature distribution model and the ViT-BERT model to obtain state information and work environment information of the robots.

2. The construction robot collaborative work system of claim 1, wherein, The BIM modeling module comprises: a semantic segmentation submodule for obtaining three-dimensional point cloud data of the BIM model, performing semantic segmentation on the three-dimensional point cloud data, extracting geometric features and semantic features, and constructing a multi-dimensional feature vector containing spatial position and structural semantics; a learning submodule for abstracting three-dimensional point cloud data as nodes in a graph structure based on the multi-dimensional feature vector output by the semantic segmentation submodule, learning the spatial dependency relationship between nodes through a graph neural network, and generating a graph representation containing structural logic; a division submodule for performing multi-scale feature extraction on the BIM model using a three-dimensional convolutional neural network based on the graph representation output by the learning submodule, and introducing a hierarchical clustering algorithm to perform multi-scale spatial region hierarchical division to divide into a plurality of work areas.

3. The construction robot cooperative work system according to claim 2, wherein The division submodule comprises: The three-dimensional convolutional neural network is used for multi-scale feature extraction of the BIM model, and the three-dimensional convolutional neural network comprises multiple convolutional layers and pooling layers, local detail features are extracted through the shallow convolutional layers, global structure features are extracted through the deep convolutional layers, and the features of different scales are integrated through a pyramid pooling module to generate multi-scale feature representations; Based on the extracted multi-scale features, each small region is taken as an initial clustering unit, the feature similarity between each small region is calculated by using a cosine similarity, the clustering units with the highest similarity are merged in a bottom-up manner to form larger regions, and a region division result is obtained.

4. The construction robot cooperative work system according to claim 1, wherein The central control module comprises: An initialization submodule is configured to decompose the construction engineering task into multiple subtasks, each subtask corresponding to a work area, and initialize ant colony parameters; A selection submodule is configured to select a next subtask to be executed according to the pheromone concentration and heuristic information of each subtask, and each ant follows a roulette wheel selection strategy when selecting a task; An assignment submodule is configured to update the pheromone concentration between each subtask according to the task assignment result after all ants complete a task assignment; An iteration submodule is configured to obtain an optimal task assignment scheme after multiple iterations, assign each robot with a subtask and a work area, and plan a work order and a path of the robot.

5. The construction robot cooperative work system according to claim 4, wherein The ant colony parameters at least include the number of ants, the initial concentration of pheromones, a heuristic factor and a pheromone evaporation factor.

6. The construction robot cooperative work system according to claim 1, wherein The information sharing module comprises: An update submodule is configured to update a neighbor node list in real time through signal strength detection when the robot moves, dynamically adjust a communication link by using a Q-learning algorithm, and form an adaptive Mesh network topology; A calculation submodule is configured to calculate a conflict resolution weight according to a task priority, a current load of the robot and a resource occupation duration when two robots request the same resource, and put a robot with a low priority into a waiting queue; An implementation submodule is configured to obtain an optimal yielding strategy by using a Nash equilibrium algorithm if a path planning conflict occurs, and realize real-time communication and cooperative control of multiple robots.

7. The construction robot cooperative work system according to claim 6, wherein The implementation submodule comprises: Each robot uses a Kalman filter to predict a trajectory point sequence in a future time window based on a current position, a speed and a target path, and generates a space-time occupation grid map; The space-time occupation grid maps of all robots are spatially superimposed to identify an area with space-time overlap, and a conflict state vector is constructed; Based on the conflict state vector, a set of selectable strategies is defined for each robot, the revenue of the robot is calculated for each strategy combination, and a multi-dimensional revenue matrix is constructed; The multi-dimensional revenue matrix is traversed to solve a Nash equilibrium point, a final executed Nash equilibrium strategy combination is determined, and an optimal yielding strategy is obtained.

8. A method of implementing a construction robot cooperative work system according to claim 1, characterized by, The method comprises the following steps: A three-dimensional model of the construction engineering is established by using a BIM technology, and the BIM model is divided into multiple work areas; State information and work environment information of the robot are collected in real time by a sensor equipped on each robot; Receive the state information and working environment information sent by each robot, combine the working area division and working requirements in the BIM model, and use the ant colony algorithm to perform task allocation and working planning; Each robot moves to the corresponding working area for work according to the allocated task and working planning, and the robots share the working progress, state information and environment information through real-time communication between the robots during the working process; Real-time monitor the working progress and state of each robot, evaluate the working effect of the robot, and timely adjust the task allocation and planning strategy according to the evaluation result to optimize the collaborative working process of the robot.

Images