Building logistics robot carrying path optimization method and system based on obstacle avoidance

Abstract

The application relates to the technical field of robot path planning, and discloses a building logistics robot carrying path optimization method and system based on obstacle avoidance, which comprises the following steps: acquiring an environment point cloud to construct a three-dimensional physical attribute grid map containing flatness and friction coefficient; extracting an unknown obstacle bounding box and performing cross-stage matching with a model database to grade a completed building in advance and extract construction process attributes; using an exponential decay formula to radiate outward to correct surrounding grid physical parameters and generate an environment degradation gradient; combining the current total mass and speed of the robot to calculate a maximum safe curvature limit value, cutting off branches exceeding the limit value in path searching to generate a local detour path; reporting the path to the cloud after the vehicle passes, and calculating a safe relaxation degree according to the wheelbase and total mass of the subsequent vehicle to determine whether to issue a detour template. The application solves the problems of obstacle avoidance deadlock caused by construction progress and robot side slip out of control, and reduces the total computing power consumption of multi-machine cooperation.

Description

Technical Field

[0001] This invention relates to the field of robot path planning technology, specifically to a method and system for optimizing the handling path of a construction logistics robot based on obstacle avoidance. Background Technology

[0002] With the intelligent development of the construction industry, logistics robots have been widely used in material handling at construction sites. The complex and ever-changing environment of construction sites places extremely high demands on the autonomous navigation and path planning of robots. Currently, most logistics robot path planning relies on pre-imported building information models for global navigation, combined with onboard sensors for local obstacle avoidance.

[0003] However, in actual construction, the on-site progress often deviates from the theoretically planned progress, with some building entities frequently being completed ahead of schedule. When the robot scans these entities, because they cannot be matched in the current stage's preset model, conventional algorithms often misclassify them as personnel or temporary debris, leading to strategies such as waiting in place or frequent replanning, resulting in obstacle avoidance deadlock. Simultaneously, construction work affects the physical state of the surrounding ground, causing a decrease in local road surface friction or a deterioration in smoothness. Existing path planning methods typically only perform simple geometric obstacle avoidance, failing to quantify the deteriorating impact on the surrounding surface physical properties in conjunction with construction processes. Without combining dynamic ground physical parameters with real-time vehicle loads, heavy-duty robots are highly susceptible to lateral slippage or overturning and loss of control when turning on contaminated surfaces using conventionally planned trajectories. Furthermore, multiple heterogeneous logistics robots often operate collaboratively on construction sites. When faced with the same complex and unknown obstacle area, the existing independent planning mode requires each robot to perform a complete pathfinding calculation. It lacks a mechanism for sharing safety experience among multiple robots and does not consider the dynamic differences in anti-rollover capabilities of vehicles with different masses and wheelbases, resulting in serious redundancy consumption of the overall computing power of the cluster. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for optimizing the handling path of a construction logistics robot based on obstacle avoidance. This solves the problems of obstacle avoidance deadlock caused by asynchronous model progress in the construction environment, loss of control due to heavy-load sideslip caused by failure to consider the deterioration of road surface physical properties, and high computing power consumption due to the lack of heterogeneous safe reuse mechanism for multi-machine independent pathfinding.

[0005] To achieve the above objectives, the first aspect of the present invention provides a method for optimizing the handling path of a construction logistics robot based on obstacle avoidance, comprising the following steps: The system acquires environmental point cloud data collected by the sensor group, discretizes the environmental point cloud data into a three-dimensional spatial grid in the spatial dimension, extracts point cloud features based on a preset material mapping matrix, calculates the basic flatness index and basic friction coefficient of the corresponding grid, and constructs a three-dimensional physical property grid map containing physical parameters by combining the spatial benchmark in the model database. During the movement of the logistics robot, unknown obstacles ahead of the path are identified and the geometric bounding boxes of the obstacle point cloud are extracted. The geometric bounding boxes of the obstacle point cloud are geometrically matched with the component data of the current construction stage in the model database. When the matching of the current stage fails, the component data of the adjacent construction cycle is requested from the model database for cross-stage rematch. If the matching is successful, the unknown obstacle is classified as a building completed ahead of schedule. For obstacles identified as the prematurely completed building, the construction process attributes of the prematurely completed building are extracted from the model database. Based on the extracted construction process attributes, the prematurely completed building boundary is used as the reference to radiate outward in the three-dimensional physical attribute grid map. The foundation flatness index and foundation friction coefficient of each grid within the influence radius are corrected using a distance-based exponential decay formula to generate a physical environment associated degradation gradient. The current total mass data of the logistics robot and the driving speed command issued by the system are read. Combined with the corrected friction coefficient updated in the target extended grid, the maximum safe curvature limit for the logistics robot to maintain lateral slippage is calculated. The maximum safe curvature limit is introduced as a condition parameter into the cost function of the heuristic path search algorithm. Search branches with curvature exceeding the maximum safe curvature limit are truncated to generate local detour paths. After the logistics robot completes its journey along the local detour path, it reports the coordinate data of the local detour path and its own vehicle physical parameters to the cloud server. The cloud server receives the data and generates a local detour template. When other logistics robots request to approach the associated area, it obtains the current wheelbase and current total vehicle weight of the requesting vehicle to calculate the safety slack. Based on the algebraic determination result of the safety slack, it executes the dispatch instruction to issue the local detour template or refuse to issue it.

[0006] Preferably, the calculation of the base flatness index and base friction coefficient of the corresponding grid specifically includes: Extract all point cloud coordinate data falling within any single three-dimensional space grid, calculate the statistical variance of the extracted point cloud in the vertical elevation direction, and assign the statistical variance value as the basic flatness index of the current three-dimensional space grid. The average laser reflection intensity and average infrared echo absorption rate of all point clouds within the three-dimensional spatial grid are calculated and combined into a feature vector. The feature vector is input into a preset material mapping matrix, and the road surface material category corresponding to the current three-dimensional spatial grid is matched by the nearest neighbor interpolation method. The basic friction coefficient corresponding to the road surface material category is then output.

[0007] Preferably, the geometrical matching of the obstacle point cloud bounding box with the component data of the current construction stage in the model database, and the cross-stage re-matching, specifically include: Calculate the 3D intersection-union ratio (IUU) between the geometric bounding box of the obstacle point cloud and the bounding boxes of each component in the current construction stage. When all 3D IUU are lower than a preset first matching threshold, the matching in the current stage is determined to be a failure. Extract and merge component data from a fixed number of adjacent construction cycles before and after the current construction phase, and establish a sliding time window that includes historical completed progress and future construction progress. The 3D intersection-union ratio of the obstacle point cloud geometric bounding box and the bounding boxes of all components within the sliding time window is recalculated, and the maximum value is extracted as the maximum geometric matching degree.

[0008] Preferably, classifying the unknown obstacle as a pre-completed building upon successful matching specifically includes: When the maximum geometric matching degree is higher than the preset second matching threshold, the cross-stage rematch is confirmed to be successful, and the unknown obstacle is classified as an early-completed building caused by the actual construction progress exceeding the model database plan. The preset second matching threshold is greater than or equal to the preset first matching threshold. The state of the area occupied by the prematurely completed building in the three-dimensional physical attribute raster map is forcibly corrected to a static, unremovable state.

[0009] Preferably, the step of extracting the construction process attributes of the prematurely completed building and generating the physical environment-related degradation gradient specifically includes: The construction operation type and material information of the pre-completed building are extracted as specific construction process attributes. The corresponding deterioration influence radius, friction attribute attenuation coefficient and flatness deterioration coefficient are retrieved from the pre-set process influence factor mapping table. When multiple construction process keywords are parsed, the worst-case envelope extraction strategy is executed, and the largest value among them is selected as the final degradation parameter.

[0010] Preferably, the step of correcting the base flatness index and the base friction coefficient of each grid within the influence radius using a distance-based exponential decay formula specifically involves: For affected grids whose minimum Euclidean distance from the center coordinate to the boundary of the pre-completed building is less than or equal to the deterioration radius, the corrected flatness index is equal to the base flatness index multiplied by a first product factor, where the first product factor is a product of the flatness deterioration coefficient and the negative power of the natural constant, and the exponent of the negative power is the ratio of the minimum Euclidean distance to the deterioration radius. Its corrected friction coefficient is equal to the base friction coefficient multiplied by a second product factor, the second product factor being a product of the friction property attenuation coefficient and the negative power of the natural constant. When the affected grid is simultaneously within the deterioration radius of multiple prematurely completed buildings, the maximum value among multiple independent corrected flatness indices is extracted as the final corrected flatness index, and the minimum value among multiple independent corrected friction coefficients is extracted as the final corrected friction coefficient.

[0011] Preferably, the calculation of the maximum safe curvature limit for the logistics robot to maintain lateral slippage specifically involves: The commanded driving speed of the logistics robot is obtained. When the commanded driving speed is lower than the minimum safe speed threshold preset by the system, the minimum safe speed threshold is forcibly used for algebraic calculation. The maximum safe curvature limit is equal to the product of the corrected friction coefficient of the target extended grid and the gravitational acceleration constant, divided by the square of the maximum of the commanded driving speed and the minimum safe speed lower limit threshold.

[0012] Preferably, the step of truncating search branches with curvature exceeding the maximum safe curvature limit to generate local detour paths specifically includes: Calculate the actual curvature of the local trajectory extending from the current parent node to the surrounding child nodes. When the actual curvature is greater than the maximum safe curvature limit, assign the transfer cost to infinity to truncate the branch. When the actual curvature is less than or equal to the maximum safe curvature limit, the comprehensive transfer cost is calculated; the comprehensive transfer cost is the sum of the product of the distance weighting coefficient and the Euclidean distance between nodes, the product of the flatness weighting coefficient and the modified flatness index, and the product of the curvature penalty weighting coefficient and the proportion of the actual curvature to the maximum safe curvature limit; The local detour path is generated by iteratively expanding the combined transfer costs of each effective branch and extracting the node sequence with the minimum total cost using a backtracking algorithm.

[0013] Preferably, the calculation of the safety slack, and the execution of issuing the local detour template or rejecting the issuance of the scheduling instruction based on the algebraic determination result of the safety slack, specifically includes: The safety slack is equal to the ratio of the total mass of the reference robot to the current total mass of the requesting robot, multiplied by the ratio of the current wheelbase of the requesting robot to the wheelbase of the reference robot, and finally the product is reduced by one. When the safety slack is greater than or equal to zero, the path of the local detour template is pushed down to the requesting robot; When the safety relaxation is less than zero, the local detour template is rejected, and a replanning instruction is sent to the requesting robot to autonomously perform path reconstruction calculation.

[0014] A second aspect of the present invention provides a path optimization system for a construction logistics robot based on obstacle avoidance, comprising: The physics initialization module is used to acquire environmental point cloud data collected by the sensor group, discretize the environmental point cloud data into a three-dimensional spatial grid in the spatial dimension, extract point cloud features based on the preset material mapping matrix, calculate the basic flatness index and basic friction coefficient of the corresponding grid, and construct a three-dimensional physical property grid map containing physical parameters by combining the spatial benchmark in the model database. The obstacle classification module is used to identify unknown obstacles ahead of the path and extract the geometric bounding boxes of the obstacle point cloud during the movement of the logistics robot. The geometric bounding boxes of the obstacle point cloud are geometrically matched with the component data of the current construction stage in the model database. When the matching of the current stage fails, the module requests the component data of the adjacent construction cycle from the model database for cross-stage rematch. If the matching is successful, the unknown obstacle is classified as a building completed ahead of schedule. The degradation gradient generation module is used to extract the construction process attributes of the prematurely completed buildings from the model database for obstacles identified as prematurely completed buildings. Based on the extracted construction process attributes, the module radiates outward from the boundary of the prematurely completed building in the 3D physical property grid map. It uses a distance-based exponential decay formula to correct the base flatness index and base friction coefficient of each grid within the influence radius, thereby generating a physical environment-related degradation gradient. The path reconstruction module is used to read the current total mass data of the logistics robot and the driving speed command issued by the system. Combined with the updated corrected friction coefficient in the target extended grid, it calculates the maximum safe curvature limit for the logistics robot to maintain lateral slippage. The maximum safe curvature limit is introduced as a condition parameter into the cost function of the heuristic path search algorithm. The search branches with curvature exceeding the maximum safe curvature limit are truncated to generate local detour paths. The multi-robot scheduling module is used to report the coordinate data of the local detour path and the physical parameters of the vehicle to the cloud server after the logistics robot completes its journey along the local detour path. The cloud server receives the data and generates a local detour template. When other logistics robots request to approach the associated area, the module obtains the current wheelbase and total mass of the requesting vehicle to calculate the safety slack. Based on the algebraic determination result of the safety slack, the module executes the scheduling instruction to issue the local detour template or refuse to issue it.

[0015] This invention provides a method and system for optimizing the handling path of a construction logistics robot based on obstacle avoidance. It has the following beneficial effects: 1. This invention re-matches the geometric bounding box of the point cloud of unknown obstacles with the component data of adjacent construction cycles across stages, which can accurately identify buildings that are completed ahead of schedule due to the actual construction progress. Therefore, it can solve the problem that conventional obstacle avoidance algorithms misjudge fixed building entities as temporary obstacles and cause deadlock due to the inconsistency between the actual progress on site and the progress of the theoretical model. It reduces the robot's ineffective waiting and frequent planning actions in front of unknown obstacles and improves the stability of path optimization in dynamic construction environments.

[0016] 2. This invention generates a distance-based physical environment degradation gradient by extracting the construction process attributes of the building, and derives the maximum safe curvature limit by combining the robot's current total mass and expected speed. In this way, the actual physical state such as the decrease in surface friction can be transformed into the curvature cutoff condition in the pathfinding algorithm, so that the generated detour trajectory can actively avoid areas with the risk of sideslip, effectively preventing the heavy-duty logistics robot from lateral slippage or loss of control and overturning when passing through construction-polluted road surfaces, and ensuring the physical safety of chassis control.

[0017] 3. This invention constructs a safety relaxation assessment and multi-machine scheduling mechanism based on heterogeneous vehicle parameters. The cloud server uses the difference in total vehicle mass and wheelbase between the baseline robot and the requesting robot to determine the dynamic safety margin. Therefore, it can ensure that each heterogeneous robot, under the premise of meeting the anti-rollover requirements, allows subsequent vehicles with qualified safety relaxation to directly reuse the verified local detour template, avoiding a large amount of repetitive environmental perception and path search calculation, reducing the overall computing power consumption of the robot cluster, and improving the passage efficiency of multi-agents in complex construction areas. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a flowchart of the method for initializing a three-dimensional physical property raster map according to the present invention; Figure 4This is a flowchart of the obstacle classification and cross-stage matching method of the present invention; Figure 5 This is a flowchart of the degradation gradient generation method of the present invention; Figure 6 This is a flowchart of the local path reconstruction method of the present invention; Figure 7 This is a two-dimensional contour map of the corrected friction coefficient of the present invention; Figure 8 This is a comparison diagram of the local bypass spatial trajectories of the logistics robot under different algorithms of the present invention; Figure 9 This is a time-series response curve of lateral centrifugal acceleration according to the present invention; Figure 10 This is a bar chart showing the average computing power consumption per single trip for multiple machines according to the present invention.

[0019] Among them, 10 is a logistics robot; 11 is a sensor group; 12 is a pressure sensor; 13 is an industrial control computer; 20 is a cloud server; 30 is a model database; 100 is a physics field initialization module; 200 is an obstacle classification module; 300 is a degradation gradient generation module; 400 is a path reconstruction module; and 500 is a multi-machine scheduling module. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] See attached document Figure 1 , Figure 1 This is a schematic diagram of the system architecture of a construction logistics robot handling path optimization system based on obstacle avoidance, provided by an embodiment of the present invention. The present invention provides a construction logistics robot handling path optimization system based on obstacle avoidance, the system comprising: The physical field initialization module 100 is used to acquire environmental point cloud data collected by sensor group 11, discretize the environmental point cloud data into three-dimensional grid in spatial dimension, extract point cloud features based on preset material mapping matrix, calculate the basic flatness index and basic friction coefficient of the corresponding grid, and construct a grid map containing physical parameters by combining the spatial benchmark in model database 30. The obstacle classification module 200 is used to identify unknown obstacles ahead of the path and extract the geometric bounding box of the obstacle point cloud during the movement of the logistics robot 10 by using the sensor group 11. The obstacle point cloud geometric bounding box is geometrically matched with the component data of the current construction stage in the model database 30. When the matching of the current stage fails, the module requests the component data of the adjacent construction cycle from the model database 30 for cross-stage rematch. If the matching is successful, the unknown obstacle is classified as an early-completed building caused by the deviation of the construction schedule. The degradation gradient generation module 300 is used to extract the construction process attributes of the prematurely completed buildings from the model database 30 for obstacles identified as prematurely completed buildings. Based on the extracted construction process attributes, the module radiates outward from the boundary of the prematurely completed building in the three-dimensional physical attribute grid map. The module uses a distance-based exponential decay formula to correct the base flatness index and base friction coefficient of each grid within the influence radius, thereby generating a physical environment-related degradation gradient. The path reconstruction module 400 is used to read the current total mass data fed back by the pressure sensor 12 and the driving speed command issued by the system. It calculates the maximum safe curvature limit for the logistics robot 10 to maintain lateral slippage by combining the updated corrected friction coefficient within the target extended grid. The maximum safe curvature limit is introduced as a condition parameter into the cost function of the heuristic path search algorithm. The search branch with curvature exceeding the maximum safe curvature limit is truncated to generate a local detour path that adapts to the current load and ground physical state. The multi-machine scheduling module 500 is used to report the coordinate data of the local detour path and its own vehicle physical parameters to the cloud server 20 after the logistics robot 10 completes its journey along the local detour path. The cloud server 20 receives the data and generates a local detour template. When other logistics robots 10 request to approach the associated area, the module obtains the current wheelbase and current total mass of the requesting vehicle to calculate the safety slack. Based on the algebraic determination result of the safety slack, the module executes the scheduling instruction to issue the local detour template or refuse to issue the template.

[0022] See attached document Figure 2 , Figure 2 This is a flowchart illustrating a method for optimizing the handling path of a construction logistics robot based on obstacle avoidance, as provided in an embodiment of the present invention. The present invention provides a method for optimizing the handling path of a construction logistics robot based on obstacle avoidance, the method comprising: Step S10: The industrial control computer 13 acquires the environmental point cloud data collected by the sensor group 11, discretizes the environmental point cloud data into a three-dimensional grid in the spatial dimension, extracts point cloud features based on the preset material mapping matrix, calculates the basic flatness index and basic friction coefficient of the corresponding grid, and constructs a grid map containing physical parameters by combining the spatial benchmark in the model database 30. In step S20, during the movement of the logistics robot 10, the industrial control computer 13 identifies unknown obstacles ahead of the path through the sensor group 11 and extracts the geometric bounding box of the obstacle point cloud. The geometric bounding box of the obstacle point cloud is geometrically matched with the component data of the current construction stage in the model database 30. When the matching of the current stage fails, the computer requests the component data of the adjacent construction cycle from the model database 30 for cross-stage rematch. If the matching is successful, the unknown obstacle is classified as an early-completed building caused by the deviation of the construction progress. Step S30: For obstacles identified as prematurely completed buildings, the industrial control computer 13 extracts the construction process attributes of the prematurely completed buildings from the model database 30. Based on the extracted construction process attributes, it radiates outward from the boundary of the prematurely completed buildings in the three-dimensional physical attribute grid map. It uses the distance-based exponential decay formula to correct the foundation flatness index and foundation friction coefficient of each grid within the influence radius, and generates a physical environment associated degradation gradient. Step S40: The industrial control computer 13 reads the current total mass data fed back by the pressure sensor 12 and the driving speed command issued by the system. It calculates the maximum safe curvature limit for the logistics robot 10 to maintain lateral slippage by combining the updated corrected friction coefficient within the target extended grid. The maximum safe curvature limit is introduced as a condition parameter into the cost function of the heuristic path search algorithm. The search branch with curvature exceeding the maximum safe curvature limit is truncated to generate a local detour path that adapts to the current load and ground physical state. In step S50, after the logistics robot 10 completes its journey along the partial detour path, the industrial control computer 13 reports the coordinate data of the partial detour path and its own vehicle physical parameters to the cloud server 20. The cloud server 20 receives the data and generates a partial detour template. When other logistics robots 10 request to approach the associated area, the cloud server 20 obtains the current wheelbase and current total vehicle weight of the requesting vehicle, calculates the safety slack, and executes the dispatch instruction to issue a partial detour template or refuse to issue it based on the algebraic judgment result of the safety slack.

[0023] To further clarify the implementation of each technical aspect of the present invention, the following will provide a detailed description of the implementation of each functional module involved above and its internal processing flow.

[0024] See attached document Figure 3 , Figure 3 This is a flowchart of a method for initializing a three-dimensional physical property raster map according to an embodiment of the present invention. In this embodiment, step S10 is executed by the physics field initialization module 100 running inside the industrial control computer 13. To construct a bottom-level map that reflects the true physical properties of the environment, step S10 specifically includes the following steps: In step S101, the physics initialization module 100 acquires environmental point cloud data through the sensor group 11. Specifically, the environmental point cloud data includes reflection point cloud data acquired by the 3D LiDAR scan and absorptivity data acquired by the infrared rangefinder scan. The physics initialization module 100 fuses the reflection point cloud data and absorptivity data in the same physical coordinate system and divides the fused point cloud data into a 3D spatial grid with fixed side lengths in the spatial dimension. As a preferred method, the value range of the fixed side length is determined by the ground contact area of ​​the wheels of the logistics robot 10, and is usually set between 0.1 meters and 0.5 meters to ensure that the grid resolution can reflect the road surface details without causing computational redundancy. For the time synchronization and spatial extrinsic parameter calibration of the multi-sensor data of the 3D LiDAR and the infrared rangefinder, those skilled in the art can use existing joint calibration algorithms, which are well-known technologies in the field and will not be described in detail here.

[0025] Step S102: After spatial discretization of the point cloud data, to quantify the impact of ground surface undulation on vehicle driving posture, the physics initialization module 100 extracts the coordinate data of all point clouds falling within any single three-dimensional spatial grid after partitioning. Based on the extracted point cloud coordinate data, the physics initialization module 100 calculates the statistical variance of the extracted point cloud in the vertical elevation direction. Physically, the statistical variance in the vertical elevation direction reflects the longitudinal fluctuation of discrete points within a specific area; a larger fluctuation value indicates a more uneven road surface. The physics initialization module 100 assigns the calculated statistical variance value as the basic smoothness index of the current three-dimensional spatial grid. It should be noted that to ensure the validity of the statistical data, when the total number of point clouds falling within the three-dimensional spatial grid is less than a preset threshold, the three-dimensional spatial grid is not included in the calculation and is marked as an unknown area. The formula for calculating the basic smoothness index is as follows: ; In the formula, Indicates the first The basic flatness index of a three-dimensional spatial grid; Indicates falling into the first The total number of point clouds within a 3D spatial grid; Indicates the first Within the three-dimensional spatial grid, the first Vertical elevation values ​​of point clouds; Indicates the first The average vertical elevation of all point clouds within a three-dimensional spatial grid.

[0026] Step S103, besides elevation smoothness, the frictional characteristics of the road surface material are also key constraints for subsequent underlying kinematics optimization. The physics initialization module 100 calculates the... The average laser reflection intensity and average infrared echo absorption rate of all point clouds within a 3D spatial grid are calculated. The industrial control computer 13 stores a preset material mapping matrix. This material mapping matrix consists of data on the reflection characteristics and friction coefficient ranges of various building materials obtained from offline calibration experiments. Subsequently, the physics initialization module 100 combines the calculated average laser reflection intensity and average infrared echo absorption rate into a feature vector. Then, the physics initialization module 100 inputs this feature vector into the preset material mapping matrix, calculates the Euclidean distance between the feature vector and the sample data in the material mapping matrix using nearest neighbor interpolation, thereby matching the road surface material category corresponding to the current 3D spatial grid and outputting the basic friction coefficient corresponding to that road surface material category. In practical applications, the value range of this basic friction coefficient is usually normalized to the range of 0 to 1. The calculation logic of the basic friction coefficient is expressed by the following formula: ; In the formula, Indicates the first The basic friction coefficient of a three-dimensional spatial grid; Represents the mapping function corresponding to the preset material mapping matrix; Indicates the first The average laser reflection intensity of a point cloud within a three-dimensional spatial grid; Indicates the first Average infrared echo absorption rate of point clouds within a three-dimensional spatial grid.

[0027] In step S104, to form a complete map structure for subsequent path search algorithms, the physics initialization module 100 acquires the geometric occupancy status of each three-dimensional spatial grid. As a specific implementation, the physics initialization module 100 determines the difference between the maximum and minimum point cloud elevations within the three-dimensional spatial grid. When the elevation difference exceeds a set vehicle passage threshold, the geometric occupancy status of the three-dimensional spatial grid is confirmed as occupied. The physics initialization module 100 integrates the calculated basic flatness index, basic friction coefficient, and geometric occupancy status according to the global coordinate system and projects the integrated data into the model database 30. At this time, the physics initialization module 100 aligns and registers the projected data with the original spatial reference in the model database 30, generating a three-dimensional physical attribute grid map containing physical parameters. For the transformation and projection calculation of the global coordinate system, those skilled in the art can use conventional three-dimensional coordinate rotation and translation matrices. The coordinate transformation process is a well-known technique in the field and will not be elaborated here.

[0028] See attached document Figure 4 , Figure 4This is a flowchart of an obstacle classification and cross-stage matching method according to an embodiment of the present invention. In this embodiment, step S20 is executed by the obstacle classification module 200 running inside the industrial control computer 13. To solve the robot obstacle avoidance deadlock problem caused by the spatiotemporal asynchrony between the model's planned progress and the actual on-site progress in a construction environment, step S20 specifically includes the following steps: In step S201, during the movement of the logistics robot 10, when the sensor group 11 detects an unmarked abnormal point cloud cluster in the three-dimensional physical attribute grid map, the obstacle classification module 200 determines and identifies an unknown obstacle ahead of the path. To ensure the mathematical validity of subsequent geometric calculations, the obstacle classification module 200 counts the total number of contained points in the abnormal point cloud cluster. When the total number of contained points exceeds a preset threshold for the number of effective points for dimensionality reduction, the obstacle classification module 200 performs dimensionality reduction processing on the point cloud data of the unknown obstacle using principal component analysis. By extracting the orthogonal direction with the largest distribution variance of the point cloud in three-dimensional space, the obstacle classification module 200 calculates the eigenvalues ​​and eigenvectors of the point cloud covariance matrix, thereby determining the main direction distribution of the unknown obstacle. The obstacle classification module 200 generates a three-dimensional bounding box that tightly encloses the unknown obstacle along the main direction and extracts this three-dimensional bounding box as the geometric bounding box of the obstacle point cloud. After completing the feature extraction of the spatial dimension, the obstacle classification module 200 obtains the current timestamp of the system and queries and extracts the component bounding box data of the current construction stage corresponding to the timestamp from the model database 30. The obstacle classification module 200 calculates the 3D intersection-union ratio (IUGR) of the obstacle point cloud geometric bounding boxes with the bounding boxes of each component in the current construction stage. The IUGR is the ratio of the intersecting volume to the merged volume of two bounding boxes in space, reflecting the degree of spatial overlap between the physically measured boundary and the theoretical boundary of the model. The obstacle classification module 200 then compares the calculated IUGR with a preset first matching threshold. When all IUGRs are lower than the preset first matching threshold, the obstacle classification module 200 determines that the current stage matching has failed, indicating that the unknown obstacle does not belong to the preset building components in the current construction plan. The specific matrix solution process for point cloud clustering and principal component analysis can be implemented using existing point cloud processing algorithm libraries by those skilled in the art; its mathematical derivation is well-known in the field and will not be elaborated here. As a preferred approach, the preset first matching threshold is set between 0.6 and 0.8 to tolerate geometric deformation caused by point cloud scanning errors.

[0029] In step S202, if matching fails at the current stage, conventional obstacle avoidance algorithms often misclassify unknown obstacles as temporary debris or moving personnel, thus adopting strategies such as waiting in place or frequent replanning. To accurately identify the physical attributes of obstacles, the obstacle classification module 200 initiates a cross-stage data request to the model database 30, extracting component data from a fixed number of adjacent construction cycles before and after the current construction stage. Let the preset time extension span be... In each construction cycle, specifically, the obstacle classification module 200 obtains the lower and upper time limits for the current project from the system's overall construction cycle schedule. When extracting data, the obstacle classification module 200 performs boundary truncation; if the forward or backward cycle exceeds the lower or upper time limit, extraction only continues up to the boundary. The obstacle classification module 200 will extract... The component datasets for each construction cycle are merged to create a sliding time window that includes historical completed progress and future construction progress. The obstacle classification module 200 recalculates the 3D intersection-union ratio (IUU) of the obstacle point cloud geometric bounding boxes with the bounding boxes of all components within the sliding time window, and extracts the maximum value as the maximum geometric matching degree. The formula for calculating the maximum geometric matching degree is as follows: ; In the formula, This represents the maximum geometric fit of an unknown obstacle within the sliding time window; This indicates the number of construction cycles that the sliding time window extends forward and backward along the time axis; Represents the geometric bounding box of the extracted obstacle point cloud; This indicates the timestamp corresponding to the current construction phase; Indicates the first time within the sliding time window The component enclosure box set corresponding to each construction cycle; This represents a function for calculating the geometric volume of a space. This represents the operation of finding the geometric intersection; This represents the geometric union operation. As a specific implementation, the time span is extended. The value range is set to 1 to 3 construction cycles to ensure that the search range covers common construction progress deviation ranges.

[0030] In step S203, after obtaining the maximum matching value within the sliding time window, it is also necessary to define the final attributes of the unknown obstacle. Therefore, the obstacle classification module 200 compares the calculated maximum geometric matching degree with a preset second matching threshold. As a preferred method, to avoid misjudgment, the preset second matching threshold should be greater than or equal to the preset first matching threshold, typically set to 0.75 to 0.9. When the maximum geometric matching degree is higher than the preset second matching threshold, the obstacle classification module 200 confirms that the unknown obstacle highly matches the planned components of a future construction phase in terms of geometric shape and spatial location. Therefore, the obstacle classification module 200 classifies the unknown obstacle as a prematurely completed building due to the actual construction progress exceeding the plan in the model database 30. To prevent the local path search algorithm from attempting to traverse this area in subsequent runs, the obstacle classification module 200, based on the classification result, forcibly corrects the state of the area occupied by the prematurely completed building in the 3D physical attribute raster map to an unremovable static occupation. This state correction mechanism eliminates the possibility of misclassifying building entities as temporary obstacles, providing stable and reliable spatial boundary conditions for the subsequent collaborative optimization of underlying physical properties and paths. Conversely, when the maximum geometric matching degree is lower than or equal to a preset second matching threshold, the obstacle classification module 200 classifies the unknown obstacle as an unrecorded temporary dynamic obstacle and maintains the original updatable detection properties of the area, allowing the local path search algorithm to restore passage after subsequent scanning determines that it has been removed.

[0031] In this embodiment, step S30 is executed by the degradation gradient generation module 300 running inside the industrial control computer 13. Since the construction work on building components typically has spillover effects on the physical state of the surrounding local environment, and in order to transform this unknown environmental disturbance into a calculable mathematical constraint, step S30 specifically includes the following steps: See attached document Figure 5 , Figure 5 This is a flowchart of a method for generating degradation gradients according to an embodiment of the present invention. In step S301, for obstacles that have been classified and confirmed as prematurely completed buildings, the degradation gradient generation module 300 obtains the ID of the building component that has successfully undergone cross-stage rematching. Specifically, the degradation gradient generation module 300 uses this building component ID as the query primary key and synchronously extracts the non-geometric text data of the building component within the corresponding construction period from the model database 30. The non-geometric text data in the model database 30 may include the current construction operation type and material information of the building component. The degradation gradient generation module 300 confirms the extracted non-geometric text data as the specific construction process attribute corresponding to the prematurely completed building. The specific implementation mechanism for text data extraction and database retrieval can be accomplished by those skilled in the art using standard structured query language, which is a well-known technology in the field and will not be elaborated here.

[0032] Step S302: Since different construction processes will have varying degrees of damage or pollution to the surrounding ground, it is necessary to quantify this impact. Specifically, the industrial control computer 13 can store a pre-set process impact factor mapping table. This mapping table records the correspondence between common construction process keywords and multi-dimensional degradation parameters. Therefore, the degradation gradient generation module 300 parses the extracted construction process attributes using a string matching algorithm, retrieves the corresponding process entries in the mapping table, and assigns them the corresponding degradation impact radius, friction attribute attenuation coefficient, and flatness deterioration coefficient. In the actual parsing process, multiple construction operations may be carried out in parallel on the same building component. When the degradation gradient generation module 300 parses multiple construction process keywords, the system executes the most unfavorable working condition envelope extraction strategy, that is, compares the impact radius and attenuation coefficient corresponding to each process respectively, and selects the largest value as the final degradation parameter of the currently prematurely completed building. As a specific implementation method, when the construction process attribute includes the keyword "concrete pouring," mud splashing is likely to occur around the site, and the degradation gradient generation module 300 assigns a larger degradation influence radius and a higher friction attribute attenuation coefficient. When the construction process attribute includes the keyword "rebar cutting," the surrounding area is mostly scattered solid debris, and the degradation gradient generation module 300 assigns a smaller degradation influence radius and a higher smoothness deterioration coefficient. As a preferred method, the friction attribute attenuation coefficient is set to a range of 0.1 to 0.4, and the smoothness deterioration coefficient is set to a range of 0.2 to 1.5 to match the surface variation range of a typical construction site.

[0033] In step S303, after obtaining the specific degradation parameters, the degradation gradient generation module 300 calculates the influence field outward from the boundary of the prematurely completed building in the three-dimensional physical property raster map. The degradation gradient generation module 300 traverses the surrounding graticles in the three-dimensional physical property raster map and calculates the minimum Euclidean distance from the center coordinates of each surrounding graticle to the boundary of the prematurely completed building. For graticles whose minimum Euclidean distance is less than or equal to the degradation influence radius, the degradation gradient generation module 300 determines them as affected graticles. From the perspective of physical diffusion laws, the concentration of dust, mud, and debris generated during construction exhibits a non-linear decay with increasing distance. Therefore, the degradation gradient generation module 300 substitutes the minimum Euclidean distance of the affected graticles into the distance-based exponential decay formula to numerically correct the foundation smoothness index and foundation friction coefficient generated in step S10, generating a physical environment-related degradation gradient and outputting the corrected smoothness index and corrected friction coefficient. The correction calculation is expressed by the following formula: ; ; In the formula, Indicates the first Corrected flatness index for each affected raster; Indicates the first The base flatness index of the affected grid; Indicates the first Corrected friction coefficient for each affected grid cell; Indicates the first The base friction coefficient of each affected grid; This represents the flatness deterioration coefficient assigned to the current construction process attribute, and ; This represents the frictional property attenuation coefficient assigned to the current construction process attribute, and ; Indicates the first The minimum Euclidean distance from the center of each affected grid cell to the boundary of the prematurely completed building; This indicates the radius of deterioration impact assigned to the current construction process attribute; Represents the natural constant.

[0034] Considering the possibility of multiple adjacent prematurely completed buildings at the construction site, when a surrounding grid cell is simultaneously within the degradation influence radius of multiple prematurely completed buildings, the degradation gradient generation module 300 calculates the independent correction values ​​for that grid cell caused by the influence of each building. The degradation gradient generation module 300 extracts the maximum value among multiple independent corrected smoothness indices as the final corrected smoothness index for that grid cell, and extracts the minimum value among multiple independent corrected friction coefficients as the final corrected friction coefficient for that grid cell. When the minimum Euclidean distance of a surrounding grid cell is greater than all associated degradation influence radii, the degradation gradient generation module 300 determines that the surrounding grid cell has not been affected by the overflow of construction work, and directly uses the original base smoothness index and base friction coefficient as the final corrected values. By establishing this continuously changing physical property attenuation field, the gradual deterioration process of the surrounding environmental gradient in the construction area can be more realistically reflected, thus providing accurate terrain physical boundary constraint inputs for underlying motion control and local path planning.

[0035] See attached document Figure 6 , Figure 6 This is a flowchart of a local path reconstruction method according to an embodiment of the present invention. In this embodiment, step S40 is executed by the path reconstruction module 400 running inside the industrial control computer 13. In order to plan a trajectory in the degraded local physical field that can both avoid obstacles and prevent vehicle sideslip and bump damage, step S40 specifically includes the following steps: S401, the path reconstruction module 400 acquires the robot's real-time total mass and desired speed. When the logistics robot 10 performs a handling task, its empty mass and the load of the goods it carries together constitute the total mass of the vehicle. Therefore, the path reconstruction module 400 can continuously read the total load data of the vehicle body and goods collected by the pressure sensors installed at the bottom of the logistics robot 10's platform via the underlying interface, calculate and output the robot's real-time total mass. Subsequently, the underlying control system of the logistics robot 10 dynamically adjusts its torque output and maximum speed limit based on the calculated real-time total mass. On this basis, the path reconstruction module 400 communicates with the motion control chassis of the logistics robot 10 to synchronously acquire the desired travel speed issued by the control command under the current load state. As for the sensor data reading and the parsing of the underlying communication protocol, those skilled in the art can use standard bus communication protocols, which are well-known technologies in the field and will not be elaborated upon here.

[0036] Step S402 transforms the abstract physical friction into specific spatial geometric constraints. Specifically, the path reconstruction module 400 constructs a lateral slip critical dynamic model to solve for the maximum safe curvature limit.

[0037] Specifically, since a vehicle inevitably generates lateral centrifugal force when performing a steering maneuver around an obstacle, when this lateral centrifugal force exceeds the maximum static friction between the tires and the ground, the vehicle will slip laterally, thus triggering a risk of loss of control under heavy load. Therefore, the path reconstruction module 400 constructs a critical dynamic model for lateral slip based on the extracted desired driving speed and the corrected friction coefficient of the corresponding three-dimensional spatial grid generated in step S30. In this dynamic model, the critical condition for maintaining the vehicle from slipping is that the centripetal acceleration does not exceed the maximum friction acceleration that the ground can provide. The path reconstruction module 400 multiplies the corrected friction coefficient of the current three-dimensional spatial grid by the gravitational acceleration and divides it by the square of the desired driving speed to obtain the maximum safe curvature limit for maintaining the logistics robot 10 from slipping under the current total mass state of the vehicle and the current physical properties of the road surface. To avoid mathematical overflow errors caused by the denominator of the formula approaching zero when the vehicle is stationary or starting at extremely low speed, the path reconstruction module 400 sets a minimum safe speed lower limit threshold for the desired driving speed. When the expected driving speed is lower than the lower threshold, the system forces the use of this lower threshold for algebraic calculations. The formula for calculating the maximum safe curvature limit is as follows: ; In the formula, This indicates that the logistics robot 10 is in the Maximum safe curvature limit when driving within a three-dimensional spatial grid; Indicates the first Corrected friction coefficient for a three-dimensional spatial grid; Represents the gravitational acceleration constant; This indicates the desired driving speed given by the current control command; This represents the minimum safe speed threshold preset by the system, typically set to 0.1 m / s; This represents the function that takes the maximum value.

[0038] In step S403, after obtaining the physical properties and dynamic constraints of each 3D spatial grid, the system needs to integrate them into the cost evaluation system of the pathfinding algorithm. Specifically, the path reconstruction module 400 uses a heuristic search algorithm with a curvature penalty truncation mechanism for pathfinding. The path reconstruction module 400 generates a search tree by expanding nodes towards the target point in the map space. When generating a local trajectory extending from the current parent node to surrounding child nodes, the path reconstruction module 400 calculates the actual curvature corresponding to the local trajectory. The path reconstruction module 400 compares the calculated actual curvature with the maximum safe curvature limit of the grid where the child node is located. If the actual curvature is greater than the maximum safe curvature limit, it indicates that the transfer branch has a risk of sideslip. The path reconstruction module 400 directly assigns the transfer cost of the transfer path to infinity, thereby truncating the branch in the algorithm search tree and preventing the logistics robot 10 from attempting to enter the trajectory with a risk of sideslip. Conversely, if the actual curvature is less than or equal to the maximum safe curvature limit, the path reconstruction module 400 determines that the local trajectory has lateral stability under the current friction conditions. The path reconstruction module 400 further combines the physical distance between nodes, the corrected flatness index of the raster where the child node is located, and the proportion of the actual curvature in the maximum safe curvature limit to calculate the comprehensive transfer cost. The formula for calculating the comprehensive transfer cost is as follows: ; In the formula, This indicates the extension from the parent node to the child node at the specified position. The overall transfer cost of a three-dimensional spatial grid; , and Let represent the distance weighting coefficient, the flatness weighting coefficient, and the curvature penalty weighting coefficient, respectively. ; This represents the Euclidean distance between a parent node and a child node. Indicates the first Corrected flatness index of a three-dimensional spatial grid; This represents the actual curvature of the current local trajectory; Indicates the first The maximum safe curvature limit for a three-dimensional spatial grid. As a preferred approach, considering the high safety requirements for anti-tipping in heavy-duty logistics scenarios, the curvature penalty weighting coefficient... The value range is usually set to 0.4 to 0.6, which is the distance weighting coefficient. Smoothness weighting coefficient The remaining quota will then be allocated to ensure that dynamic physical security is prioritized during path planning.

[0039] As the search tree progresses, the path reconstruction module 400 continuously calculates and accumulates the comprehensive transfer cost of each effective branch, iteratively expanding from the starting node to the target node. When the search reaches the area where the target node is located, the path reconstruction module 400 extracts the node sequence with the minimum total cost using a backtracking algorithm, outputting the optimal detour path that takes into account the degradation of the underlying physical environment. This optimal detour path spatially avoids pre-completed buildings that are statically occupied, and selects a driving trajectory that meets the requirements of dynamic stability within the surrounding degraded environment, thereby ensuring the smooth execution of heavy-duty logistics handling tasks in building scenarios.

[0040] In this embodiment, step S50 is executed by the multi-machine scheduling module 500 running inside the cloud server 20. Since reusing verified safe paths can significantly reduce the computing power consumption of a single machine in complex multi-machine collaborative construction scenarios, step S50 specifically includes the following steps to achieve safe sharing of underlying obstacle avoidance experience among heterogeneous robot clusters: In step S501, after a logistics robot 10 successfully exits the environmentally deteriorated area on the local path generated in step S40, the multi-robot scheduling module 500 marks it as the baseline robot. In actual engineering applications, multiple robots may pass through the same area successively. To ensure the optimality of the cloud-based experience database, the multi-robot scheduling module 500 continuously compares the comprehensive transfer costs of the uploaded paths of each robot and selects the robot with the lowest historical comprehensive transfer cost as the final baseline robot for that area. After determining the baseline object, the multi-robot scheduling module 500 extracts the path coordinate sequence of the baseline robot successfully completing the detour task, and simultaneously collects the physical attribute parameters of the baseline robot when performing the task, specifically including the total vehicle mass and the vehicle wheelbase. To ensure the rigor of the data definition and the legality of subsequent calculations, the total vehicle mass is defined as the total mass of the vehicle plus the weight of the empty vehicle and the weight of the cargo, and this value is always greater than zero. The multi-robot scheduling module 500 packages the extracted path coordinate sequence, total vehicle mass, and vehicle wheelbase into a local detour template and simultaneously uploads it to the model database of the cloud server 20 for storage. As for the wireless data transmission and synchronization mechanism between the robot and the cloud server 20, those skilled in the art can use standard wireless communication protocols and message queue telemetry transmission technology, which are well-known technologies in the field and will not be elaborated here.

[0041] In step S502, as construction progresses, when a new logistics robot 10 (i.e., a requesting logistics robot 10) is detected approaching the physically deteriorated area, the system needs to assess the physical safety of directly reusing historical experience paths. Since robots carrying different tasks are often deployed together on-site, the multi-robot scheduling module 500 performs a safety relaxation determination driven by heterogeneous parameters. Specifically, the requesting logistics robot 10 sends a request command to the cloud server 20 and simultaneously reports its real-time total vehicle mass and wheelbase data. Subsequently, the cloud server 20 extracts the wheelbase and total vehicle mass of the requesting robot approaching the area. Specifically, based on the physical principles of vehicle dynamics and rollover stability, compared to the baseline robot, if the requesting robot has a lighter total vehicle mass (i.e., less lateral slip inertia and centrifugal force) and a wider wheelbase (i.e., a larger chassis support polygon area), its anti-rollover capability is stronger when traveling along the same trajectory. Based on this physical logic, the multi-machine scheduling module 500 divides the total mass of the reference robot by the total mass of the requesting robot, multiplies this ratio by the ratio of the wheelbase of the requesting robot to the wheelbase of the reference robot, and finally subtracts one from the product to calculate the dimensionless safety slack. The formula for calculating the safety slack is as follows: ; In the formula, This indicates the safety slack required for the robot to reuse the local bypass template; This indicates the total mass of the reference robot. This indicates the total mass of the requested robot vehicle; This indicates the requested wheelbase of the robot. This indicates the wheelbase of the reference robot.

[0042] In step S503, after obtaining the quantified safety assessment indicators, the system can execute differentiated path distribution instructions. Specifically, the multi-machine scheduling module 500 executes an adaptive distribution strategy based on the numerical polarity of the safety relaxation. When the calculated safety relaxation is greater than or equal to zero, it indicates that the dynamic stability of the requesting robot is superior to or equal to that of the benchmark robot. The multi-machine scheduling module 500 determines that the empirical path in the local detour template has physical downlink compatibility and directly pushes the execution path to the requesting robot. After receiving the path, the requesting robot directly executes line-following motion, thereby saving the computing power required for local environment perception and path reconstruction.

[0043] When the calculated safety relaxation is less than zero, it indicates that the requesting robot's center of gravity load conditions are worse than the baseline robot. Blindly reusing this empirical path poses a risk of rollover when traversing sharp curves or severely bumpy grids. To ensure the driving safety of the execution layer, the multi-machine scheduling module 500 refuses to issue this local detour template and sends a replanning instruction to the requesting robot. After receiving the replanning instruction, the requesting robot calls its own path reconstruction module 400 to autonomously perform path reconstruction calculations based on its actual load constraints and the surrounding 3D physical attribute grid map, generating a local obstacle avoidance path that conforms to the current physical constraints. Through the above-mentioned experience solidification and adaptive scheduling mechanism, the system improves the overall passage efficiency of the multi-agent cluster in complex construction environments while ensuring the physical safety boundaries of individuals.

[0044] To enable those skilled in the art to more clearly understand the purpose, technical solution, and advantages of this invention, the present invention will be further described in detail below with reference to specific application embodiments, real experimental test data, and corresponding drawings. It should be noted that the embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0045] I. Specific Application Examples Suppose an indoor commercial complex is currently undergoing simultaneous construction of electromechanical installation and secondary masonry work. Logistics robots A, B, and C are deployed on-site to handle building material transport tasks.

[0046] 1. Physics field initialization (corresponding to step S10) The industrial control computer 13 fuses LiDAR and infrared data, dividing the ground into a 0.2m × 0.2m three-dimensional grid. The current area is primarily ordinary concrete; the system calculates the base friction coefficient of this area using a pre-set material mapping matrix. Basic flatness index .

[0047] 2. Obstacle classification and cross-stage matching (corresponding to step S20) Logistics robot A (lead vehicle, total mass 500kg, wheelbase 1.0m) is traveling at a desired speed of 1.2m / s. Sensors ahead of its path detect an obstacle point cloud with a volume of approximately 2m × 2m × 3m. A query of model database 30 indicates that this location should be empty during the current construction cycle (match failed). The system automatically extends the BIM progress by two construction cycles (i.e., the next 14 days) and finds a C30 load-bearing column at this location. The calculated 3D intersection-union ratio (IUU) between the actual point cloud and the model bounding box is 0.88, higher than the second matching threshold of 0.80. The system classifies this obstacle as an early-completed building and changes its grid status to statically occupied.

[0048] 3. Degradation gradient generation (corresponding to step S30) The system extracts the construction process attribute of the C30 load-bearing column as concrete pouring and vibration. The influence radius of this process is obtained by referring to a table. Friction property attenuation coefficient Smoothness deterioration coefficient The system radiates outward from the column boundary. For example, at a grid point 1.5m away from the column, the exponential decay formula is used to calculate: Corrected coefficient of friction: (Friction is significantly reduced, posing a risk of slippery mud). Corrected flatness index .

[0049] 4. Local path reconstruction (corresponding to step S40) When passing through the aforementioned degraded grid, the desired speed of logistics robot A is 1.2 m / s. Substituting into the critical dynamics formula for lateral slip, the maximum safe curvature limit of the grid is obtained. In heuristic pathfinding, if a detour branch requires the robot to make a sharp turn (actual curvature...) Since 4.0 > 3.47, the system determines that there is a risk of sideslip and truncates the branch by assigning an infinite cost. Ultimately, the system plans a safe local detour path for logistics robot A that has a gentle curvature and is close to the periphery of the impact zone.

[0050] 5. Multi-machine scheduling and experience reuse (corresponding step S50) After robot A safely passes through, it will compare the path coordinates with its own parameters ( kg, m) is uploaded to the cloud server as a partial detour template 20.

[0051] Scenario 1: Robot B requests to approach the area. B is an empty robot with a total mass of [missing information]. kg, wheelbase m. Cloud computing security relaxation: The cloud directly sends A's template to B, allowing B to bypass pathfinding computation and safely follow the path.

[0052] Scenario 2: Robot C requests to approach the area. Robot C is fully loaded with heavy objects; the total mass of the vehicle is... kg, and the chassis has a narrow wheelbase m. Cloud computing security relaxation: The cloud refused to issue the command, and instruction C replanned the path based on its own overload characteristics (C ultimately planned an arc that was farther away from the pillar and had a smaller curvature).

[0053] II. Experimental Verification and Effect Comparison To verify the technical effects of the present invention, a simulation experiment was also conducted in this embodiment for comparison.

[0054] The test procedure was set as follows: a 50m×50m three-dimensional physical simulation environment for a construction site was constructed using the Robotics-System-Toolbox and Navigation-Toolbox in MATLAB-R2025a software. Real construction condition parameters (including building component dimensions, initial ground friction coefficient, and flatness index) were imported into the simulation environment, and a cluster of 20 heterogeneous logistics robots with different total vehicle weights and wheelbases was deployed to perform the handling task. Three comparison groups were set up: Control group 1 used the traditional A* geometric obstacle avoidance algorithm (treating obstacles only as impassable meshes, without considering the degradation of surrounding physical properties); Control group 2 used the Dynamic Window Method (DWA, which considers the robot's kinematic constraints but lacks the cross-stage recognition and multi-robot collaborative experience reuse mechanism for early-completed buildings); the experimental group used the obstacle avoidance-based construction logistics robot handling path optimization system and method proposed in this invention.

[0055] The results are as follows: Reference Figure 7 As shown, Figure 7 This is a two-dimensional contour map of the corrected friction coefficient, generated using MATLAB software. Figure 7 In the diagram, the central highlighted area represents the static occupied area of ​​a building that was classified as prematurely completed after cross-stage matching. The contour data clearly shows that the closer the grid area is to the building's boundary, the cooler its color (representing a significant drop in the corrected friction coefficient to approximately 0.4 due to the influence of construction mud, etc.); while with increasing Euclidean distance, the corrected friction coefficient exhibits an exponential gradual change and smoothly recovers to the initial baseline value of 0.65 (warm color area). This indicates that the degradation gradient generation module of this invention can accurately and continuously quantify the physical spillover interference of construction operations on the surrounding ground in space, fully conforming to the attenuation law of the actual physical environment at the construction site.

[0056] Reference Figure 8 and Figure 9 As shown, Figure 8 Comparison of local bypass trajectories of logistics robots under different algorithms. Figure 9 This is the corresponding time-series response curve for lateral centrifugal acceleration. Combined with... Figure 8 It can be seen that, in pursuit of the shortest travel distance, the control group 1 (A* algorithm) planned a trajectory that closely followed the edge of obstacle degradation, resulting in a trajectory with a peak curvature as high as 5.2 rad / m; while the trajectory of the experimental group of this invention, due to the introduction of a maximum safe curvature limit truncation mechanism in the cost function, spontaneously shifted outwards, and the maximum trajectory curvature was strictly limited to within the dynamically permissible 3.5 rad / m. Further reference... Figure 9The timing response data shows that when the robot travels within this locally degraded area, the lateral acceleration curves of both control group 1 and control group 2 significantly exceed the centripetal acceleration threshold that can be provided by the maximum static friction force of the current grid. Figure 9 The red dashed line in the image indicates an overshoot of approximately 42%, which manifests as severe sideslip and loss of control in the simulation physics engine. In contrast, the acceleration curve of the experimental group using this invention consistently remained smoothly below the safety threshold, with a safety margin exceeding 15%. This demonstrates that the present invention effectively prevents the generation of high-risk local trajectories and completely eliminates the risks of sideslip and overturning under heavy load conditions.

[0057] Reference Figure 10 As shown, Figure 10 The bar chart shows the average computing power consumption per single detour by multiple robots. In a scenario simulating 20 heterogeneous logistics robots successively requesting to pass through the same physically deteriorated area, under the traditional independent planning mode (i.e., without cloud scheduling strategy), the average computing power consumption time per robot per trip is stable at about 145ms. However, in the experimental group of this invention, apart from the first baseline robot taking 152ms to complete the reconstruction, among the subsequent 19 robots requesting passage, 12 had their safety slack greater than zero after calculation by the cloud server (indicating that their total vehicle weight is lighter or their wheelbase is wider, and their anti-overturning ability is stronger), thus directly reusing the local detour template issued by the cloud, and their single pathfinding computing power consumption time plummeted to only 8ms (including only communication and data reading time); the remaining 7 robots with safety slack less than zero were rejected and re-planned autonomously according to instructions (average time 148ms). Comprehensive data shows that the multi-machine scheduling and adaptive experience reuse mechanism of the present invention reduces the total computing power consumption of the robot cluster in the complex construction area by about 58% while ensuring the safety of the vehicle's underlying dynamics. It also reduces the deadlock time of the vehicle waiting in place for replanning in front of unknown obstacles and improves the overall passage efficiency of the multi-agent cluster.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the handling path of a construction logistics robot based on obstacle avoidance, characterized in that, Includes the following steps: The environmental point cloud data collected by the sensor group is acquired and discretized into a three-dimensional spatial grid. The basic flatness index and basic friction coefficient of the corresponding grid are calculated based on the material mapping matrix. A three-dimensional physical property grid map is constructed by combining the model database. During the movement of the logistics robot, unknown obstacles ahead of the path are identified and the geometric bounding boxes of the obstacle point cloud are extracted. The geometric bounding boxes of the obstacle point cloud are matched with the component data in the model database across stages. If the match is successful, the unknown obstacle is classified as a building completed ahead of schedule. For obstacles identified as the prematurely completed building, the construction process attributes of the prematurely completed building are extracted from the model database. In the three-dimensional physical attribute grid map, the foundation flatness index and foundation friction coefficient of each grid within the influence radius are attenuated and corrected to generate a physical environment associated degradation gradient. The current total mass data and commanded driving speed of the logistics robot are read, and the maximum safe curvature limit of the logistics robot is calculated by combining the corrected friction coefficient of the target extended grid. The search branches with curvature exceeding the maximum safe curvature limit are truncated in the cost function of the heuristic path search algorithm to generate a local detour path. After the logistics robot completes its journey along the local detour path, it reports the coordinate data of the local detour path and its own vehicle physical parameters to the cloud server to generate a local detour template. When the remaining logistics robots request to approach the associated area, the current wheelbase and current total vehicle weight of the requesting vehicle are obtained to calculate the safety slack, and the dispatching command to issue the local detour template or refuse to issue it is executed based on the safety slack.

2. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 1, characterized in that, The calculation of the basic flatness index and basic friction coefficient of the corresponding grid specifically includes: Extract all point cloud coordinate data falling within any single three-dimensional spatial grid, calculate the statistical variance of the extracted point cloud in the vertical elevation direction, and assign the statistical variance value as the basic flatness index of the current three-dimensional spatial grid. The average laser reflection intensity and average infrared echo absorption rate of all point clouds within the three-dimensional spatial grid are calculated and combined into a feature vector. The feature vector is input into a preset material mapping matrix, and the road surface material category corresponding to the current three-dimensional spatial grid is matched by the nearest neighbor interpolation method. The basic friction coefficient corresponding to the road surface material category is then output.

3. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 1, characterized in that, The process of matching the obstacle point cloud geometric bounding boxes with the component data in the model database across stages specifically includes: Calculate the 3D intersection-union ratio (IUU) between the geometric bounding box of the obstacle point cloud and the bounding boxes of each component in the current construction stage. When all 3D IUU are lower than the preset first matching threshold, the matching in the current stage is determined to be a failure. Extract and merge component data from a fixed number of adjacent construction cycles before and after the current construction phase, and establish a sliding time window that includes historical completed progress and future construction progress. The 3D intersection-union ratio of the obstacle point cloud geometric bounding box and the bounding boxes of all components within the sliding time window is recalculated, and the maximum value is extracted as the maximum geometric matching degree.

4. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 3, characterized in that, The step of classifying the unknown obstacle as a prematurely completed building upon a successful match specifically includes: When the maximum geometric matching degree is higher than the preset second matching threshold, the cross-stage rematch is confirmed to be successful, and the unknown obstacle is classified as an early-completed building caused by the actual construction progress exceeding the model database plan. The preset second matching threshold is greater than or equal to the preset first matching threshold. The state of the area occupied by the prematurely completed building in the three-dimensional physical attribute raster map is forcibly corrected to a static, unremovable state.

5. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 1, characterized in that, The extraction of construction process attributes of the prematurely completed building and the generation of the physical environment-related degradation gradient specifically include: The construction operation type and material information of the pre-completed building are extracted as specific construction process attributes. The corresponding deterioration influence radius, friction attribute attenuation coefficient and flatness deterioration coefficient are retrieved from the pre-set process influence factor mapping table. When multiple construction process keywords are parsed, the worst-case envelope extraction strategy is executed, and the maximum value is selected as the final degradation parameter.

6. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 5, characterized in that, The attenuation correction of the basic flatness index and basic friction coefficient of each grid within the influence radius is specifically performed using a distance-based exponential attenuation formula, including: For affected grids whose minimum Euclidean distance from the center coordinate to the boundary of the pre-completed building is less than or equal to the deterioration radius, the corrected flatness index is equal to the base flatness index multiplied by a first product factor, where the first product factor is a product of the flatness deterioration coefficient and the negative power of the natural constant, and the exponent of the negative power is the ratio of the minimum Euclidean distance to the deterioration radius. The corrected friction coefficient is equal to the base friction coefficient multiplied by a second product factor, where the second product factor is a product of the friction property attenuation coefficient and the negative power of the natural constant. When the affected grid is simultaneously within the deterioration radius of multiple prematurely completed buildings, the maximum value among multiple independent corrected flatness indices is extracted as the final corrected flatness index, and the minimum value among multiple independent corrected friction coefficients is extracted as the final corrected friction coefficient.

7. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 6, characterized in that, The calculation of the maximum safe curvature limit for the logistics robot to maintain lateral slippage is specifically as follows: The commanded driving speed of the logistics robot is obtained. When the commanded driving speed is lower than the minimum safe speed threshold preset by the system, the minimum safe speed threshold is forcibly used for algebraic calculation. The maximum safe curvature limit is equal to the product of the corrected friction coefficient of the target extended grid and the gravitational acceleration constant, divided by the square of the maximum of the commanded driving speed and the minimum safe speed lower limit threshold.

8. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 7, characterized in that, The search branch whose truncation curvature exceeds the maximum safe curvature limit generates a local detour path, specifically including: Calculate the actual curvature of the local trajectory extending from the current parent node to the surrounding child nodes. When the actual curvature is greater than the maximum safe curvature limit, assign the transfer cost to infinity to truncate the branch. When the actual curvature is less than or equal to the maximum safe curvature limit, the comprehensive transfer cost is calculated; the comprehensive transfer cost is the sum of the product of the distance weighting coefficient and the Euclidean distance between nodes, the product of the flatness weighting coefficient and the modified flatness index, and the product of the curvature penalty weighting coefficient and the proportion of the actual curvature to the maximum safe curvature limit; The local detour path is generated by iteratively expanding the combined transfer costs of each effective branch and extracting the node sequence with the minimum total cost using a backtracking algorithm.

9. The method for optimizing the handling path of a construction logistics robot based on obstacle avoidance according to claim 1, characterized in that, The calculation of the safety slack, and the execution of the dispatch instruction to issue the local detour template or refuse to issue it based on the algebraic determination result of the safety slack, specifically includes: The safety slack is equal to the ratio of the total mass of the reference robot to the current total mass of the requesting robot, multiplied by the ratio of the current wheelbase of the requesting robot to the wheelbase of the reference robot, and finally the product is reduced by one. When the safety slack is greater than or equal to zero, the path of the local detour template is pushed down to the requesting robot; When the safety relaxation is less than zero, the local detour template is rejected, and a replanning instruction is sent to the requesting robot to autonomously perform path reconstruction calculation.

10. A system for implementing the obstacle avoidance-based material handling path optimization method for construction logistics robots as described in any one of claims 1-9, characterized in that, include: The physics initialization module is used to acquire environmental point cloud data collected by the sensor group and discretize it into a three-dimensional spatial grid. Based on the material mapping matrix, it calculates the basic flatness index and basic friction coefficient of the corresponding grid and constructs a three-dimensional physical property grid map in combination with the model database. The obstacle classification module is used to identify unknown obstacles ahead of the path and extract the geometric bounding boxes of the obstacle point cloud during the movement of the logistics robot. The obstacle point cloud geometric bounding boxes are then matched with the component data in the model database across stages. If the match is successful, the unknown obstacle is classified as a pre-completed building. The degradation gradient generation module is used to extract the construction process attributes of the prematurely completed buildings from the model database for obstacles identified as prematurely completed buildings. It then radiates outward from the boundary of the prematurely completed building in the 3D physical property grid map, and attenuates and corrects the base flatness index and base friction coefficient of each grid within the influence radius to generate a physical environment-related degradation gradient. The path reconstruction module is used to read the current total mass data and commanded driving speed of the logistics robot, calculate the maximum safe curvature limit of the logistics robot by combining the corrected friction coefficient of the target extended grid, and truncate the search branch with curvature exceeding the maximum safe curvature limit in the cost function of the heuristic path search algorithm to generate a local detour path. The multi-robot scheduling module is used to report the coordinate data of the local detour path and its own vehicle physical parameters to the cloud server after the logistics robot completes its journey along the local detour path to generate a local detour template. When the remaining logistics robots request to approach the associated area, the module obtains the current wheelbase and current total vehicle weight of the requesting vehicle to calculate the safety slack, and executes the scheduling instruction to issue the local detour template or refuse to issue it based on the safety slack.

Images