A construction robot autonomous path planning and obstacle avoidance method and system based on a BIM model

Abstract

The application discloses a kind of based on BIM model's construction robot autonomous path planning and obstacle avoidance method and system, comprising: obtaining BIM model, extracts the three-dimensional geometric information and semantic information of construction area;Initial global path is generated based on the geometry and semantic information of BIM model;Construction robot carries multiple sensors and real-time collection environmental data;Fusion BIM model prior information and real-time sensor data, adopt SLAM technology to realize the accurate positioning of robot in BIM coordinate system, and local real-time obstacle avoidance and path dynamic adjustment are carried out in combination with dynamic window method and artificial potential field method;Path planning information is transmitted to robot control system for execution by communication module.The application makes full use of the geometry and semantic information of BIM model to improve the accuracy and efficiency of path planning, realizes high-precision positioning and real-time dynamic obstacle avoidance by BIM prior information assisted multi-sensor fusion SLAM, effectively guarantees the safety of robot in dynamic construction environment, significantly reduces artificial intervention, and improves construction automation level.

Description

Technical Field

[0001] This invention belongs to the field of intelligent construction and robot navigation technology, specifically relating to a method and system for autonomous path planning and obstacle avoidance of construction robots based on BIM models. Background Technology

[0002] As a crucial pillar of the global economy, the construction industry has long faced challenges such as labor shortages, low productivity, and frequent safety accidents. In recent years, robotics and automation technologies have been increasingly introduced into the construction field, aiming to assist or replace manual labor in completing repetitive, labor-intensive, and hazardous construction tasks. Autonomous navigation is the core technological foundation for unmanned operation of construction robots, while path planning and obstacle avoidance are key aspects of autonomous navigation. However, the construction environment is characterized by dynamically changing spatial layouts, diverse and unfixed obstacles, and constantly adjusting work area boundaries as construction progresses. This results in traditional robot path planning methods facing limitations in adaptability, path safety, and planning efficiency in construction scenarios.

[0003] Currently, there are several solutions for path planning and navigation of construction robots, but they all have limitations to varying degrees.

[0004] Building Information Modeling (BIM), as a digital representation containing geometric and semantic information of building components, has been gradually introduced into the field of robot navigation as a global prior knowledge source for path planning. CN-115855068-A discloses a BIM-based autonomous robot path navigation method and system, which determines the effective movement range and generates an effective navigation image for path planning by performing a two-way spatial comparison between the BIM model and the robot's point cloud model. However, this method has limited utilization of the semantic information in the BIM model, failing to fully utilize semantic labels in the BIM model regarding obstacle types, work areas, and prohibited areas to optimize path planning, and lacks the ability to adjust in real time when facing dynamic obstacles in the construction environment. CN-111121780-A discloses a mobile robot navigation method that directly obtains path segments from drawing files to generate navigation points for movement. While this simplifies the navigation process, this method has limited environmental representation capabilities and struggles to handle complex 3D construction scenarios. CN120084329A discloses an automatic navigation method for construction site robots based on BIM, which achieves spatiotemporal joint modeling through environmental perception and prediction. However, this method focuses on candidate path generation under multi-objective optimization and still has shortcomings in the in-depth utilization of BIM semantic information.

[0005] The current BIM-based construction robot path planning and obstacle avoidance technologies have the following shortcomings: (1) The semantic information of the BIM model, such as the categories and locations of permanent structural components like walls and columns, the boundaries of the construction work area, and prohibited areas, has not been fully extracted and effectively utilized in the entire process of path planning and obstacle avoidance, resulting in a lack of consideration for the safety and rationality of the generated path; (2) The fusion depth between the prior information of the BIM model and the real-time data of multiple sensors is insufficient, and there is a lack of a complete technical solution that uses BIM semantic prior as the basis for SLAM positioning constraints and obstacle avoidance strategies; (3) The obstacle avoidance strategy lacks semantic perception capabilities and cannot adopt differentiated obstacle avoidance strategies based on the semantic type of obstacles, such as permanent structural components and temporary stacked materials; (4) Existing technical solutions are mostly improvements of a single link and lack a complete technical link from BIM information extraction, global path planning, multi-sensor fusion positioning to dynamic obstacle avoidance. Summary of the Invention

[0006] To address the problems existing in the prior art, the present invention aims to provide a method and system for autonomous path planning and obstacle avoidance of construction robots based on BIM models. By deeply mining the geometric and semantic information in the BIM model and integrating it into the entire process of global path planning, multi-sensor fusion SLAM localization, local dynamic obstacle avoidance, and path adjustment, a complete technical solution from BIM prior information to real-time autonomous navigation of the robot is constructed.

[0007] In a first aspect, the present invention provides a method for autonomous path planning and obstacle avoidance of a construction robot based on a BIM model, comprising the following steps: Step S1: Obtain the BIM model of the target construction area, and extract the three-dimensional geometric information and semantic information from the BIM model. The semantic information includes at least the category and location of permanent structural components, the marked area of ​​movable obstacles, the boundary of the construction work area, and the prohibited passage area.

[0008] Step S2: Construct a hierarchical grid map based on the three-dimensional geometric and semantic information, wherein different grids are assigned differentiated access cost weights according to the semantic information, grids corresponding to permanent structural components are set as impassable, grids in the working area are set as low access cost, and grids in prohibited areas are set as high access cost; an improved A is applied to the hierarchical grid map. * The algorithm or RRT algorithm generates an initial global path from the starting position to the target job position; The hierarchical raster map includes a geometric layer, a semantic layer, and a cost layer; wherein the geometric layer stores three-dimensional geometric information, the semantic layer stores the semantic labels of each raster, and the cost layer stores the passage cost weights calculated based on the semantic labels; the improved A * The algorithm or RRT algorithm searches on the cost layer; The improvement A* The evaluation function of the algorithm is: , in For nodes The comprehensive evaluation function value, From the starting node to the node The actual cost, For the node Heuristic cost estimation to the target node The weighting coefficients of the heuristic function are dynamically adjusted based on BIM semantic information. For semantic information-based penalty terms, when node When located in a restricted area or near permanent structural members Take the larger value.

[0009] The improved RRT algorithm described in step 2 includes: The semantic bias sampling strategy is based on the semantic access cost of each raster. Calculate sampling weights Random sampling points are selected in free space using a roulette wheel method, which makes the sampling tend to favor low-cost working areas; During the expansion of the RRT tree, cost-weighted distance is used for nearest neighbor selection, and the cost of newly generated nodes is evaluated. When a new node falls into a high-cost grid, it is selected with probability. Accept extensions; Where, p i For grid The relative probability weights of those selected as random sampling points; costi For grid The permissible cost value in the hierarchical cost layer; λ This is the sampling bias attenuation coefficient; μ This is the probability decay coefficient for acceptance. The initially generated RRT path is pruned using BIM semantic information to remove redundant nodes, and a smooth path is generated in the working area through B-spline interpolation to ensure that the safe distance from permanent structural components is not less than a preset threshold.

[0010] Step S3: The construction robot is equipped with a multi-sensor system, which includes at least a lidar, an inertial measurement unit, and a vision camera to collect environmental data in real time during movement along the initial global path. The multi-sensor system further includes an ultra-wideband positioning tag and / or a wheeled odometer to provide auxiliary positioning information in scenarios where lidar data is sparse or environmental features are not obvious. The environmental data includes lidar 3D point cloud data, IMU six-axis inertial data, visual image data, and UWB ranging data.

[0011] Step S4: Using the semantic information of the BIM model as a priori constraint, the real-time collected environmental data is fused together, and the robot's pose is estimated in real time and the local environment map is updated using SLAM technology based on BIM prior constraints. The obstacles detected by the real-time sensors are matched and compared with the BIM semantic information to identify permanent fixed obstacles that have been marked in the BIM model and newly added dynamic / temporary obstacles that have not been marked in the BIM model. The BIM-based SLAM technology specifically includes: using the geometric information of permanent structural components in the BIM model as the initial map prior; using LiDAR-Inertial Odometry for front-end odometry estimation; performing real-time registration of the estimated pose with the geometric features extracted from the BIM model; constructing a factor graph optimization problem containing BIM prior constraints; solving the robot's optimal pose estimation through nonlinear least squares optimization; and using semantic information from the BIM model to perform semantic annotation and dynamic object filtering on the SLAM-constructed map.

[0012] Step S5: During the robot's movement, when the newly added dynamic / temporary obstacle is detected, the local path is replanned using the dynamic window method and the artificial potential field method. While maintaining a safe distance from permanent fixed obstacles, the local path is dynamically adjusted to bypass the newly added dynamic / temporary obstacle. At the same time, BIM semantic information is used to optimize the obstacle avoidance path so that it remains within the working area. The obstacle avoidance strategy integrating the dynamic window method and the artificial potential field method specifically includes: First, using BIM semantic information to differentiate the repulsive potential field in the artificial potential field method, the high-intensity impassable repulsive field generated by permanent structural components, the traversable medium-intensity repulsive field generated by newly added dynamic / temporary obstacles, and the low-intensity boundary repulsive field generated by the boundary of the work area; then, in the velocity sampling space of the dynamic window method, the potential field gradient of the artificial potential field method is used as a cost index for trajectory evaluation, which together with the target orientation cost, velocity cost, and obstacle distance cost constitutes a comprehensive evaluation function for the velocity command; in each control cycle, the velocity command that optimizes the comprehensive evaluation function is selected from the velocity sampling space to drive the robot's motion.

[0013] Step S6: The planned path information is transmitted to the robot control system in real time through the communication module, and the robot control system drives the robot actuator to complete the construction task according to the planned path.

[0014] Furthermore, the construction robot is an autonomous mobile construction robot with a wheeled mobile chassis equipped with a work execution mechanism, which is a bricklaying robot arm, a spraying robot arm, or a ground leveling actuator; the robot control system receives path planning instructions through the robot operating system and converts them into speed control instructions for the robot chassis and action instructions for the work execution mechanism.

[0015] Secondly, the present invention also provides an autonomous path planning and obstacle avoidance system for construction robots based on BIM models, comprising: The BIM model processing module is used to acquire BIM models and extract three-dimensional geometric and semantic information. A global path planning module is used to generate an initial global path based on the three-dimensional geometric information and semantic information; A multi-sensor data acquisition module, mounted on a construction robot, includes at least a lidar, an inertial measurement unit, and a vision camera, for real-time acquisition of environmental data; a BIM sensor fusion localization and mapping module, used to use the semantic information of the BIM model as prior constraints, fuse the real-time environmental data, and employ SLAM technology based on BIM prior constraints to estimate the robot's pose in real time and update the local environmental map; the BIM sensor fusion localization and mapping module further includes: The front-end odometry submodule is used for front-end pose estimation based on lidar point cloud data and IMU inertial data; The BIM registration submodule is used to perform real-time scanning and matching of the estimated pose with the geometric features in the BIM model. The factor graph optimization submodule is used to construct a factor graph containing BIM prior constraints and solve for the optimal pose through nonlinear least squares optimization. The semantic annotation submodule is used to perform semantic annotation and dynamic object filtering on the SLAM-built map using semantic information from the BIM model.

[0016] A dynamic obstacle avoidance and path adjustment module is used to perform local path replanning using a dynamic window method and an artificial potential field method when new dynamic / temporary obstacles are detected, and to optimize the obstacle avoidance path by combining BIM semantic information; the dynamic obstacle avoidance and path adjustment module further includes: The artificial potential field calculation submodule is used to calculate differentiated gravitational and repulsive potential fields for different types of obstacles and target points based on BIM semantic information; The dynamic window velocity sampling submodule is used to sample candidate velocity commands in the velocity space under robot kinematic constraints; The trajectory evaluation submodule is used to score candidate trajectories by comprehensively considering the target orientation, obstacle distance, potential field gradient, and velocity smoothness, and select the optimal velocity command.

[0017] The communication and control module is used to transmit the planned path information to the robot control system to drive the robot actuators.

[0018] This invention relates to a BIM semantic information-driven hierarchical grid map construction and global path planning. Traditional methods, when using BIM models for path planning, typically only extract geometric information to construct an occupation grid map or point cloud map, ignoring the rich semantic information within BIM. This invention innovatively proposes a three-layer grid map structure comprising a geometric layer, a semantic layer, and a cost layer: the geometric layer stores 3D geometric information; the semantic layer stores semantic labels for each grid, such as permanent structural components, movable obstacle marking areas, construction work area boundaries, and prohibited areas; the cost layer calculates differentiated access cost weights based on the semantic labels, setting permanent structural components as impassable, work areas as low access cost, and prohibited areas as high access cost. Based on this hierarchical grid map, improvements are made to A... * The evaluation function of the algorithm or the sampling strategy of the RRT algorithm enables global path planning to prioritize guiding the path through the work area and away from the prohibited area while ensuring the feasibility of the path, thereby improving the construction adaptability and safety of the path from the source.

[0019] This invention utilizes deep fusion of BIM prior information and multi-sensor real-time data for SLAM localization. Existing BIM-SLAM integration schemes mainly focus on geometric registration or initialization, lacking a systematic utilization of BIM semantic priors during SLAM estimation. This invention uses the geometric information of permanent structural components (walls, columns, beams, etc.) in the BIM model as initial map priors, employs LiDAR-Inertial Odometry for front-end odometry estimation, and performs real-time registration of the estimated pose with geometric features extracted from the BIM model, such as wall planes and column edges. A factor graph optimization problem incorporating BIM prior constraints is constructed, and the optimal robot pose is solved using nonlinear least squares optimization. Simultaneously, BIM semantic information is used to semantically annotate and filter dynamic objects on the local map constructed by SLAM, effectively distinguishing between permanently fixed obstacles already labeled in the BIM and dynamic / temporary obstacles added during construction, providing a semantic basis for subsequent differentiated obstacle avoidance strategies.

[0020] This invention employs a BIM semantic-guided dynamic window-artificial potential field fusion obstacle avoidance strategy. Traditional obstacle avoidance methods use a uniform strategy for all obstacles, failing to balance obstacle avoidance safety and path smoothness. This invention proposes a differentiated obstacle avoidance strategy guided by BIM semantics: First, BIM semantic information is used to differentiate the repulsive potential field in the artificial potential field method—a high-intensity, impassable repulsive field is generated for permanent structural components to achieve "hard obstacle avoidance," a medium-intensity, traversable repulsive field is generated for newly added dynamic / temporary obstacles to achieve "soft obstacle avoidance," and a low-intensity boundary repulsive field is generated at the work area boundary to guide the robot to remain within a safe operating range. Then, in the velocity sampling space of the dynamic window method, the potential field gradient of the artificial potential field method is used as a cost indicator for trajectory evaluation, enabling the robot to make more reasonable obstacle avoidance decisions using semantic information while avoiding dynamic obstacles—for example, when faced with multiple obstacle avoidance direction choices, priority is given to directions towards the interior of the work area and away from permanent structural components.

[0021] Compared with the prior art, the present invention has the following significant advantages: By leveraging the precise 3D geometric information of the BIM model as a global prior, and combining it with semantic information to construct a layered raster map and differentiate access costs, the generated global paths are not only geometrically feasible but also conform to the spatial logic of the construction task. This improves path planning efficiency by approximately 35% compared to traditional methods, and also increases path safety margin compared to traditional A / B methods. * The algorithm is improved by approximately 80%.

[0022] By setting up differentiated artificial potential fields guided by BIM semantic information, robots can accurately distinguish between permanent structural components and temporary dynamic obstacles, and adopt different obstacle avoidance strategies. This significantly reduces the risk of collisions between robots and building structures. In simulated construction scenarios, the safe distance between robots and obstacles is increased by more than 80% compared to traditional uniform obstacle avoidance strategies.

[0023] By using BIM semantic priors as constraints for SLAM factor graph optimization and utilizing the geometric features of permanent structural components for real-time scanning and matching, the positioning accuracy and robustness under difficult conditions such as construction dust, occlusion, and weak texture are effectively improved. The positioning accuracy of BIM-assisted SLAM is about 40% higher than that of pure laser SLAM without BIM priors.

[0024] This invention integrates BIM model information, global path planning, multi-sensor fusion SLAM, dynamic obstacle avoidance, and robot control into a complete autonomous construction system, achieving seamless connection from BIM design model to autonomous robot construction, significantly reducing the need for manual intervention, and is applicable to various construction scenarios such as bricklaying, spraying, and ground leveling, thus powerfully promoting the development of intelligent construction technology. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a module architecture diagram of the system of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] Example 1: An autonomous path planning and obstacle avoidance system for construction robots based on BIM models, applying the method of this invention to an indoor bricklaying construction scenario. The construction robot is an autonomous bricklaying robot mounted on a wheeled mobile chassis with a six-DOF bricklaying robotic arm. The chassis is equipped with a 16-line LiDAR with a sampling frequency of 10Hz, including a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer; a nine-axis IMU with a sampling frequency of 200Hz; a 1280×720 resolution, 30fps RGB-D visual camera; an ultra-wideband positioning tag; and a wheeled odometer, including: The BIM model processing module is used to acquire BIM models and extract three-dimensional geometric and semantic information. A global path planning module is used to generate an initial global path based on the three-dimensional geometric information and semantic information; A multi-sensor data acquisition module, mounted on a construction robot, includes at least a lidar, an inertial measurement unit, and a vision camera, for real-time acquisition of environmental data; a BIM sensor fusion localization and mapping module, used to use the semantic information of the BIM model as prior constraints, fuse the real-time environmental data, and employ SLAM technology based on BIM prior constraints to estimate the robot's pose in real time and update the local environmental map; the BIM sensor fusion localization and mapping module further includes: The front-end odometry submodule is used for front-end pose estimation based on lidar point cloud data and IMU inertial data; The BIM registration submodule is used to perform real-time scanning and matching of the estimated pose with the geometric features in the BIM model. The factor graph optimization submodule is used to construct a factor graph containing BIM prior constraints and solve for the optimal pose through nonlinear least squares optimization. The semantic annotation submodule is used to perform semantic annotation and dynamic object filtering on the SLAM-built map using semantic information from the BIM model.

[0028] A dynamic obstacle avoidance and path adjustment module is used to perform local path replanning using a dynamic window method and an artificial potential field method when new dynamic / temporary obstacles are detected, and to optimize the obstacle avoidance path by combining BIM semantic information; the dynamic obstacle avoidance and path adjustment module further includes: The artificial potential field calculation submodule is used to calculate differentiated gravitational and repulsive potential fields for different types of obstacles and target points based on BIM semantic information; The dynamic window velocity sampling submodule is used to sample candidate velocity commands in the velocity space under robot kinematic constraints; The trajectory evaluation submodule is used to score candidate trajectories by comprehensively considering the target orientation, obstacle distance, potential field gradient, and velocity smoothness, and select the optimal velocity command.

[0029] The communication and control module is used to transmit the planned path information to the robot control system to drive the robot actuators.

[0030] The method used is as follows: Step S1: Obtain the BIM model of the target construction floor in IFC format, and analyze the 3D geometric information of the bricklaying construction area in the model, including the location and dimensions of walls, the location and dimensions of door and window openings, the location of columns and beams, and semantic information including the IFC type identifier of the walls, material properties, construction work elevation surface, reserved opening location, bricklaying stacking area, and no-passage area. The extracted semantic information also includes the construction sequence markings of the walls, such as the bottom-up layered bricklaying sequence and brick type and size parameters.

[0031] Step S2: Divide the bricklaying construction area into a 0.1m × 0.1m × 0.1m three-dimensional grid, and construct a three-layer grid map consisting of a geometric layer, a semantic layer, and a cost layer. In the semantic layer, areas with existing walls are marked as "permanent structural components," meaning they are impassable; locations where walls are to be built are marked as "working areas," meaning they have low access costs with a cost coefficient of 1.0; door and window openings are marked as "passable areas," with a cost coefficient of 0.8; brick stacking areas are marked as "material areas," with a cost coefficient of 1.2; and the area outside the construction zone is marked as "prohibited areas," with a cost coefficient of ∞.

[0032] Adopting improved A * The algorithm performs a global path search at the cost layer, with the following evaluation function: , Dynamically adjust according to the construction sequence of the wall to be built. The value is taken from the area closest to the current construction elevation. Smaller values ​​are used to encourage paths to move closer to the target area; penalty term When approaching existing walls and columns, the value is set to 1.5 times the grid cost coefficient to ensure the robot maintains a safe distance from completed structures. The generated initial global path starts from the bricklaying area and passes through each wall to be built sequentially along the construction sequence.

[0033] Step S3: During the robot's movement along the global path, the LiDAR collects 3D point cloud data at a frequency of 10Hz, the IMU collects inertial data at a frequency of 200Hz, and the camera collects visual images at a frequency of 30Hz. The geometric information of the permanent structural components, including the walls and columns already constructed, in the BIM model is used as the initial SLAM map prior. LiDAR-inertial odometry is used for front-end pose estimation. The estimated pose is then scanned and matched in real-time with the wall plane equations and column edge features extracted from the BIM model to construct a factor graph optimization problem. The nodes in the factor graph include the robot's six-DOF pose at each time step, and the edges include IMU pre-integration edges, LiDAR scan matching edges, and BIM geometric feature registration edges. Real-time optimization is performed using the iSAM2 incremental nonlinear least squares solver to obtain the robot's accurate positioning result in the BIM coordinate system.

[0034] Step S4: Semantically annotate the local map constructed by SLAM using BIM semantic information. Project the LiDAR point cloud into the BIM coordinate system using the current estimated pose, and determine the spatial relationship between each point and permanent structural components in the BIM: if a point falls within the geometric range of a BIM wall / column and the distance is less than the threshold of 0.05m, it is marked as a "permanent fixed obstacle"; if a clustered point cloud region does not match any permanent BIM component, it is marked as a "new dynamic / temporary obstacle". Through this semantic annotation mechanism, the difference between the BIM model and the actual construction site can be perceived.

[0035] Step S5: During the robot's movement, when the semantic annotation identifies "new dynamic / temporary obstacles" such as brick pallets temporarily stacked by workers or moving auxiliary equipment, the BIM semantic guidance fusion obstacle avoidance algorithm is triggered.

[0036] First, an artificial potential field is constructed based on semantic type differentiation: permanent structural components annotated by BIM generate a high-intensity repulsive potential field, with the potential field function as follows: , in Range of repulsive force ,when The potential field value is infinite to achieve hard constraints); new temporary obstacles generate a medium-intensity repulsive potential field. , It is traversable; a low-intensity boundary repulsive potential field is generated at the boundary of the working area. , The target location, i.e., the current masonry work site, generates an attractive potential field.

[0037] In the Dynamic Window Method (DWA) framework, the robot's velocity sampling space is constrained by chassis kinematics, i.e., a maximum linear velocity of 0.5 m / s and a maximum angular velocity of 1.0 rad / s. The gradient information of the aforementioned artificial potential field is used as a cost metric in the trajectory evaluation function. The comprehensive evaluation function for velocity commands is: , Wherein, the linear velocity is With angular velocity; heading ( v , ω ) represents the target azimuth angle; dist( v , ω ) represents the distance to the obstacle; vel( v , ω () represents the speed smoothing value; The potential field value at the endpoint of the candidate trajectory, weight , , , Solving within each control cycle, i.e., 100ms, makes... The optimal speed command drives the robot's movement.

[0038] Step S6: The planned path information is transmitted to the robot control system via the WiFi communication module. The chassis controller drives the four-wheel drive omnidirectional wheels to move along the planned path, while the robotic arm controller controls the spray gun tip to maintain a constant working distance of 0.3m from the wall. After the robot completes the spraying work on the current wall, it automatically moves to the next wall to be sprayed to continue the work, realizing automated spraying construction throughout the house.

[0039] Performance Verification: A comparative spraying operation experiment was conducted in a 4m×5m indoor room. The room included four walls to be sprayed, two windows, one doorway, and three BIM-marked areas for switches and sockets that should not be sprayed. The comparison scheme was a traditional full-coverage path planning + DWA obstacle avoidance scheme without BIM semantic information guidance. The experimental results are shown in Table 1 below: Table 1 Coating coverage 91.50% 99.30% Increase by 8.5% Number of times of accidental spraying 3.2 times / room 0.2 times / room Reduced by 93.8% Total task time (min) 28.5 22.3 Shortened by 21.8% Obstacle avoidance response time (ms) 320 180 Shortened by 43.8% Positioning accuracy RMSE (m) 0.18 0.06 Increased by 66.7% As shown in Table 1, the spray coverage increased from 91.5% to 99.3%, the number of accidental sprayings decreased by 93.8%, and the total operation time was shortened by 21.8%. These data indicate that the introduction of BIM semantic information, such as the location of switches and sockets and wall materials, significantly improved the coverage integrity and construction quality of the spraying operation.

[0040] Example 2 is essentially the same as Example 1, except that the spraying operation employs a fixed, full-coverage path strategy for the wall surface. The wall area and paint coverage parameters extracted from the BIM model adaptively adjust the row spacing and spray gun movement speed of the spraying path. For large, continuous walls (area ≥ 10m²), a larger row spacing of 0.4m and a higher movement speed of 0.4m / s are used; for small or irregularly shaped walls (area < 3m²), a smaller row spacing of 0.2m and a lower movement speed (0.2m / s) are used to ensure uniform spraying. Furthermore, the SLAM positioning scheme can automatically switch to a fusion positioning mode dominated by visual-inertial odometry and supplemented by BIM geometric features when the quality of the lidar data is severely degraded due to paint mist interference.

[0041] Example 3: This example is applied to a large-area concrete ground laser leveling construction scenario. It is basically the same as Example 1, except that the construction robot is an autonomous ground leveling robot with a tracked mobile chassis equipped with a laser leveling actuator. The chassis is equipped with a 10Hz multi-line lidar, a 200Hz IMU, a 25fps binocular vision camera, an RTK-GNSS module, and a wheeled odometer.

[0042] Step S1 differs in that: the BIM model of the target construction area is obtained, and the three-dimensional geometric information of the ground to be leveled is extracted, including the ground boundary outline, design elevation, slope requirements, location and size of drainage ditches and sump pits, and semantic information including the boundary and elevation requirements of different functional zones, the location of embedded parts, the location of reserved openings, and the boundary line between the completed pouring area and the area to be leveled.

[0043] Step S2 differs in that the construction area is divided into a 0.5m × 0.5m two-dimensional grid. In the semantic layer, the area to be leveled is marked as "Working Area" with a cost coefficient of 1.0; the area that has been leveled and where the concrete has initially set is marked as "No-Passage Area" with a cost coefficient of ∞; drainage ditches and sump pits are marked as "No-Passage Area" with a cost coefficient of ∞; the 0.3m range around the location of embedded parts is marked as "Cautionary Passage Area" with a cost coefficient of 3.0; and the area outside the boundary of the construction area is marked as "Prohibited Area" with a cost coefficient of ∞.

[0044] In ground leveling scenarios, robots typically need to perform a full-coverage scan of the area to be leveled. Therefore, the global path planning adopts a "U-shaped" coverage path strategy: using an improved A* algorithm, a U-shaped full-coverage path is generated by spiraling outwards from the center of the target area, with the path row spacing set according to the effective working width of the leveling actuator, which is 2.0m. The evaluation function includes a semantic penalty term. Used to ensure that the path remains within the area to be leveled and does not cross the "no-entry zone" that has already been leveled.

[0045] Step S3 differs in that the ground leveling robot is equipped with an RTK-GNSS module for global positioning in open outdoor areas, with GNSS positioning accuracy at the centimeter level. When the robot enters indoor areas or areas with limited GNSS signals, it automatically switches to a BIM-constrained laser-inertial SLAM mode. The SLAM module uses structured geometric features extracted from the BIM, such as column grid positions and wall boundaries, for scanning and matching to achieve sub-decimeter-level indoor positioning.

[0046] The difference in step S4 is that, in terms of semantic perception, embedded parts and reserved openings on the ground are detected in real time through LiDAR scanning and visual images, and compared with the positions marked in BIM. Embedded parts with position deviations exceeding 0.05m are marked and the map is updated.

[0047] The difference in step S5 is that during construction, the ground leveling robot typically works in collaboration with multiple pieces of equipment, such as concrete transport trucks and smoothing robots. This invention utilizes BIM semantic information to employ differentiated obstacle avoidance strategies for different types of construction equipment. When the leveling robot moves along a U-shaped path, if it detects a large, slow-moving, and predictable concrete transport truck ahead, it predicts the truck's direction of movement based on the transport route information marked in the BIM and plans a detour path in advance. If it detects a smoothing robot already in operation with a relatively fixed position, it avoids the obstacle using a dynamic window method with a smaller detour radius of approximately 1.5 times the robot's width. During obstacle avoidance, BIM semantic information is used to ensure that the detour path does not cross the boundary of the area to be leveled and does not trample on areas that have already been leveled and are beginning to set.

[0048] The difference in step S6 is that the planned path information is transmitted to the robot control system via a WiFi / 4G dual-mode communication module. The chassis controller controls the tracked chassis to move along the planned path and maintain the working posture of the leveling actuator. The leveling accuracy is stably controlled within ±5mm, and the levelness pass rate reaches over 98%.

[0049] Comparative Experiment: A comparative experiment was conducted in a 200m² concrete floor construction area, including three drainage ditches and five embedded parts. The comparative scheme was a GNSS-based fixed-route full-coverage scheme (without BIM semantic information guidance, without SLAM-assisted positioning, and without BIM semantic guidance for collaborative obstacle avoidance). The experimental results are shown in Table 2 below: Table 2 Leveling coverage 94.80% 99.50% Increase by 5.0% Number of times embedded parts are accidentally crushed 1.8 times / 200m² 0.1 times / 200m² Reduced by 94.4% GNSS positioning error in confined areas (m) 0.52 0.08 Reduced by 84.6% Multi-machine obstacle avoidance success rate 76.30% 97.80% Increased by 28.2% Total time spent on leveling operations (min) 52.6 41.8 Shortened by 20.5% Leveling accuracy pass rate 93.50% 98.80% Increased by 5.7% As shown in Table 2, the leveling coverage increased from 94.8% to 99.5%, the positioning accuracy in GNSS-restricted areas improved by 84.6% compared to the pure GNSS solution, and the obstacle avoidance success rate in multi-robot collaborative operations increased from 76.3% to 97.8%. These data indicate that BIM semantic information and BIM prior constraint SLAM have a significant effect on improving positioning accuracy and work quality in large-area ground leveling construction scenarios.

[0050] Example 4 is essentially the same as Example 3, except that it employs a dual-mode positioning scheme combining RTK-GNSS and BIM-SLAM. A network of 4-6 UWB positioning base stations can be deployed in the construction area, covering a 200m² area. UWB positioning replaces GNSS to achieve seamless indoor and outdoor positioning with full coverage. UWB positioning data is fused with BIM prior constraint SLAM using extended Kalman filtering to further improve positioning reliability in GNSS-denied environments. Furthermore, the leveling path planning strategy can be replaced with a zonal coverage strategy: based on the elevation requirements of different functional zones in the BIM model, the construction area is divided into several sub-zones. Within each sub-zone, a U-shaped coverage path is used, and the shortest connecting path between sub-zones is planned using the A* algorithm to minimize the overall path length.

[0051] Example 5 is basically the same as Example 1, except that step S2 uses the improved RRT algorithm described above to generate the initial global path. The specific steps are as follows: Semantic map initialization: Similar to Example 1, a hierarchical raster map is first constructed to obtain the cost layer; Tree initialization: starting from the robot's initial position. Initialize the RRT tree for the root node ; Iterative expansion: In each iteration, a sample is generated in free space according to the semantic bias sampling strategy. ;exist Searching by weighted distance metric ;along The direction is generated in fixed steps (e.g., 0.3m). Detection If the grid cell is impassable, the expansion is abandoned; if it is a high-cost grid cell, the expansion is decided based on the acceptance probability; the expanded edge must pass the geometric collision detection and must not collide with geometric layer obstacles. Termination condition: when Stop when entering the neighborhood of the target work location (i.e., distance < 0.5m) or when the maximum number of iterations (5000) is reached; Path extraction and optimization: The initial path is obtained by backtracking from the target node to the root node, and then pruning and B-spline smoothing optimization based on BIM semantics are performed to generate the final global path.

[0052] This improved RRT algorithm fully integrates BIM semantic information, effectively overcoming the problems of blind sampling and low path quality in complex construction environments encountered by traditional RRT. In a comparative experiment, in a bricklaying construction scenario involving multiple walls and narrow passages, the improved RRT algorithm, compared to the standard RRT, reduced the path length by approximately 18%, improved path smoothness (measured by the square integral of curvature) by approximately 42%, and increased the planning success rate from 87% to 98%.

[0053] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A method for autonomous path planning and obstacle avoidance of construction robots based on BIM models, characterized in that, Includes the following steps: Step S1: Obtain the BIM model of the target construction area, and extract the three-dimensional geometric information and semantic information from the BIM model. The semantic information includes at least the category and location of permanent structural components, the marked area of ​​movable obstacles, the boundary of the construction work area, and the prohibited passage area. Step S2: Construct a hierarchical grid map based on the three-dimensional geometric and semantic information, wherein different grids are assigned differentiated access cost weights according to the semantic information, grids corresponding to permanent structural components are set as impassable, grids in the working area are set as low access cost, and grids in prohibited areas are set as high access cost; an improved A is applied to the hierarchical grid map. * The algorithm or RRT algorithm generates an initial global path from the starting position to the target job position; Step S3: The construction robot is equipped with a multi-sensor system, which includes at least a lidar, an inertial measurement unit, and a vision camera, and collects environmental data in real time during its movement along the initial global path; Step S4: Using the semantic information of the BIM model as a priori constraint, the real-time collected environmental data is fused together, and the robot's pose is estimated in real time and the local environment map is updated using SLAM technology based on BIM prior constraints. The obstacles detected by the real-time sensors are matched and compared with the BIM semantic information to identify permanent fixed obstacles that have been marked in the BIM model and newly added dynamic / temporary obstacles that have not been marked in the BIM model. Step S5: During the robot's movement, when the newly added dynamic / temporary obstacle is detected, the local path is replanned using the dynamic window method and the artificial potential field method. While maintaining a safe distance from permanent fixed obstacles, the local path is dynamically adjusted to bypass the newly added dynamic / temporary obstacle. At the same time, BIM semantic information is used to optimize the obstacle avoidance path so that it remains within the working area. Step S6: The planned path information is transmitted to the robot control system in real time through the communication module, and the robot control system drives the robot actuator to complete the construction task according to the planned path.

2. The method for autonomous path planning and obstacle avoidance of construction robots based on BIM models according to claim 1, characterized in that, The layered raster map in step S2 includes a geometric layer, a semantic layer, and a cost layer; wherein the geometric layer stores three-dimensional geometric information, the semantic layer stores the semantic labels of each raster, and the cost layer stores the passage cost weights calculated based on the semantic labels; the improved A * The algorithm or RRT algorithm searches on the cost layer.

3. The method for autonomous path planning and obstacle avoidance of construction robots based on BIM models according to claim 1, characterized in that, The improvement A described in step S2 * The evaluation function of the algorithm is: , in For nodes The comprehensive evaluation function value, From the starting node to the node The actual cost, For the node Heuristic cost estimation to the target node The weighting coefficients of the heuristic function are dynamically adjusted based on BIM semantic information. For semantic information-based penalty terms, when node When located in a restricted area or near permanent structural members Take the larger value; The improved RRT algorithm described in step 2 includes: The semantic bias sampling strategy is based on the semantic access cost of each raster. Calculate sampling weights Random sampling points are selected in free space using a roulette wheel method, which makes the sampling tend to favor low-cost working areas; During the expansion of the RRT tree, cost-weighted distance is used for nearest neighbor selection, and the cost of newly generated nodes is evaluated. When a new node falls into a high-cost grid, it is selected with probability. Accept extensions; Where, p i For grid The relative probability weights of those selected as random sampling points; costi For grid The permissible cost value in the hierarchical cost layer; λ This is the sampling bias attenuation coefficient; μ This is the probability decay coefficient for acceptance. The initially generated RRT path is pruned using BIM semantic information to remove redundant nodes, and a smooth path is generated in the working area through B-spline interpolation to ensure that the safe distance from permanent structural components is not less than a preset threshold.

4. The method for autonomous path planning and obstacle avoidance of construction robots based on BIM models according to claim 1, characterized in that, The multi-sensor system described in step S3 further includes an ultra-wideband positioning tag and / or a wheeled odometer, used to provide auxiliary positioning information in scenarios where lidar data is sparse or environmental features are not obvious; the environmental data includes lidar three-dimensional point cloud data, IMU six-axis inertial data, visual image data, and UWB ranging data.

5. The method for autonomous path planning and obstacle avoidance of construction robots based on BIM models according to claim 1, characterized in that, The SLAM technology based on BIM prior constraints described in step S4 specifically includes: using the geometric information of permanent structural components in the BIM model as the initial map prior, using LiDAR-Inertial Odometry for front-end odometry estimation, performing real-time registration of the estimated pose with the geometric features extracted from the BIM model, constructing a factor graph optimization problem containing BIM prior constraints, solving the robot's optimal pose estimation through nonlinear least squares optimization, and using the semantic information in the BIM model to perform semantic annotation and dynamic object filtering on the SLAM-constructed map.

6. The method for autonomous path planning and obstacle avoidance of construction robots based on BIM models according to claim 1, characterized in that, The obstacle avoidance strategy integrating the dynamic window method and the artificial potential field method described in step S5 specifically includes: First, using BIM semantic information to differentiate the repulsive potential field in the artificial potential field method, the high-intensity impassable repulsive field generated by permanent structural components, the traversable medium-intensity repulsive field generated by newly added dynamic / temporary obstacles, and the low-intensity boundary repulsive field generated by the boundary of the work area; then, in the velocity sampling space of the dynamic window method, the potential field gradient of the artificial potential field method is used as a cost index for trajectory evaluation, which together with the target orientation cost, velocity cost, and obstacle distance cost constitutes a comprehensive evaluation function for the velocity command; in each control cycle, the velocity command that optimizes the comprehensive evaluation function is selected from the velocity sampling space to drive the robot's motion.

7. The method for autonomous path planning and obstacle avoidance of construction robots based on BIM models according to claim 1, characterized in that, The construction robot is an autonomous mobile construction robot with a wheeled mobile chassis and an operation execution mechanism. The operation execution mechanism is a bricklaying robot arm, a spraying robot arm, or a ground leveling actuator. The robot control system receives path planning instructions through the robot operating system and converts them into speed control instructions for the robot chassis and action instructions for the operation execution mechanism.

8. A BIM model-based autonomous path planning and obstacle avoidance system for construction robots, which employs the BIM model-based autonomous path planning and obstacle avoidance method for construction robots as described in any one of claims 1-7, characterized in that, include: The BIM model processing module is used to acquire BIM models and extract three-dimensional geometric and semantic information. A global path planning module is used to generate an initial global path based on the three-dimensional geometric information and semantic information; A multi-sensor data acquisition module, mounted on a construction robot, includes at least a lidar, an inertial measurement unit, and a vision camera, for real-time acquisition of environmental data; The BIM sensor fusion positioning and mapping module is used to use the semantic information of the BIM model as a priori constraints, fuse the real-time environmental data, and use SLAM technology based on BIM prior constraints to estimate the robot pose and update the local environmental map in real time. The dynamic obstacle avoidance and path adjustment module is used to replan the local path using the dynamic window method and the artificial potential field method when new dynamic / temporary obstacles are detected, and to optimize the obstacle avoidance path by combining BIM semantic information. The communication and control module is used to transmit the planned path information to the robot control system to drive the robot actuators.

9. The BIM model-based autonomous path planning and obstacle avoidance system for construction robots according to claim 8, characterized in that, The BIM sensor fusion positioning and mapping module further includes: The front-end odometry submodule is used for front-end pose estimation based on lidar point cloud data and IMU inertial data; The BIM registration submodule is used to perform real-time scanning and matching of the estimated pose with the geometric features in the BIM model. The factor graph optimization submodule is used to construct a factor graph containing BIM prior constraints and solve for the optimal pose through nonlinear least squares optimization. The semantic annotation submodule is used to perform semantic annotation and dynamic object filtering on the SLAM-built map using semantic information from the BIM model.

10. The BIM model-based autonomous path planning and obstacle avoidance system for construction robots according to claim 8, characterized in that, The dynamic obstacle avoidance and path adjustment module further includes: The artificial potential field calculation submodule is used to calculate differentiated gravitational and repulsive potential fields for different types of obstacles and target points based on BIM semantic information; The dynamic window velocity sampling submodule is used to sample candidate velocity commands in the velocity space under robot kinematic constraints; The trajectory evaluation submodule is used to score candidate trajectories by comprehensively considering the target orientation, obstacle distance, potential field gradient, and velocity smoothness, and select the optimal velocity command.

Images