Intelligent Construction Method for Roadbed Filling Based on Unmanned Earthmoving Vehicle Clusters

Abstract

This invention provides an intelligent construction method for roadbed filling based on an unmanned earthmoving vehicle cluster. The method acquires environmental perception data, engineering design data, and unmanned earthmoving vehicle operation data of the roadbed filling area. A three-dimensional construction model is constructed through spatiotemporal fusion, generating construction status data corresponding to the work area, earthwork requirements, and construction constraints. Based on the construction status data and vehicle operation status, task decomposition, vehicle matching, work sequence generation, and collaborative path planning are performed. Unmanned loaders are controlled to perform earthmoving, transportation, and placement, while unmanned bulldozers are controlled to perform earthmoving, filling, and leveling. Vehicle positioning, environmental perception, and working device status are integrated to implement collaborative closed-loop control of the vehicle chassis and working devices. The three-dimensional construction model, filling tasks, collaborative paths, and control parameters are continuously updated based on the difference between the actual work surface and the target filling surface. This invention is applicable to deep pit filling and grid paving, improving the collaborative control accuracy, safety, and quality traceability of roadbed filling.

Description

Technical Field

[0001] This invention relates to the field of unmanned earthmoving vehicle construction, and in particular to an intelligent construction method for roadbed filling based on a cluster of unmanned earthmoving vehicles. Background Technology

[0002] Roadbed filling is a crucial step in road construction, and its quality directly impacts the roadbed's bearing capacity, stability, and durability. Traditional construction typically involves dump trucks, loaders, and bulldozers operating collaboratively under manual driving and on-site command. The construction process is heavily influenced by the driver's experience, field of vision, and on-site coordination skills, easily leading to problems such as vehicle path intersections, waiting for avoidance maneuvers, repeated driving, and discontinuous work processes. Furthermore, manual operation struggles to consistently and accurately control the placement of fill material, the filling range, the thickness of each layer, and the flatness of the work surface, easily resulting in construction quality issues such as missed filling, overfilling, and uneven filling. In deep pits, sloping terrain, and complex obstacle environments, there are also risks of vehicle collisions, rollovers, and personnel safety.

[0003] With the development of autonomous driving, environmental perception, and digital construction technologies, some construction machinery is now capable of autonomous driving, path tracking, and single-machine construction operations. However, existing solutions mostly focus on the autonomous driving or pre-set path execution of individual vehicles, lacking unified modeling for the continuously changing terrain, earthwork distribution, vehicle operating status, and construction quality during roadbed filling. This makes it difficult to perform real-time earthwork demand calculations, vehicle task matching, work sequence coordination, and construction parameter updates based on changes between the current working surface and the target filling surface. Particularly when unmanned loaders and unmanned bulldozers are used in joint construction, a complete closed-loop technical solution covering environmental perception, task decomposition, collaborative path planning, joint control of the chassis and working devices, and feedback of construction results is still lacking.

[0004] Existing technology CN110221610A discloses an intelligent operation system and method for unmanned construction machinery. It acquires operational environment data through drones and construction machinery, and a control center performs task planning and controls various unmanned construction machinery to perform navigation, obstacle avoidance, and collaborative operations. While this existing technology enables centralized management and dynamic operation control of construction machinery, it primarily targets general construction machinery operations. It does not establish a relationship between the current working surface and the target filling surface for layered roadbed filling, nor does it clearly address the task matching, operation sequence, and collaborative control of filling thickness between unmanned loaders' earthmoving, transportation, and placement and unmanned bulldozers' pushing and leveling.

[0005] Existing technology CN118238153B discloses an autonomous construction method and system for an embodied intelligent bulldozer robot. It generates work instructions based on the construction plan and site environment information, generates the robot's travel path and bulldozer blade motion plan through a motion planning model, and outputs adjustment strategies based on the robot's actual state, such as continuing work, replanning, or terminating work. While this existing technology enables autonomous planning and feedback adjustment for the bulldozer robot, it primarily targets the construction operation of a single bulldozer robot. It does not form a multi-equipment collaborative system consisting of drones, unmanned loaders, and unmanned bulldozers, nor does it address earthwork demand analysis, task-vehicle matching, multi-vehicle path occupancy coordination, or switching between deep pit filling and grid paving modes based on a dynamic three-dimensional construction model. Summary of the Invention

[0006] The main objective of this invention is to provide an intelligent construction method for roadbed filling based on an unmanned earthmoving vehicle cluster. The technical problem to be solved by this application is: how to achieve integrated unmanned construction in a dynamically changing roadbed filling environment, including construction scenario modeling, earthmoving task allocation, collaborative operation of unmanned loaders and unmanned bulldozers, closed-loop control of vehicles and working devices, and construction quality feedback updates.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an intelligent construction method for roadbed filling based on an unmanned earthmoving vehicle cluster, comprising the following steps: S1. Acquire environmental perception data, engineering design data, and unmanned earthmoving vehicle operation data of the roadbed filling area. Perform spatiotemporal fusion on the acquired data to construct a three-dimensional construction model that is dynamically updated with the construction process. Based on the three-dimensional construction model, form construction status data that characterizes the work area, earthwork demand, and construction constraints. S2, based on construction status data and the operational status of the unmanned earthmoving vehicle cluster, decomposes the roadbed filling task and matches vehicles, generating the operation sequence and collaborative path of unmanned loaders and unmanned bulldozers. S3 controls the unmanned loader and unmanned bulldozer to enter the corresponding work area according to the work sequence and collaborative path. The unmanned loader performs earthwork shoveling, transportation and material placement, and the unmanned bulldozer performs earthwork filling and leveling. S4 integrates the positioning information, environmental perception information and working device status information of unmanned earthmoving vehicles to determine the joint operation status of the vehicle and working device, and performs collaborative closed-loop control of the vehicle chassis and working device based on the joint operation status, collaborative path and target filling surface. S5 obtains the actual working surface after earthwork placement and filling, as well as the task execution results of unmanned earthwork vehicles. Based on the difference between the actual working surface and the target filling surface, it continuously updates the three-dimensional construction model, roadbed filling task, collaborative path, and closed-loop control parameters until the current filling layer meets the construction quality requirements.

[0008] In the preferred embodiment, step S1 includes: Before construction, drone photography equipment and drone lidar were used to survey the roadbed filling area to obtain initial environmental data characterizing the topography, boundaries, slope and pothole conditions of the construction area, and to connect it with the engineering design data corresponding to the roadbed cross section, layered filling, base treatment and drainage layout. An initial three-dimensional model of the roadbed filling area is established based on initial environmental data and engineering design data; During construction, the onboard environmental sensing equipment and vehicle status acquisition equipment of the unmanned earthmoving vehicles are used to continuously acquire information on changes in the work surface, earthwork distribution, obstacle status, and the operating status of the unmanned earthmoving vehicles. Data from different data sources before and after construction are sequentially unified in coordinates, aligned in time, spatially correlated, and merged and updated. The initial 3D model is then recursively corrected based on the fusion results to generate a 3D construction model that dynamically changes with the construction process.

[0009] In the preferred embodiment, construction status data is generated based on a three-dimensional construction model, including: Spatial difference calculations are performed between the current working face in the 3D construction model and the target filling face in the engineering design data to form a filling state field that reflects the distribution of earthwork gaps in the roadbed filling area. The work units are divided according to the spatial continuity of the filling site, the operational capacity of unmanned earthmoving vehicles, and the traffic conditions of the construction area, and the earthwork requirements corresponding to each work unit are determined. Based on the spatial shape and earthwork requirements of each work unit, generate the material placement benchmark, the filling benchmark, and the filling thickness control benchmark. Construction constraints are generated based on construction boundaries, foundation conditions, drainage facility protection range, obstacle distribution, and vehicle traffic conditions. By linking work units, earthwork requirements, material placement benchmarks, filling benchmarks, filling thickness control benchmarks, and construction constraints to a three-dimensional construction model, construction status data is generated for filling task generation, collaborative path planning, and construction control.

[0010] In the preferred embodiment, step S2 includes: Based on the construction status data, the roadbed filling construction process is analyzed and converted into machine-executable task units with operation location, target amount of work, operation type and construction connection relationship. A task chain is established based on the construction connection relationship. Establish vehicle capability status based on the equipment type, current location, availability, operational capabilities, and task execution progress of unmanned earthmoving vehicles; Based on the task chain and vehicle capability status, establish the matching relationship between tasks and vehicles, with the goal of reducing the comprehensive construction cost caused by vehicle idling, operation waiting, energy consumption and path conflicts, and determine the task allocation results and operation sequence of unmanned loaders and unmanned bulldozers. When construction status data, vehicle capacity status, or construction environment changes, the vehicle matching is re-performed for the task units affected by the changes, and the task allocation results and operation sequence are updated on a rolling basis.

[0011] In the preferred scheme, the collaborative path is generated using a two-layer path planning approach that combines global planning with local adjustments, including: Based on the 3D construction model and construction constraints, a passage cost field is established that can be updated according to the working surface, earthwork distribution and obstacle status. Based on the passage cost field, a global reference path is generated for each unmanned earthmoving vehicle to complete the corresponding task. Based on the expected operation process of each unmanned earthmoving vehicle along the global reference path, establish a path occupancy status with time attributes; When the path occupancy status of different unmanned earthmoving vehicles overlaps, the passage order, waiting position or driving route of the unmanned earthmoving vehicles shall be adjusted according to the construction connection relationship of the corresponding tasks and the vehicle operation status. When the vehicle-mounted environmental perception device detects a change in traffic conditions, it performs local reconstruction of the path in the affected area and connects the locally reconstructed path with the unaffected global reference path to form a dynamically updated cooperative path.

[0012] In the preferred embodiment, the combined operating state of the vehicle and the working device in step S4 is determined as follows: Acquire multi-source positioning observation information formed by real-time dynamic carrier phase differential positioning equipment, inertial measurement unit, vehicle mileage acquisition equipment and vehicle-mounted environmental perception equipment; The observation credibility is determined based on the effectiveness and stability of multi-source positioning observation information, and joint state estimation is performed within a continuous time window to obtain the continuous spatial pose of unmanned earthmoving vehicles in the three-dimensional construction model. When some multi-source positioning observation information is abnormal, reduce the impact of abnormal information on joint state estimation, and use the remaining effective information to maintain continuous output of spatial pose. Once the abnormal information is recovered, the deviation between the recovered positioning result and the current spatial pose is progressively corrected. The combined operating state of the vehicle chassis and working device is determined based on the spatial position, vehicle motion state, and the attitude and load state of the bucket or blade.

[0013] In the preferred embodiment, coordinated closed-loop control of the vehicle chassis and the working device includes: A vehicle-working device coupled prediction model is established based on the motion response of unmanned earthmoving vehicles, the action response of working devices, and the earthmoving load. The trajectory deviation of the unmanned earthmoving vehicle relative to the collaborative path is determined based on the joint operation status, and the operational deviation of the working device is determined based on the elevation difference between the actual working surface and the target filling surface. Rolling predictive control is used to generate steering, drive and braking control quantities for the vehicle chassis, and adaptive feedback control is used to generate attitude control quantities for the bucket or blade. The vehicle chassis control quantity is sent to the drive-by-wire chassis, and the working device control quantity is sent to the electro-hydraulic actuator, so that the unmanned earthmoving vehicle runs along the cooperative path and performs earthmoving, placement, pushing or leveling according to the filling thickness control benchmark. Based on the vehicle motion response, the load of the working device, and the actual changes in the working surface, the vehicle-working device coupling prediction model and the control quantity for the next control cycle are modified.

[0014] In the preferred embodiment, when the 3D construction model identifies a depressed area to be filled within the roadbed filling area, a deep pit filling operation mode is adopted. The deep pit filling operation mode includes: Based on the spatial shape of the sunken area to be filled, the earthwork gap and the surrounding traffic conditions, the unloading area is determined outside the sunken area to be filled, and the earthwork transfer and placement area is determined at the edge of the sunken area to be filled. Guide transport vehicles to unload soil in the unloading area, and control unmanned loaders to scoop up and transport the unloaded soil to the soil transfer and placement area. Based on the remaining earthwork gap in the sunken area to be filled and the filling thickness control benchmark, the unmanned bulldozer is controlled to push the earthwork in layers from the earthwork transfer and placement area into the sunken area to be filled and level it. Obtain the actual working surface after each filling, update the remaining earthwork gap in the sunken area to be filled, and adjust the subsequent material placement position, filling direction, and filling number.

[0015] In the preferred embodiment, when the 3D construction model identifies the roadbed filling area as a continuously unfolding work surface, a grid paving operation mode is adopted. The grid paving operation mode includes: Based on the spatial range of the continuously unfolding work surface, the filling thickness control benchmark, and the operational capabilities of unmanned earthmoving vehicles, multiple interconnected grid work units are divided. The earthwork requirements are determined based on the current working face and the target filling face corresponding to each grid operation unit, and the corresponding quantitative material placement task is generated. Control the unmanned loader to transport the earthwork to each grid work unit and execute quantitative material placement; control the unmanned bulldozer to perform filling and leveling in sequence according to the spatial connection relationship between the grid work units. The system obtains the actual elevation, filling thickness, and leveling status of each grid operation unit after the completion of the operation. Based on the actual filling status, it identifies underfilled or overfilled areas and generates material replenishment tasks, material transfer tasks, or re-leveling tasks. After each grid work unit meets the construction quality requirements of the current filling layer, the corresponding task execution data and construction quality data are recorded.

[0016] An intelligent construction system for roadbed filling based on an unmanned earthmoving vehicle cluster includes a dynamic three-dimensional model acquisition system for the work scene, an intelligent scheduling system, and an unmanned earthmoving vehicle cluster. The dynamic 3D model acquisition system for the work scene includes a drone acquisition device and a model processing device, which are used to acquire construction environment data of the roadbed filling area, calculate earthwork volume, and construct a dynamic 3D construction model. The intelligent scheduling system communicates with the dynamic 3D model acquisition system for the work scenario, and is used to generate construction tasks, work sequence and collaborative paths based on the dynamic 3D construction model and the operating status of the unmanned earthmoving vehicle cluster. The unmanned earthmoving vehicle cluster includes unmanned loaders and unmanned bulldozers. Both unmanned loaders and unmanned bulldozers are equipped with onboard perception, navigation and control devices to obtain the equipment operating status and construction environment status. They perform earthwork placement, filling and leveling according to the construction task, operation sequence and collaborative path, and feed back the task execution results to the intelligent scheduling system and the dynamic three-dimensional model acquisition system of the operation scene to update the dynamic three-dimensional construction model and subsequent construction control data.

[0017] This invention provides an intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters. This application integrates UAV mapping data, vehicle-mounted environmental perception data, engineering design data, and unmanned earthmoving vehicle operation data to construct a three-dimensional construction model that updates with the construction process. Based on the difference between the current working surface and the target filling surface, earthwork demand and construction constraints are formed, so that the construction tasks, working areas, and filling control benchmarks can be adjusted synchronously with the site conditions.

[0018] This application decomposes and matches filling tasks based on construction status and vehicle capacity, coordinates the shoveling, transporting and placing operations of unmanned loaders with the pushing and leveling operations of unmanned bulldozers, and reduces operational conflicts between vehicles through global planning, local adjustments and path occupation coordination, thereby improving the continuous collaborative capability of the fleet.

[0019] This application integrates vehicle positioning, environmental perception, and working device status to perform collaborative closed-loop control of the drive-by-wire chassis and bucket or blade. It also continuously corrects the three-dimensional construction model, filling task, collaborative path, and control parameters based on the difference between the actual working surface and the target filling surface. This helps to improve the control stability of the layered filling thickness and flatness, and realizes unmanned, dynamic, and traceable construction process. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a schematic diagram of the composition of the intelligent roadbed filling construction system based on an unmanned earthmoving vehicle cluster of the present invention.

[0021] Figure 2 This is a schematic diagram of the fixed-point unloading of transport vehicles during the deep pit filling process of the roadbed according to the present invention.

[0022] Figure 3 This is a schematic diagram illustrating the unmanned loader's shoveling, transporting, and placing of materials during the deep pit filling process of the roadbed according to the present invention.

[0023] Figure 4 This is a schematic diagram of the unmanned bulldozer pushing and leveling during the deep pit filling process of the roadbed according to the present invention.

[0024] Figure 5 This is a schematic diagram of the large-area roadbed grid paving operation of the present invention.

[0025] Figure 6 This is a flowchart illustrating the construction of a dynamic 3D model of the work scenario and the generation of construction status data for this invention.

[0026] Figure 7 This is a flowchart illustrating the roadbed filling task decomposition, vehicle matching, and collaborative path planning of the present invention.

[0027] Figure 8 This is a flowchart of the collaborative closed-loop control and construction feedback update process for the vehicle and working device of the present invention. Detailed Implementation

[0028] Example 1 like Figure 1-8 As shown, an intelligent construction method for roadbed filling based on an unmanned earthmoving vehicle cluster includes the following steps: S1. Acquire environmental perception data, engineering design data, and unmanned earthmoving vehicle operation data of the roadbed filling area. Perform spatiotemporal fusion on the acquired data to construct a three-dimensional construction model that is dynamically updated with the construction process. Based on the three-dimensional construction model, form construction status data that characterizes the work area, earthwork demand, and construction constraints. S2, based on construction status data and the operational status of the unmanned earthmoving vehicle cluster, decomposes the roadbed filling task and matches vehicles, generating the operation sequence and collaborative path of unmanned loaders and unmanned bulldozers. S3 controls the unmanned loader and unmanned bulldozer to enter the corresponding work area according to the work sequence and collaborative path. The unmanned loader performs earthwork shoveling, transportation and material placement, and the unmanned bulldozer performs earthwork filling and leveling. S4 integrates the positioning information, environmental perception information and working device status information of unmanned earthmoving vehicles to determine the joint operation status of the vehicle and working device, and performs collaborative closed-loop control of the vehicle chassis and working device based on the joint operation status, collaborative path and target filling surface. S5 obtains the actual working surface after earthwork placement and filling, as well as the task execution results of unmanned earthwork vehicles. Based on the difference between the actual working surface and the target filling surface, it continuously updates the three-dimensional construction model, roadbed filling task, collaborative path, and closed-loop control parameters until the current filling layer meets the construction quality requirements.

[0029] In the preferred embodiment, step S1 includes: Before construction, drone photography equipment and drone lidar were used to survey the roadbed filling area to obtain initial environmental data characterizing the topography, boundaries, slope and pothole conditions of the construction area, and to connect it with the engineering design data corresponding to the roadbed cross section, layered filling, base treatment and drainage layout. An initial three-dimensional model of the roadbed filling area is established based on initial environmental data and engineering design data; During construction, the onboard environmental sensing equipment and vehicle status acquisition equipment of the unmanned earthmoving vehicles are used to continuously acquire information on changes in the work surface, earthwork distribution, obstacle status, and the operating status of the unmanned earthmoving vehicles. Data from different data sources before and after construction are sequentially unified in coordinates, aligned in time, spatially correlated, and merged and updated. The initial 3D model is then recursively corrected based on the fusion results to generate a 3D construction model that dynamically changes with the construction process.

[0030] In the preferred embodiment, construction status data is generated based on a three-dimensional construction model, including: Spatial difference calculations are performed between the current working face in the 3D construction model and the target filling face in the engineering design data to form a filling state field that reflects the distribution of earthwork gaps in the roadbed filling area. The work units are divided according to the spatial continuity of the filling site, the operational capacity of unmanned earthmoving vehicles, and the traffic conditions of the construction area, and the earthwork requirements corresponding to each work unit are determined. Based on the spatial shape and earthwork requirements of each work unit, generate the material placement benchmark, the filling benchmark, and the filling thickness control benchmark. Construction constraints are generated based on construction boundaries, foundation conditions, drainage facility protection range, obstacle distribution, and vehicle traffic conditions. By linking work units, earthwork requirements, material placement benchmarks, filling benchmarks, filling thickness control benchmarks, and construction constraints to a three-dimensional construction model, construction status data is generated for filling task generation, collaborative path planning, and construction control.

[0031] In the preferred embodiment, step S2 includes: Based on the construction status data, the roadbed filling construction process is analyzed and converted into machine-executable task units with operation location, target amount of work, operation type and construction connection relationship. A task chain is established based on the construction connection relationship. Establish vehicle capability status based on the equipment type, current location, availability, operational capabilities, and task execution progress of unmanned earthmoving vehicles; Based on the task chain and vehicle capability status, establish the matching relationship between tasks and vehicles, with the goal of reducing the comprehensive construction cost caused by vehicle idling, operation waiting, energy consumption and path conflicts, and determine the task allocation results and operation sequence of unmanned loaders and unmanned bulldozers. When construction status data, vehicle capacity status, or construction environment changes, the vehicle matching is re-performed for the task units affected by the changes, and the task allocation results and operation sequence are updated on a rolling basis.

[0032] In the preferred scheme, the collaborative path is generated using a two-layer path planning approach that combines global planning with local adjustments, including: Based on the 3D construction model and construction constraints, a passage cost field is established that can be updated according to the working surface, earthwork distribution and obstacle status. Based on the passage cost field, a global reference path is generated for each unmanned earthmoving vehicle to complete the corresponding task. Based on the expected operation process of each unmanned earthmoving vehicle along the global reference path, establish a path occupancy status with time attributes; When the path occupancy status of different unmanned earthmoving vehicles overlaps, the passage order, waiting position or driving route of the unmanned earthmoving vehicles shall be adjusted according to the construction connection relationship of the corresponding tasks and the vehicle operation status. When the vehicle-mounted environmental perception device detects a change in traffic conditions, it performs local reconstruction of the path in the affected area and connects the locally reconstructed path with the unaffected global reference path to form a dynamically updated cooperative path.

[0033] In the preferred embodiment, the combined operating state of the vehicle and the working device in step S4 is determined as follows: Acquire multi-source positioning observation information formed by real-time dynamic carrier phase differential positioning equipment, inertial measurement unit, vehicle mileage acquisition equipment and vehicle-mounted environmental perception equipment; The observation credibility is determined based on the effectiveness and stability of multi-source positioning observation information, and joint state estimation is performed within a continuous time window to obtain the continuous spatial pose of unmanned earthmoving vehicles in the three-dimensional construction model. When some multi-source positioning observation information is abnormal, reduce the impact of abnormal information on joint state estimation, and use the remaining effective information to maintain continuous output of spatial pose. Once the abnormal information is recovered, the deviation between the recovered positioning result and the current spatial pose is progressively corrected. The combined operating state of the vehicle chassis and working device is determined based on the spatial position, vehicle motion state, and the attitude and load state of the bucket or blade.

[0034] In the preferred embodiment, coordinated closed-loop control of the vehicle chassis and the working device includes: A vehicle-working device coupled prediction model is established based on the motion response of unmanned earthmoving vehicles, the action response of working devices, and the earthmoving load. The trajectory deviation of the unmanned earthmoving vehicle relative to the collaborative path is determined based on the joint operation status, and the operational deviation of the working device is determined based on the elevation difference between the actual working surface and the target filling surface. Rolling predictive control is used to generate steering, drive and braking control quantities for the vehicle chassis, and adaptive feedback control is used to generate attitude control quantities for the bucket or blade. The vehicle chassis control quantity is sent to the drive-by-wire chassis, and the working device control quantity is sent to the electro-hydraulic actuator, so that the unmanned earthmoving vehicle runs along the cooperative path and performs earthmoving, placement, pushing or leveling according to the filling thickness control benchmark. Based on the vehicle motion response, the load of the working device, and the actual changes in the working surface, the vehicle-working device coupling prediction model and the control quantity for the next control cycle are modified.

[0035] In the preferred embodiment, when the 3D construction model identifies a depressed area to be filled within the roadbed filling area, a deep pit filling operation mode is adopted. The deep pit filling operation mode includes: Based on the spatial shape of the sunken area to be filled, the earthwork gap and the surrounding traffic conditions, the unloading area is determined outside the sunken area to be filled, and the earthwork transfer and placement area is determined at the edge of the sunken area to be filled. Guide transport vehicles to unload soil in the unloading area, and control unmanned loaders to scoop up and transport the unloaded soil to the soil transfer and placement area. Based on the remaining earthwork gap in the sunken area to be filled and the filling thickness control benchmark, the unmanned bulldozer is controlled to push the earthwork in layers from the earthwork transfer and placement area into the sunken area to be filled and level it. Obtain the actual working surface after each filling, update the remaining earthwork gap in the sunken area to be filled, and adjust the subsequent material placement position, filling direction, and filling number.

[0036] In the preferred embodiment, when the 3D construction model identifies the roadbed filling area as a continuously unfolding work surface, a grid paving operation mode is adopted. The grid paving operation mode includes: Based on the spatial range of the continuously unfolding work surface, the filling thickness control benchmark, and the operational capabilities of unmanned earthmoving vehicles, multiple interconnected grid work units are divided. The earthwork requirements are determined based on the current working face and the target filling face corresponding to each grid operation unit, and the corresponding quantitative material placement task is generated. Control the unmanned loader to transport the earthwork to each grid work unit and execute quantitative material placement; control the unmanned bulldozer to perform filling and leveling in sequence according to the spatial connection relationship between the grid work units. The system obtains the actual elevation, filling thickness, and leveling status of each grid operation unit after the completion of the operation. Based on the actual filling status, it identifies underfilled or overfilled areas and generates material replenishment tasks, material transfer tasks, or re-leveling tasks. After each grid work unit meets the construction quality requirements of the current filling layer, the corresponding task execution data and construction quality data are recorded.

[0037] An intelligent construction system for roadbed filling based on an unmanned earthmoving vehicle cluster includes a dynamic three-dimensional model acquisition system for the work scene, an intelligent scheduling system, and an unmanned earthmoving vehicle cluster. The dynamic 3D model acquisition system for the work scene includes a drone acquisition device and a model processing device, which are used to acquire construction environment data of the roadbed filling area, calculate earthwork volume, and construct a dynamic 3D construction model. The intelligent scheduling system communicates with the dynamic 3D model acquisition system for the work scenario, and is used to generate construction tasks, work sequence and collaborative paths based on the dynamic 3D construction model and the operating status of the unmanned earthmoving vehicle cluster. The unmanned earthmoving vehicle cluster includes unmanned loaders and unmanned bulldozers. Both unmanned loaders and unmanned bulldozers are equipped with onboard perception, navigation and control devices to obtain the equipment operating status and construction environment status. They perform earthwork placement, filling and leveling according to the construction task, operation sequence and collaborative path, and feed back the task execution results to the intelligent scheduling system and the dynamic three-dimensional model acquisition system of the operation scene to update the dynamic three-dimensional construction model and subsequent construction control data.

[0038] Example 2 Further explanation in conjunction with Example 1, such as Figure 1-8 As shown, the entire text adopts a consistent style. Indicates time Environmental perception data, using Representing engineering design data, using Indicates time The operation data of unmanned earthmoving vehicles, using Indicates time The three-dimensional construction model is adopted. Indicates time The construction status data, with the same Chinese meaning, will be represented by the same symbols throughout the text.

[0039] The intelligent construction method for roadbed filling is implemented sequentially, including construction data acquisition, task planning, vehicle execution, closed-loop control, and construction feedback. Environmental perception data is acquired before and during construction. Engineering design data and unmanned earthmoving vehicle operation data A three-dimensional construction model is generated through spatiotemporal fusion processing, and its processing relationship can be expressed as follows: ; in, This represents the spatiotemporal fusion processing procedure, used to unify coordinates between different data sources, align acquisition times, establish spatial relationships, and update status. Construction status data can be represented as follows: ; in, This describes the process of extracting the work area, earthwork requirements, and construction constraints based on the 3D construction model and engineering design data. The intelligent scheduling system... and The roadbed filling task is decomposed, and different tasks are matched to unmanned loaders and unmanned bulldozers respectively, generating operation sequences and collaborative paths. Unmanned loaders perform earthmoving, transportation, and placement, while unmanned bulldozers perform earthmoving, filling, and leveling. During construction, vehicle position, vehicle attitude, working device attitude, and actual working surface are continuously acquired, and the control parameters of the vehicle chassis and working devices are adjusted accordingly. After completing one round of placement or filling, the actual working surface is compared with the target filling surface, and the earthwork demand, task allocation, collaborative path, and control parameters are recalculated based on the comparison results, thus forming a dynamic construction closed loop.

[0040] Before construction, continuous images of the construction area are acquired using drone photography equipment, and 3D point clouds of the construction area are obtained using drone LiDAR. The image data and point cloud data are then correlated to the engineering coordinate system using drone positioning information. Spatial points in any sensor coordinate system Spatial points after transformation to engineering coordinate system It can be represented as ; in, Represents a spatial point in the data source coordinate system. Represents a spatial point in the engineering coordinate system. This represents the direction transformation matrix from the data source coordinate system to the engineering coordinate system. This represents the corresponding spatial translation. After coordinate unification, the UAV data is overlaid with the engineering design data corresponding to the roadbed cross-section, layered filling, base treatment, and drainage layout to establish an initial 3D model. During construction, vehicle-mounted environmental sensing equipment acquires new observation elevations of the work surface. and in accordance with Update the current work surface. In the formula, Representing a spatial point At any moment The current working surface elevation, This indicates that the vehicle-mounted environmental perception equipment is at all times The obtained observation elevation, This indicates the elevation of the work surface at the last update time. This represents the fusion weight of the current observation data. When the observation data is stable and valid, the weight is increased. When observation data is affected by obstruction, dust, or vehicle vibration, the data loss is reduced. Through continuous iterative updates, the 3D construction model can reflect real-time changes in terrain, earthwork distribution, obstacles, and construction progress.

[0041] Spatial difference calculations are performed between the current working surface in the 3D construction model and the target filling surface in the engineering design data. The elevation of the target filling surface is determined using... It is indicated that the current working face elevation adopts... Indicates a spatial point The difference in filling elevation at the location is expressed as ; when A value greater than zero indicates that there is earthwork to be filled at that location; when When the value is less than zero, it indicates that the location is higher than the target fill surface; when When the deviation is within the allowable range, it indicates that the location meets the requirements of the current fill layer. This is based on the total spatial locations. Together, they form the filling state field. For any work unit The corresponding earthwork demand can be expressed as ; In the formula, Indicates work unit At any moment The earthwork demand, This represents the loose-lay correction factor determined based on the loose state of the filling material. This represents a small area unit of the working surface. Working units are divided based on the spatial continuity of the filling state field, vehicle turning range, vehicle passable area, and the working range of the working equipment. The material placement datum is used to determine the unloading position and single material placement range of the unmanned loader; the filling datum is used to determine the filling direction and termination position of the unmanned bulldozer; and the filling thickness control datum is used to limit the target loose-lay thickness of the current filling layer. Construction boundaries, weak foundation areas, drainage facility protection areas, obstacle areas, and impassable areas are all associated with the 3D construction model as construction constraints.

[0042] Construction status data is first converted into task units that can be performed by unmanned earthmoving vehicles. Each task unit has at least the work location, target quantity, work type, and construction connection relationship. The task units of unmanned loaders mainly include shoveling, transporting, and placing tasks, while the task units of unmanned bulldozers mainly include filling and leveling tasks. Task chains are established between tasks according to the construction sequence. Must precede the task Upon completion, the corresponding job sequence satisfies ; in, Indicates task At the beginning of Indicates task The completion time. Task unit. With unmanned earthmoving vehicles The overall construction cost between them can be expressed as: ; In the formula, Representing task unit Assigned to unmanned earthmoving vehicles The overall construction cost, This represents the cost corresponding to the empty driving distance. This indicates the cost of waiting for a task. This indicates the cost corresponding to energy consumption. This represents the cost corresponding to a path conflict. , , and These represent the weights of the corresponding costs. Task matching can be represented as... ; in, Represents task allocation variables; when task unit Assigned to unmanned earthmoving vehicles hour, When not allocated, Each task unit is assigned only to an unmanned earthmoving vehicle that meets the equipment type and operational capacity requirements. When a vehicle malfunctions, runs out of power, changes location, or the task schedule changes, only the matching relationship of the affected task units is recalculated; there is no need to regenerate the entire construction task.

[0043] The collaborative path is generated using a combination of global planning and local adjustments. A passage cost field is constructed based on the 3D construction model. The calculation relationship can be expressed as follows: ; In the formula, This indicates the cost of passage caused by terrain slope and undulation of the work surface. This indicates the cost of passage resulting from the distribution of earthwork and the bearing capacity of the ground. This represents the cost of passage due to the formation of an obstacle. This indicates the cost of access created by the construction boundary and the protected area. , , and These represent the weights of the corresponding passage costs. Unmanned earthmoving vehicles. The global reference path can be represented as: ; in, Indicates no unmanned earthmoving vehicles Candidate paths, This represents the global reference path with the minimum overall travel cost. This represents a minute path length. To avoid multi-vehicle conflicts, a path occupancy status with time attributes is established for each path segment. ,in, Indicates a path segment. This indicates the expected time of occupancy. When two unmanned earthmoving vehicles are both occupied on the same path segment and within the same time frame, a path conflict is determined, and the passage order, waiting position, or travel path is adjusted based on task coordination, vehicle loading status, and vehicle braking distance. When the onboard environmental perception equipment detects new obstacles or terrain changes, only the affected path segment is partially reconstructed, and the partially reconstructed path is smoothly connected to the unaffected global reference path.

[0044] Multi-source positioning and observation information is provided jointly by real-time dynamic carrier phase differential positioning equipment, inertial measurement units, vehicle mileage acquisition equipment, and onboard environmental sensing equipment. Unmanned earthmoving vehicles. The state adopts express, This includes the vehicle's position, attitude, and motion state in the engineering coordinate system. The observation information of positioning devices adopts This indicates that the corresponding observation relationship is adopted. The joint state estimate can be expressed as: ; In the formula, This represents the continuous spatial pose of the vehicle obtained from the joint state estimation. This represents the set of valid location observation information at the current moment. Indicates the first The credibility weight of location observation information This represents the constraint matrix corresponding to the observation error. The confidence weights can be expressed as... ; in, This represents the reliability score obtained based on the positioning solution status, observation continuity, and environmental matching degree. When a positioning observation is abnormal, its reliability score decreases, and the corresponding confidence weight decreases accordingly, preventing the abnormal information from dominating the joint state estimation. Once the abnormal information is recovered, [the following is used]. ; A gradual correction is made. In the formula, This indicates the vehicle's spatial orientation before the correction. This indicates the restored location result. This indicates the progressively corrected spatial orientation of the vehicle. This represents the asymptotic correction factor. By limiting... The rate of change can prevent sudden jumps in vehicle spatial pose and control commands during positioning recovery. The vehicle spatial pose, together with the attitude of the bucket or blade and the load status, constitute the joint operating state of the vehicle and the working device.

[0045] The vehicle-working device coupled prediction model is used to describe the interaction between vehicle movement, working device operation, and earthmoving load. Its discrete state relationship can be expressed as follows: ; In the formula, Indicates no unmanned earthmoving vehicles During the control cycle The coupled operating state, Indicates control cycle The vehicle chassis control quantity and the working device control quantity. This represents the operational disturbance caused by earthwork resistance, ground adhesion status, and the load on the working equipment. This represents the vehicle-working device coupling state prediction process. The working device control variables of the unmanned loader are used to adjust the bucket lifting and tilting states, while those of the unmanned bulldozer are used to adjust the blade lifting and tilting states. The objective function of the rolling predictive control can be expressed as: In the formula, Indicates no unmanned earthmoving vehicles The control target value is within the prediction range. This indicates the number of control cycles covered by the predicted control. This indicates the trajectory deviation of the vehicle's actual operating state relative to the cooperative path. This indicates the operational deviation of the actual working surface relative to the target filling surface. This represents the control increment between adjacent control cycles. Represents the trajectory deviation weight matrix. Represents the job deviation weight matrix. This represents the control increment weight matrix. It decreases by rolling. This generates steering, drive, and braking control variables for the vehicle chassis, as well as attitude control variables for the bucket or blade. After executing the first set of control variables in each control cycle, the vehicle motion response, working device load, and actual working surface changes are reacquired, and the prediction results for the next control cycle are updated accordingly.

[0046] When the 3D construction model identifies the recessed area to be filled, this area is designated as the deep pit filling area. The remaining earthwork requirement for the deep pit filling area is determined according to the aforementioned unified earthwork requirement calculation relationship, i.e. ; In the formula, Indicates the time of the deep pit filling area The remaining earthwork demand, This indicates the deep pit filling area. Based on the deep pit boundary, surrounding slope, and vehicle access conditions, an unloading area for transport vehicles is determined outside the deep pit, and a soil transfer and placement area is determined at a location where unmanned loaders can safely reach and unmanned bulldozers can continuously push and fill. Unmanned loaders transport the unloaded soil in batches to the soil transfer and placement area, while unmanned bulldozers push the soil in layers into the deep pit according to the filling thickness control benchmark. The amount of soil that can be handled in a single filling operation is... The expected number of submissions can be expressed as: ; in, Indicates the expected number of submissions. This indicates rounding up. The actual working surface is re-acquired and recalculated after each backfilling operation. This allows for dynamic adjustment of the subsequent placement of materials, the direction of filling, and the number of filling operations, avoiding omissions or overfills caused by continuous operation based on the initial earthwork volume.

[0047] The continuous deployment surface is divided into multiple grid operation units based on the spatial range, fill thickness control benchmarks, and unmanned earthmoving vehicle operation capabilities. The earthwork demand for each grid work unit will still be calculated using a unified formula. ; Calculation. The effective material placement capacity of an unmanned loader in a single operation is calculated using... This indicates that the grid operation unit The number of times the fabric loading needs to be performed can be expressed as: ; In the formula, Represents grid operation unit The corresponding number of loading operations. The unmanned loader sequentially places materials according to the earthwork requirements of each grid work unit, while the unmanned bulldozer continuously performs pushing, filling, and leveling according to the spatial connection between the grids. The average filling elevation difference of each grid work unit can be expressed as... ; In the formula, Represents grid operation unit Average filling elevation difference Represents grid operation unit The area. When When the corresponding grid is identified as an underfilled area, a replenishment task is generated; when When the corresponding grid is identified as an overfilled area, a material transfer task is generated; when When the flatness meets the requirements, the corresponding grid will be identified as a qualified area. This indicates the average allowable elevation deviation of the current fill layer. After each grid is completed, the task execution process, vehicle trajectory, material placement quantity, actual elevation, and fill thickness are recorded, enabling the construction quality to be traced according to the grid operation unit.

[0048] The dynamic 3D model acquisition system for the work scenario, the intelligent scheduling system, and the unmanned earthmoving vehicle cluster are connected through a construction communication network. The dynamic 3D model acquisition system uses unmanned aerial vehicles (UAVs) to acquire construction environment data and uses a model processing device to calculate earthwork volume and construct a 3D construction model. The intelligent scheduling system receives the 3D construction model. and unmanned earthmoving vehicle operation data The system generates construction tasks, work sequences, and collaborative paths. The unmanned earthmoving vehicle cluster includes unmanned loaders and unmanned bulldozers. Each vehicle is equipped with onboard perception, navigation, and control devices to acquire equipment operating status and construction environment conditions, and to control the vehicle chassis and working devices to execute corresponding construction tasks. Task execution results are displayed using... The feedback update relationship of the 3D construction model can be represented as follows: ; In the formula, This represents the time interval between adjacent model update times. This describes the process of updating the 3D construction model based on the task execution results. Subsequent construction plans will then be adopted. This indicates that its update relationship can be represented as ; In the formula, This indicates the updated construction tasks, work sequence, and collaboration paths. This represents the task generation and path planning process executed by the intelligent scheduling system. Therefore, the dynamic 3D model acquisition system for the work scenario is responsible for construction environment perception and model updating, the intelligent scheduling system is responsible for construction decision-making, and the unmanned earthmoving vehicle cluster is responsible for construction execution and result feedback. Together, these three components form a complete technical system corresponding to the aforementioned methodological steps.

[0049] Example 3 Further explanation is provided in conjunction with Examples 1-2, such as Figure 1-8 As shown, this embodiment takes a section of a highway embankment construction project as the implementation object. This construction section includes both a continuously unfolded working face and a locally sunken area to be filled. The entire text adopts... Indicates time Environmental perception data, using Representing engineering design data, using Indicates time The operation data of unmanned earthmoving vehicles, using Indicates time The three-dimensional construction model is adopted. Indicates time The construction status data, and the above symbols maintain the same meaning throughout this embodiment.

[0050] like Figure 1 As shown, the intelligent construction system for roadbed filling based on an unmanned earthmoving vehicle cluster includes a dynamic 3D model acquisition system for the work scene, an intelligent scheduling system, and the unmanned earthmoving vehicle cluster. The dynamic 3D model acquisition system for the work scene includes a drone acquisition device and a model processing device. The drone acquisition device is equipped with drone photography equipment and drone LiDAR. The intelligent scheduling system is communicatively connected to the dynamic 3D model acquisition system for the work scene. The unmanned earthmoving vehicle cluster includes unmanned loaders and unmanned bulldozers. Both unmanned loaders and unmanned bulldozers are equipped with onboard environmental perception equipment, real-time dynamic carrier phase differential positioning equipment, inertial measurement units, vehicle mileage acquisition equipment, onboard controllers, drive-by-wire chassis, and working device actuators. The working device actuators of the unmanned loaders control the bucket, and the working device actuators of the unmanned bulldozers control the blade.

[0051] Before construction begins, in accordance with Figure 6 The process shown utilizes UAV photography equipment and UAV LiDAR to survey the roadbed filling area. The UAV photography equipment acquires continuous images of the construction area, while the UAV LiDAR acquires a 3D point cloud of the construction area. The UAV acquisition device simultaneously records the UAV's position and attitude, which is used to convert the continuous images and 3D point cloud to an engineering-independent coordinate system.

[0052] The relationship between spatial points in the data source coordinate system and the engineering independent coordinate system is as follows: ; in, Represents a spatial point in the data source coordinate system. Represents a spatial point in the engineering independent coordinate system. This represents the direction transformation matrix from the data source coordinate system to the engineering-independent coordinate system. This represents the spatial translation from the data source coordinate system to the engineering independent coordinate system. The model processing device uses the same coordinate transformation relationship to process UAV imagery, UAV point clouds, and vehicle-mounted environmental perception data, enabling data collected by different devices to be mapped to the same construction location.

[0053] Engineering design data This includes roadbed cross-section data, layered filling requirements, subgrade treatment requirements, and drainage layout data. Roadbed cross-section data is used to determine the roadbed filling boundary and target filling surface; layered filling requirements are used to determine the target loose-lay thickness of the current filling layer; subgrade treatment requirements are used to determine the weak subgrade areas requiring pretreatment; and drainage layout data is used to determine the location and protection range of drainage facilities. The model processing device combines the initial environmental data obtained from UAV mapping with the engineering design data. By establishing the association, an initial three-dimensional model of the roadbed filling area is created.

[0054] During construction, unmanned loaders and unmanned bulldozers continuously acquire information about changes in the work surface, soil distribution, and obstacle status through onboard environmental sensing equipment. They also continuously acquire vehicle position, attitude, and motion status through real-time dynamic carrier phase differential positioning equipment, inertial measurement units, and vehicle odometers. Furthermore, they continuously acquire the attitude and load status of the bucket or blade through sensors on the working device. This data forms the environmental sensing data. and unmanned earthmoving vehicle operation data .

[0055] Environmental sensing data Engineering design data and unmanned earthmoving vehicle operation data After coordinate unification, time alignment, spatial correlation, and fusion updates, a 3D construction model is generated. The processing relationship is expressed as follows: ; in, This represents the spatiotemporal fusion processing procedure. The spatiotemporal fusion processing procedure is used to map different data sources to the same engineering spatial location and the same construction time, and to update the current work surface, earthwork distribution, obstacle location, vehicle location, and construction progress based on the valid status of each data source.

[0056] For spatial points in the engineering independent coordinate system Vehicle-mounted environmental perception equipment is constantly The obtained observation elevations are obtained using This indicates that the three-dimensional construction model is in constant motion. The recorded current working surface elevation is adopted This indicates that the current working face elevation is updated recursively according to the following relationship: ; in, This indicates the current working surface elevation recorded at the last update time. Representing a spatial point At any moment The corresponding fusion weights. Fusion weights The value of is greater than or equal to zero and less than or equal to one. When the output of the onboard environmental perception device is stable and the corresponding spatial area is unobstructed, the fusion weight is increased. When the vehicle-mounted environmental sensing equipment is affected by construction dust, vehicle vibration, or temporary obstruction, the fusion weight should be reduced. This prevents abnormal observation data from directly replacing valid data in the 3D construction model.

[0057] The model processing device performs spatial difference calculations between the current working surface in the 3D construction model and the target filling surface in the engineering design data. The elevation of the target filling surface is determined using... It is indicated that the current working face elevation adopts... Indicates a spatial point At any moment The difference in filling elevation is expressed as:

[0058] in, This indicates the difference in filling elevation between the current working face and the target filling face. When When the value is greater than zero, it represents a spatial point. There is a gap in the earthwork at the corresponding location. When When less than zero, it represents a spatial point. The corresponding location is higher than the target fill surface. When When the elevation deviation is within the allowable range for the current fill layer, it indicates that the location meets the construction requirements for the current fill layer. This is determined by the fill elevation difference across all spatial locations. Together they form the filling state field.

[0059] The model processing device divides the roadbed filling area into multiple work units based on the spatial continuity of the filling state field, the operational capacity of unmanned earthmoving vehicles, and the traffic conditions of the construction area. Each work unit employs... express, Indicates the work unit number. Work Unit At any moment The corresponding earthwork demand is expressed as follows: ; in, Indicates work unit At any moment The earthwork demand, This represents the loose-lay correction factor for the filling material. This represents a tiny area unit on the current working surface. Used to retain the positive elevation difference of the location to be filled and to exclude the negative elevation difference that is higher than the target filling surface.

[0060] The intelligent scheduling system is based on each work unit Spatial form and earthwork demand Generates placement benchmarks, filling benchmarks, and fill thickness control benchmarks. The placement benchmark determines the placement position, range, and sequence of the unmanned loader. The filling benchmark determines the filling start point, direction, and end point of the unmanned bulldozer. The fill thickness control benchmark defines the target loose thickness of the current fill layer and serves as a common benchmark for controlling the placement amount of the unmanned loader and the blade height of the unmanned bulldozer.

[0061] The intelligent dispatching system generates construction constraints based on construction boundaries, foundation conditions, drainage facility protection zones, obstacle distribution, and vehicle traffic conditions. Construction boundaries limit the operating range of unmanned earthmoving vehicles; foundation conditions restrict unmanned earthmoving vehicles from entering areas that do not meet load-bearing requirements; drainage facility protection zones prevent vehicle chassis or working devices from encroaching on drainage facility locations; obstacle distribution creates dynamic obstacle avoidance zones; and vehicle traffic conditions limit vehicle turning, climbing, passing, and reversing processes.

[0062] Construction status data Formed according to the following relationship:

[0063] in, Indicates based on 3D construction model and engineering design data The process of generating construction status data. Construction status data. This includes work units, earthwork requirements, material placement benchmarks, filling benchmarks, filling thickness control benchmarks, and construction constraints.

[0064] according to Figure 7 As shown in the process, the intelligent scheduling system reads construction status data. and unmanned earthmoving vehicle operation data The roadbed filling construction process is analyzed, converting the overall construction process into multiple machine-executable task units. Each machine-executable task unit has a working location, target workload, task type, and construction connection relationship. The machine-executable task units corresponding to the unmanned loader include earthmoving, earthmoving, and earthmoving placement tasks. The machine-executable task units corresponding to the unmanned bulldozer include earthmoving and filling tasks and work surface leveling tasks.

[0065] The machine-executable task units are linked together to form a task chain. For tasks that need to be executed sequentially... and tasks When the task Must be in the task When they were previously completed, the two satisfied the following temporal relationship: ; in, Indicates task At the beginning of Indicates task The completion time. This time relationship is used to ensure that after the unmanned loader completes the placement of material in the designated work unit, the unmanned bulldozer enters the corresponding work unit to perform pushing and leveling.

[0066] The intelligent dispatching system establishes vehicle capability status based on the equipment type, current location, availability, operational capacity, and task execution progress of unmanned earthmoving vehicles. (Task Unit) Assigned to unmanned earthmoving vehicles The total construction cost incurred at that time is expressed as follows: ; in, Representing task unit Assigned to unmanned earthmoving vehicles The overall construction cost, This represents the cost corresponding to the vehicle's empty driving distance. This indicates the cost of waiting for vehicle operations. This indicates the cost corresponding to the vehicle's energy consumption. This represents the cost associated with a vehicle route conflict. The weight representing the cost of vehicle idling distance. The weight representing the cost of vehicle operation waiting. This indicates the weight of the cost of vehicle energy consumption. The weight represents the cost of vehicle route conflicts.

[0067] The matching relationship between tasks and vehicles is determined according to the following objectives: ; in, This represents the task allocation variable. When the task unit... Assigned to unmanned earthmoving vehicles hour, When the task unit Unassigned to unmanned earthmoving vehicles hour, The intelligent scheduling system reduces the overall construction cost of all tasks by satisfying the conditions of equipment type, vehicle availability, vehicle operation capacity, and task chain time relationship, thereby determining the task allocation results and operation sequence of unmanned loaders and unmanned bulldozers.

[0068] When construction status data When changes occur, such as changes in vehicle capability status or changes in the construction environment, the intelligent scheduling system identifies the machine-executable task units affected by the changes. It only re-matches vehicles for the affected task units and continuously updates the task allocation results and work sequence. Task units unaffected by the changes continue to execute according to the original task chain.

[0069] The intelligent scheduling system is based on the three-dimensional construction model. Establish a passage cost field based on construction constraints. Spatial point. At any moment The corresponding passage cost is expressed as: ; in, Representing a spatial point At any moment The overall cost of passage, This indicates the cost of passage caused by terrain slope and undulation of the work surface. This indicates the cost of passage resulting from the distribution of earthwork and the bearing capacity of the ground. This represents the cost of passage due to the formation of an obstacle. This indicates the cost of access created by the construction boundary and the protected area. The weights representing the terrain access costs The weights representing earthwork distribution and surface bearing capacity for traffic costs. The weights represent the cost of passing through obstacles. This indicates the weight of the access cost to the construction boundary and protected area.

[0070] Unmanned dump trucks The global reference path is generated according to the following relationship: ; in, Indicates no unmanned earthmoving vehicles Candidate paths, Indicates at time The global reference path with the lowest overall travel cost. This represents the minute path length on the candidate path. The generated global reference path satisfies the construction boundary, drainage facility protection range, minimum turning conditions for vehicles, and allowable gradient requirements for vehicles.

[0071] The intelligent scheduling system establishes a time-based path occupancy status for each unmanned earthmoving vehicle's global reference path. .in, Indicates a path segment. This indicates the route section that unmanned dump trucks are expected to occupy. At that moment, Indicates no unmanned earthmoving vehicles At any moment For path segments The state of being occupied.

[0072] When different unmanned earthmoving vehicles occupy the same path segment within the same time frame, the intelligent scheduling system determines that the path occupancy status overlaps and adjusts the vehicle passage order, waiting position, or travel route based on the task chain, vehicle loading status, and vehicle operation status. When the onboard environmental perception equipment detects new obstacles, sudden changes in the work surface, or the original path no longer meeting the vehicle passage conditions, the intelligent scheduling system retains the unaffected global reference path, performs local reconstruction of the path in the affected area, and connects the locally reconstructed path with the unaffected global reference path to form a dynamically updated collaborative path.

[0073] for Figures 2 to 4 The localized depression to be filled, as shown, is addressed by the intelligent scheduling system using a deep pit filling operation mode. The 3D construction model represents the depression to be filled as... The concave area to be filled at time The remaining earthwork requirement is expressed as follows: ; in, Indicates the concave region to be filled at time [time]. The remaining earthwork demand, This indicates the recessed area to be filled. This represents the loose-lay correction factor for the filling material. Representing a spatial point The corresponding difference in filling elevation.

[0074] The intelligent scheduling system determines the unloading area outside the sunken area, the remaining earthwork demand, and the surrounding traffic conditions based on the spatial shape of the sunken area to be filled. It also determines the earthwork transfer and placement area at the edge of the sunken area. For example... Figure 2 As shown, the transport vehicles travel to the unloading area according to the dispatch instructions and complete the unloading of the excavated soil. Figure 3As shown, the unmanned loader travels along the cooperative path to the unloading area, identifies the location of the soil pile through onboard environmental sensing equipment, controls the bucket to scoop up the soil, and transfers the soil to the soil transfer and placement area. Figure 4 As shown, the unmanned bulldozer enters the earthwork transfer and placement area along the collaborative path, pushes the earthwork in layers according to the filling thickness control benchmark into the recessed area to be filled, and then levels it.

[0075] The volume of earthwork that an unmanned bulldozer can handle in a single pass is: The estimated number of push operations required in the current state is expressed as follows: ; in, Indicates the expected number of submissions. This indicates the volume of earthwork that an unmanned bulldozer can handle in a single bulldozing operation. This indicates rounding up. After each backfilling operation, the onboard environmental sensing equipment re-acquires the actual working surface, and the model processing device recalculates the remaining earthwork requirement for the depressed area to be filled. The intelligent scheduling system adjusts the subsequent fabric placement position, filling direction, and filling frequency accordingly.

[0076] for Figure 5 The continuously deployable work surface shown employs a grid-based paving operation mode under the intelligent scheduling system. The continuously deployable work surface is divided into multiple interconnected grid work units based on spatial extent, fill thickness control benchmarks, and the operational capabilities of unmanned earthmoving vehicles. Each grid work unit The earthwork demand is all based on express.

[0077] The effective material placement capacity of the unmanned loader in a single operation is adopted. This indicates that the grid operation unit The number of times the cloth needs to be processed is expressed as follows: ; in, Represents grid operation unit The corresponding number of times the fabric is applied, This indicates the effective material placement volume of the unmanned loader in a single operation. The unmanned loader places materials quantitatively based on the earthwork requirements of each grid work unit, while the unmanned bulldozer performs filling and leveling sequentially according to the spatial connection relationship between the grid work units.

[0078] Grid work unit The average difference in filling elevation is expressed as: ; in, Represents grid operation unit Average filling elevation difference Represents grid operation unit area, Representing a spatial point The difference in filling elevation.

[0079] when At that time, the intelligent scheduling system will divide the grid work units Identify areas with insufficient filling and generate replenishment tasks. When At that time, the intelligent scheduling system will divide the grid work units Identify the overfilled area and generate a material transfer task. When Furthermore, when the flatness of the grid work unit meets the requirements, the intelligent scheduling system will allocate the grid work unit... The area was identified as a qualified region. This indicates the average allowable elevation deviation of the current fill layer.

[0080] according to Figure 8 As shown in the process, the unmanned loader and unmanned bulldozer continuously acquire multi-source positioning observation information formed by real-time dynamic carrier phase differential positioning equipment, inertial measurement unit, vehicle mileage acquisition equipment and vehicle-mounted environmental perception equipment during construction.

[0081] Unmanned dump trucks The operating status adopts It means that the first The positioning observation information output by the positioning device adopts It means that the first The observation relationship corresponding to the positioning device adopts It indicates that there are no unmanned earthmoving vehicles. The continuous spatial pose is obtained according to the following relationship: ; in, This indicates the unmanned earthmoving vehicle obtained through joint state estimation. Continuous spatial pose Indicates time The collection of effective location observation information Indicates the first The credibility weight of location observation information Indicates the first The observation error constraint matrix corresponding to the positioning observation information.

[0082] Credible weight Determined according to the following relationship: ; in, Indicates the first Reliability score of location-based observation information This represents the traversal sequence number within the set of valid location observation information. Indicates the first The reliability score for positioning observation information is determined based on the positioning solution status, observation continuity, and environmental compatibility.

[0083] When a certain location observation information becomes abnormal, the corresponding reliability score is... Reduce, credibility weight Synchronous reduction prevents abnormal location observation information from dominating joint state estimation, and utilizes the remaining effective location observation information to maintain continuous output of vehicle spatial pose.

[0084] After the anomaly location observation information is recovered, the recovered location results are progressively corrected according to the following relationship: ; in, This indicates the vehicle's spatial orientation before the gradual correction. This indicates the restored location result. This indicates the progressively corrected spatial orientation of the vehicle. This represents the asymptotic correction factor. By restricting the asymptotic correction factor... The rate of change of the position allows the recovered positioning results to be gradually incorporated into the vehicle's spatial pose calculation, thus preventing sudden changes in the vehicle's control state.

[0085] The onboard controller determines the combined operating status of the vehicle chassis and working device based on the vehicle's spatial position, motion state, and the attitude and load status of the bucket or blade. (Unmanned earthmoving vehicle) During the control cycle The coupled operation state adopts It indicates that the control quantities of the vehicle chassis and working device are adopted. This indicates that the operational disturbance caused by earthwork resistance, ground adhesion status, and working device load is adopted. express.

[0086] The vehicle-working device coupled prediction model is expressed as follows: ; in, Indicates no unmanned earthmoving vehicles In the coupled operation state of the next control cycle, This describes the coupled state prediction process formed by the vehicle's motion response, the working device's action response, and the interaction between the vehicle and the earthwork. For unmanned loaders, the working device control variables are used to adjust the bucket lifting and tilting states. For unmanned bulldozers, the working device control variables are used to adjust the blade lifting and tilting states.

[0087] The onboard controller calculates the trajectory deviation of the unmanned earthmoving vehicle relative to the cooperative path based on the joint operation status, and calculates the operational deviation of the working device based on the elevation difference between the actual working surface and the target filling surface. The objective function of rolling predictive control is expressed as: ;in, Indicates no unmanned earthmoving vehicles The control target value is within the prediction range. This indicates the number of control cycles included in the forecast range. Indicates the control cycle number within the prediction range. Indicates no unmanned earthmoving vehicles During the control cycle relative to the trajectory deviation of the cooperative path, This indicates the operational deviation of the actual working surface corresponding to the working device relative to the target filling surface. This represents the control increment between adjacent control cycles. Represents the trajectory deviation weight matrix. Represents the job deviation weight matrix. Represents the control increment weight matrix, symbol This indicates the matrix transpose.

[0088] The onboard controller achieves the control target value through rolling solution. The reduced vehicle chassis control and working device control send the vehicle chassis steering, driving and braking control to the drive-by-wire chassis, and the bucket or blade attitude control to the electro-hydraulic actuator, so that the unmanned earthmoving vehicle runs along the cooperative path and performs earthmoving, placement, filling or leveling according to the fill thickness control benchmark.

[0089] After executing the current control quantity in each control cycle, the on-board controller reacquires the vehicle motion response, the load of the working device, and the actual changes in the working surface, and corrects the vehicle-working device coupling prediction model and the control quantity for the next control cycle accordingly, thereby forming a coordinated closed-loop control of the vehicle chassis and the working device.

[0090] After completing a round of earthmoving, filling, or leveling, the dynamic 3D model acquisition system captures the actual work surface, and the intelligent scheduling system obtains the task execution results of the unmanned earthmoving vehicles. The task execution results are then processed using... The 3D construction model is updated based on the following relationships: ; in, This represents the time interval between adjacent model update times. This refers to the process of updating the 3D construction model based on the actual work surface and task execution results. This represents the 3D construction model after feedback and updates.

[0091] The updated roadbed filling tasks, operation sequences, and collaborative paths adopt This indicates that its generation relation is: ; in, This refers to the process of task generation, vehicle matching, and collaborative path planning performed by the intelligent dispatch system. This represents the unmanned earthmoving vehicle operation data corresponding to the model update time. This indicates the updated roadbed filling task, operation sequence, and collaborative path.

[0092] When there is still an area exceeding the allowable deviation between the actual working surface and the target filling surface, the intelligent scheduling system will adjust the configuration based on the updated 3D construction model. and unmanned earthmoving vehicle operation data Generate material replenishment tasks, material transfer tasks, or re-leveling tasks, and update the collaborative path and closed-loop control parameters simultaneously.

[0093] Once all work units in the current filling layer meet the requirements for target filling surface, filling thickness, and flatness, the intelligent scheduling system ends the construction of the current filling layer and records the data collected by the UAV, the vehicle-mounted environmental perception data, the unmanned earthmoving vehicle operation data, the task allocation results, the operation sequence, the collaborative path, the vehicle chassis control data, the working device control data, and the construction quality data.

[0094] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. An intelligent construction method for roadbed filling based on unmanned earthmoving vehicle swarms, characterized by: Includes the following steps: S1. Acquire environmental perception data, engineering design data, and unmanned earthmoving vehicle operation data of the roadbed filling area. Perform spatiotemporal fusion on the acquired data to construct a three-dimensional construction model that is dynamically updated with the construction process. Based on the three-dimensional construction model, form construction status data that characterizes the work area, earthwork demand, and construction constraints. S2, based on construction status data and the operational status of the unmanned earthmoving vehicle cluster, decomposes the roadbed filling task and matches vehicles, generating the operation sequence and collaborative path of unmanned loaders and unmanned bulldozers. S3 controls the unmanned loader and unmanned bulldozer to enter the corresponding work area according to the work sequence and collaborative path. The unmanned loader performs earthwork shoveling, transportation and material placement, and the unmanned bulldozer performs earthwork filling and leveling. S4 integrates the positioning information, environmental perception information and working device status information of unmanned earthmoving vehicles to determine the joint operation status of the vehicle and working device, and performs collaborative closed-loop control of the vehicle chassis and working device based on the joint operation status, collaborative path and target filling surface. S5 obtains the actual working surface after earthwork placement and filling, as well as the task execution results of unmanned earthwork vehicles. Based on the difference between the actual working surface and the target filling surface, it continuously updates the three-dimensional construction model, roadbed filling task, collaborative path, and closed-loop control parameters until the current filling layer meets the construction quality requirements.

2. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 1, characterized in that: steps S1 includes: Before construction, drone photography equipment and drone lidar were used to survey the roadbed filling area to obtain initial environmental data characterizing the topography, boundaries, slope and pothole conditions of the construction area, and to connect it with the engineering design data corresponding to the roadbed cross section, layered filling, base treatment and drainage layout. An initial three-dimensional model of the roadbed filling area is established based on initial environmental data and engineering design data; During construction, the onboard environmental sensing equipment and vehicle status acquisition equipment of the unmanned earthmoving vehicles are used to continuously acquire information on changes in the work surface, earthwork distribution, obstacle status, and the operating status of the unmanned earthmoving vehicles. Data from different data sources before and after construction are sequentially unified in coordinates, aligned in time, spatially correlated, and merged and updated. The initial 3D model is then recursively corrected based on the fusion results to generate a 3D construction model that dynamically changes with the construction process.

3. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 2, characterized in that: Construction status data is generated based on a 3D construction model, including: Spatial difference calculations are performed between the current working face in the 3D construction model and the target filling face in the engineering design data to form a filling state field that reflects the distribution of earthwork gaps in the roadbed filling area. The work units are divided according to the spatial continuity of the filling site, the operational capacity of unmanned earthmoving vehicles, and the traffic conditions of the construction area, and the earthwork requirements corresponding to each work unit are determined. Based on the spatial shape and earthwork requirements of each work unit, generate the material placement benchmark, the filling benchmark, and the filling thickness control benchmark. Construction constraints are generated based on construction boundaries, foundation conditions, drainage facility protection range, obstacle distribution, and vehicle traffic conditions. By linking work units, earthwork requirements, material placement benchmarks, filling benchmarks, filling thickness control benchmarks, and construction constraints to a three-dimensional construction model, construction status data is generated for filling task generation, collaborative path planning, and construction control.

4. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 1, characterized in that: Step S2 includes: Based on the construction status data, the roadbed filling construction process is analyzed and converted into machine-executable task units with operation location, target amount of work, operation type and construction connection relationship. A task chain is established based on the construction connection relationship. Establish vehicle capability status based on the equipment type, current location, availability, operational capabilities, and task execution progress of unmanned earthmoving vehicles; Based on the task chain and vehicle capability status, establish the matching relationship between tasks and vehicles, with the goal of reducing the comprehensive construction cost caused by vehicle idling, operation waiting, energy consumption and path conflicts, and determine the task allocation results and operation sequence of unmanned loaders and unmanned bulldozers. When construction status data, vehicle capacity status, or construction environment changes, the vehicle matching is re-performed for the task units affected by the changes, and the task allocation results and operation sequence are updated on a rolling basis.

5. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 4, characterized in that: The collaborative path is generated using a two-tiered path planning approach that combines global planning with local adjustments, including: Based on the 3D construction model and construction constraints, a passage cost field is established that can be updated according to the working surface, earthwork distribution and obstacle status. Based on the passage cost field, a global reference path is generated for each unmanned earthmoving vehicle to complete the corresponding task. Based on the expected operation process of each unmanned earthmoving vehicle along the global reference path, establish a path occupancy status with time attributes; When the path occupancy status of different unmanned earthmoving vehicles overlaps, the passage order, waiting position or driving route of the unmanned earthmoving vehicles shall be adjusted according to the construction connection relationship of the corresponding tasks and the vehicle operation status. When the vehicle-mounted environmental perception device detects a change in traffic conditions, it performs local reconstruction of the path in the affected area and connects the locally reconstructed path with the unaffected global reference path to form a dynamically updated cooperative path.

6. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 1, characterized in that: The combined operating status of the vehicle and the working device in step S4 is determined as follows: Acquire multi-source positioning observation information formed by real-time dynamic carrier phase differential positioning equipment, inertial measurement unit, vehicle mileage acquisition equipment and vehicle-mounted environmental perception equipment; The observation credibility is determined based on the effectiveness and stability of multi-source positioning observation information, and joint state estimation is performed within a continuous time window to obtain the continuous spatial pose of unmanned earthmoving vehicles in the three-dimensional construction model. When some multi-source positioning observation information is abnormal, reduce the impact of abnormal information on joint state estimation, and use the remaining effective information to maintain continuous output of spatial pose. Once the abnormal information is recovered, the deviation between the recovered positioning result and the current spatial pose is progressively corrected. The combined operating state of the vehicle chassis and working device is determined based on the spatial position, vehicle motion state, and the attitude and load state of the bucket or blade.

7. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 6, characterized in that: Coordinated closed-loop control of the vehicle chassis and working device includes: A vehicle-working device coupled prediction model is established based on the motion response of unmanned earthmoving vehicles, the action response of working devices, and the earthmoving load. The trajectory deviation of the unmanned earthmoving vehicle relative to the collaborative path is determined based on the joint operation status, and the operational deviation of the working device is determined based on the elevation difference between the actual working surface and the target filling surface. Rolling predictive control is used to generate steering, drive and braking control quantities for the vehicle chassis, and adaptive feedback control is used to generate attitude control quantities for the bucket or blade. The vehicle chassis control quantity is sent to the drive-by-wire chassis, and the working device control quantity is sent to the electro-hydraulic actuator, so that the unmanned earthmoving vehicle runs along the cooperative path and performs earthmoving, placement, pushing or leveling according to the filling thickness control benchmark. Based on the vehicle motion response, the load of the working device, and the actual changes in the working surface, the vehicle-working device coupling prediction model and the control quantity for the next control cycle are modified.

8. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 1, characterized in that: When the 3D construction model identifies a depressed area to be filled within the roadbed filling area, a deep pit filling operation mode is adopted. The deep pit filling operation mode includes: Based on the spatial shape of the sunken area to be filled, the earthwork gap and the surrounding traffic conditions, the unloading area is determined outside the sunken area to be filled, and the earthwork transfer and placement area is determined at the edge of the sunken area to be filled. Guide transport vehicles to unload soil in the unloading area, and control unmanned loaders to scoop up and transport the unloaded soil to the soil transfer and placement area. Based on the remaining earthwork gap in the sunken area to be filled and the filling thickness control benchmark, the unmanned bulldozer is controlled to push the earthwork in layers from the earthwork transfer and placement area into the sunken area to be filled and level it. Obtain the actual working surface after each filling, update the remaining earthwork gap in the sunken area to be filled, and adjust the subsequent material placement position, filling direction, and filling number.

9. The intelligent construction method for roadbed filling based on unmanned earthmoving vehicle clusters according to claim 1, characterized in that: When the 3D construction model identifies the subgrade filling area as a continuously unfolding work surface, a grid paving operation mode is adopted. The grid paving operation mode includes: Based on the spatial range of the continuously unfolding work surface, the filling thickness control benchmark, and the operational capabilities of unmanned earthmoving vehicles, multiple interconnected grid work units are divided. The earthwork requirements are determined based on the current working face and the target filling face corresponding to each grid operation unit, and the corresponding quantitative material placement task is generated. Control the unmanned loader to transport the earthwork to each grid work unit and execute quantitative material placement; control the unmanned bulldozer to perform filling and leveling in sequence according to the spatial connection relationship between the grid work units. The system obtains the actual elevation, filling thickness, and leveling status of each grid operation unit after the completion of the operation. Based on the actual filling status, it identifies underfilled or overfilled areas and generates material replenishment tasks, material transfer tasks, or re-leveling tasks. After each grid work unit meets the construction quality requirements of the current filling layer, the corresponding task execution data and construction quality data are recorded.

10. An intelligent construction system for roadbed filling based on an unmanned earthmoving vehicle cluster, characterized by: This includes a dynamic 3D model acquisition system for work scenarios, an intelligent scheduling system, and a fleet of unmanned earthmoving vehicles; The dynamic 3D model acquisition system for the work scene includes a drone acquisition device and a model processing device, which are used to acquire construction environment data of the roadbed filling area, calculate earthwork volume, and construct a dynamic 3D construction model. The intelligent scheduling system communicates with the dynamic 3D model acquisition system for the work scenario, and is used to generate construction tasks, work sequence and collaborative paths based on the dynamic 3D construction model and the operating status of the unmanned earthmoving vehicle cluster. The unmanned earthmoving vehicle cluster includes unmanned loaders and unmanned bulldozers. Both unmanned loaders and unmanned bulldozers are equipped with onboard perception, navigation and control devices to obtain the equipment operating status and construction environment status. They perform earthwork placement, filling and leveling according to the construction task, operation sequence and collaborative path, and feed back the task execution results to the intelligent scheduling system and the dynamic three-dimensional model acquisition system of the operation scene to update the dynamic three-dimensional construction model and subsequent construction control data.

Images