High-precision laser SLAM navigation method and system of heavy AGV

Abstract

The invention relates to the technical field of AGV navigation, in particular to a high-precision laser SLAM navigation method and system for a heavy AGV. The method comprises the following steps: continuously collecting factory panoramic monitoring images based on a high-definition camera, and carrying out terrain obstacle segmentation to generate a plurality of obstacle collision constraint frames; performing three-dimensional scene structure identification according to the factory panoramic monitoring image, performing panoramic scene mapping based on a plurality of obstacle collision constraint frames, and constructing a panoramic scene structure map; calculating the current position coordinate of the AGV, carrying out intelligent AGV navigation planning according to the panoramic scene structure map, and constructing an intelligent AGV planning path; laser sensing feedback parameters are collected in real time based on a laser radar, local building structure synchronization processing is performed on a panoramic scene structure map, and a synchronous optimization scene map is constructed. According to the invention, the navigation precision of the AGV is improved, and an efficient and safe real-time path is planned through dynamic feedback information.

Description

Technical Field

[0001] The present invention relates to the technical field of AGV navigation, and particularly to a high-precision laser SLAM navigation method and system for heavy-duty AGVs. Background Art

[0002] With the rapid development of industrial automation, intelligent manufacturing, and the logistics industry, heavy-duty automated guided vehicles (AGVs), as efficient and reliable automated handling tools, have been widely used in various fields such as warehousing, production lines, and transportation. Heavy-duty AGVs can perform high-load and high-efficiency operations in complex environments through automated path planning and autonomous navigation. Especially in the fields of modern warehousing logistics, manufacturing, and heavy material handling, the application of AGVs has gradually replaced traditional manual handling, greatly improving production efficiency and reducing labor costs. However, in the process of applying heavy-duty AGVs, how to achieve high-precision and high-efficiency navigation in complex environments has become an urgent technical problem to be solved.

[0003] Traditional heavy-duty AGV navigation systems usually rely on technologies such as lidar, magnetic strips, vision, or inertial sensors for positioning and navigation. However, due to the dynamic changes in the environment, the unpredictability of obstacles, and the complexity of the scene, these traditional navigation methods are often affected by factors such as positioning errors, environmental interference, and sensor failures when facing complex environments, resulting in problems such as insufficient navigation accuracy, unstable path planning, and untimely obstacle avoidance. These problems not only affect the efficiency of AGVs but also increase the risks of equipment failures and safety hazards. Therefore, how to ensure the stability and safety of AGVs in complex environments while maintaining high-precision navigation has become a key research direction. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a high-precision laser SLAM navigation method and system for heavy-duty AGVs to solve at least one of the above technical problems.

[0005] To achieve the above object, the present invention provides a high-precision laser SLAM navigation method for heavy-duty AGVs. The heavy-duty AGV is equipped with a high-definition camera and a lidar, and the method includes the following steps: Step S1: Continuously collect panoramic monitoring images of the factory based on the high-definition camera, and perform terrain obstacle segmentation to generate multiple obstacle collision constraint boxes; Step S2: Perform three-dimensional scene structure recognition based on the panoramic monitoring images of the factory, and build a panoramic scene map based on multiple obstacle collision constraint boxes to construct a panoramic scene structure map; Step S3: Calculate the current position coordinates of the AGV, perform intelligent AGV navigation planning based on the panoramic scene structure map, and construct an intelligent AGV planned path; Step S4: Based on the lidar, the laser sensing feedback parameters are collected in real time, and the local building structure of the panoramic scene structure map is synchronously processed to construct a synchronized and optimized scene map; Step S5: According to the synchronized and optimized scene map, the maximum passing form limit analysis and scene structure protrusion collision avoidance of the intelligent AGV planned path are carried out to generate a collision avoidance optimized path; Step S6: According to the laser sensing feedback parameters, the dynamic obstacle time-series movement is tracked, and the collision avoidance optimized path is adaptively iteratively optimized to construct an intelligent navigation optimization model.

[0006] The present invention continuously collects panoramic images through a high-definition camera, enabling the AGV to comprehensively perceive the factory environment. This provides the AGV with rich visual information, ensuring that it can accurately identify and segment various terrain obstacles, including fixed buildings, equipment, and other static objects. The terrain obstacle segmentation technology helps extract the shape and position of each obstacle from the panoramic image and generates multiple collision bounding boxes (such as cylinders, rectangular boxes, etc.) based on this information. Such collision bounding boxes provide precise geometric information for subsequent path planning and obstacle avoidance. The high-definition camera can capture real-time environmental changes (such as temporarily placed obstacles), enabling the AGV to dynamically adapt to new obstacles and environmental changes, improving the flexibility and stability of navigation. Through in-depth analysis of the panoramic image, the AGV can identify the three-dimensional structure of the factory environment, such as important features like buildings, walls, doors, and windows. This provides accurate spatial location-based information for subsequent positioning and navigation. Based on the combination of the obstacle collision bounding boxes and the three-dimensional scene structure, the AGV can generate a high-precision panoramic scene map. This map not only provides the spatial information of all obstacles in the environment but also includes detailed information about the factory layout and structure, helping the AGV to perform global positioning in a complex environment. The panoramic scene map incorporates all the important features in the factory, providing rich environmental data for the path planning system and enabling it to better guide the AGV to select the optimal path. By combining with the panoramic scene structure map, the AGV can calculate its accurate position in the factory in real time, ensuring high-precision positioning. This is crucial for the precise navigation of heavy AGVs in complex environments. Based on the current position and the panoramic scene map, the AGV can perform intelligent navigation planning. By applying advanced path planning algorithms (such as A*, D*, RRT, etc.), the AGV can plan the optimal driving path according to the real-time environment and task requirements, avoiding obstacles and other potential risks. The navigation planning system can perform real-time path adjustment according to changes in the factory (such as sudden obstacles, production area adjustments, etc.), thereby improving the flexibility and safety of the AGV. The lidar can provide high-precision environmental feedback data, including the distance, size, and shape of obstacles. By real-time collecting the feedback parameters of the laser sensor, the AGV can promptly identify new obstacles or environmental changes. Based on the lidar data, the AGV can perform local synchronization and optimization of the panoramic scene map, ensuring that the map always remains up-to-date and accurate. This is crucial for ensuring the accuracy and stability of path planning, especially in a dynamic environment. The combination of the lidar and the vision system provides a more comprehensive environmental perception ability, enabling the AGV to more accurately identify and avoid obstacles in dynamic and complex environments. Through detailed analysis of the AGV's form and the environment, the AGV can analyze the passage restrictions in the path according to its own size and shape. This enables the AGV to select paths that meet its size requirements and avoid narrow spaces or areas that are not suitable for passage.The AGV can identify possible scene structure protrusions (such as platform edges, brackets, etc.) on the synchronized and optimized scene map and avoid them. This not only avoids collisions but also ensures the smooth travel of the AGV in complex terrains. Based on morphological constraints and protrusion avoidance, the AGV can generate an optimized path. This path avoids areas that are not suitable for passage and takes into account the movement characteristics of the AGV, providing a safer and more efficient navigation solution. By continuously monitoring and tracking dynamic obstacles (such as moving AGVs, staff, or material handling equipment, etc.), the AGV can predict the future trajectories of these obstacles. Through time series analysis and trajectory prediction, the AGV can make preparations for avoidance in advance and avoid collisions with these dynamic obstacles. Based on the real-time feedback of dynamic obstacles, the AGV can adaptively adjust the path. As the environment changes (new obstacles appear, target positions change, etc.), the path planning of the AGV will be optimized in real time, making the navigation more intelligent and flexible. By continuously iterating and optimizing the path, the AGV can build a highly adaptive intelligent navigation model. This enables the AGV to continuously operate in a changing environment and adjust its behavior according to new environmental data to ensure the successful completion of tasks.

[0007] In this specification, a high-precision laser SLAM navigation system for a heavy-duty AGV is provided, which is used to execute the high-precision laser SLAM navigation method for the heavy-duty AGV as described above, including: A terrain recognition module, which is used to continuously collect panoramic monitoring images of the factory based on a high-definition camera, perform terrain obstacle segmentation, and generate multiple obstacle collision constraint boxes; A panoramic mapping module, which is used to perform three-dimensional scene structure recognition based on the panoramic monitoring images of the factory, and perform panoramic scene mapping based on multiple obstacle collision constraint boxes to construct a panoramic scene structure map; A path planning module, which is used to calculate the current position coordinates of the AGV, perform intelligent AGV navigation planning based on the panoramic scene structure map, and construct an intelligent AGV planned path; A local synchronization optimization module, which is used to continuously collect laser sensing feedback parameters based on a lidar and perform local building structure synchronization processing on the panoramic scene structure map to construct a synchronized and optimized scene map; A passage restriction analysis module, which is used to perform maximum passage form restriction analysis and scene structure protrusion collision avoidance on the intelligent AGV planned path according to the synchronized and optimized scene map, and generate a collision avoidance optimized path; A dynamic obstacle avoidance module, which is used to perform dynamic obstacle time series movement tracking according to the laser sensing feedback parameters, and perform adaptive path iteration optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

[0008] The present invention can provide high-resolution image data through a high-definition camera, enabling the AGV to more clearly identify static and dynamic obstacles in the surrounding environment. This enables the AGV to maintain a high level of stability and accuracy when facing complex or changing environments. By segmenting terrain obstacles from the image data, the AGV can accurately identify potential collision obstacles and clearly define the spatial scope of these obstacles by generating collision bounding boxes for multiple obstacles. This provides an effective reference for subsequent path planning and collision avoidance. By identifying and reconstructing the three-dimensional structure of the scene, the AGV can obtain a more comprehensive and detailed environmental model. This three-dimensional mapping can not only show the spatial layout of all obstacles and building structures in the factory, but also improve the navigation accuracy of the AGV in complex environments. The panoramic mapping based on the obstacle collision bounding boxes can not only depict the static environment, but also provide a complete view containing information such as the position, shape, and size of environmental obstacles. In this way, the AGV can clearly understand the distribution of obstacles in the entire environment during path planning and provide an accurate reference for intelligent navigation. By combining the panoramic scene structure map and the current position of the AGV, the system can accurately calculate the coordinates of the AGV and generate an optimal planned path based on the real-time map. This ensures that the AGV can move forward stably in complex environments and avoid problems such as incorrect positioning or being unable to pass through narrow paths. The intelligent navigation planning based on the panoramic scene map can dynamically adjust the path to adapt to new obstacles or changes in the environment. This intelligent planning can significantly improve the autonomous navigation ability of the AGV and enhance its flexibility and adaptability in a changing environment. The high-precision distance information provided by lidar combined with the visual data of high-definition images can accurately correct the deviation in environmental mapping. By synchronously processing the lidar data with the panoramic scene map, high-precision optimization of the building structure in the local area can be achieved, ensuring that the AGV can obtain the latest scene information. As the AGV moves, the obstacles and structures in the environment may change. By real-time synchronously optimizing the scene map, the AGV can ensure that the navigation path always remains consistent with the actual environment, thereby avoiding navigation errors caused by outdated map information. By analyzing the passage restrictions of the AGV (such as turning radius, load capacity, etc.), a path that can maximize the use of its driving space can be generated for the AGV, thereby optimizing the driving efficiency and safety. This enables the AGV to select the best path in different environments according to its specific physical form and work tasks. By analyzing the protrusions or obstacles in the synchronously optimized scene map, the AGV can intelligently avoid collisions. The collision avoidance optimized path can be adjusted in real time according to the obstacle layout around the AGV, providing a safer driving route for the AGV, which is particularly important in narrow spaces or complex factory environments. By real-time tracking the movement trajectory of dynamic obstacles, the AGV can predict their future positions and make avoidance decisions in advance.This is particularly important for dealing with occasional dynamic obstacles in the factory environment, such as the movement of personnel or equipment. Based on the laser sensing feedback parameters, the AGV can adjust its path in real time according to the dynamic changes of the obstacles during driving. Adaptive path optimization can significantly improve the navigation ability of the AGV, ensuring that it can still operate efficiently in the face of unforeseen obstacles or environmental changes. Combining the above dynamic feedback and path optimization strategies, the AGV can continuously learn and improve its navigation decisions, generating more efficient and safe navigation strategies. This optimization model can not only improve the navigation accuracy of the AGV, but also make it more adaptable to the changing and complex environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a schematic diagram of the step flow of a high-precision laser SLAM navigation method for a heavy-duty AGV of the present invention; Figure 2 It is a schematic diagram of the detailed implementation steps of step S1; Figure 3 It is a schematic diagram of the detailed implementation steps of step S2; Figure 4 It is a schematic diagram of the detailed implementation steps of step S3. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0010] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0011] The embodiments of the present application provide a high-precision laser SLAM navigation method and system for a heavy-duty AGV. The execution subjects of the high-precision laser SLAM navigation method and system for the heavy-duty AGV include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc. that carry this system, which can be regarded as general computing nodes of the present application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0012] Please refer to Figures 1 to 4 , the present invention provides a high-precision laser SLAM navigation method for a heavy-duty AGV, and the high-precision laser SLAM navigation method for the heavy-duty AGV includes the following steps: Step S1: Continuously collect panoramic monitoring images of the factory based on a high-definition camera, and perform terrain obstacle segmentation to generate multiple obstacle collision bounding boxes; Step S2: Perform three-dimensional scene structure recognition based on the panoramic monitoring images of the factory, and perform panoramic scene mapping based on multiple obstacle collision bounding boxes to construct a panoramic scene structure map; Step S3: Calculate the current position coordinates of the AGV, perform intelligent AGV navigation planning based on the panoramic scene structure map, and construct an intelligent AGV planned path; Step S4: Based on the lidar, the laser sensing feedback parameters are collected in real time, and local building structure synchronization processing is performed on the panoramic scene structure map to construct a synchronized and optimized scene map; Step S5: According to the synchronized and optimized scene map, a maximum passage form limit analysis and scene structure protrusion collision avoidance are performed on the planned path of the intelligent AGV to generate a collision avoidance optimized path; Step S6: Perform dynamic obstacle time-series movement tracking according to the laser sensing feedback parameters, and perform adaptive path iterative optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

[0013] The present invention continuously captures panoramic images through a high-definition camera, enabling the AGV to comprehensively perceive the factory environment. This provides the AGV with rich visual information, ensuring that it can accurately identify and segment various terrain obstacles, including fixed buildings, equipment, and other static objects. The terrain obstacle segmentation technology helps extract the shape and position of each obstacle from the panoramic image and generate multiple collision constraint boxes (such as cylinders, rectangular boxes, etc.) based on this information. Such collision constraint boxes provide precise geometric information for subsequent path planning and obstacle avoidance. The high-definition camera can capture real-time environmental changes (such as temporarily placed obstacles), enabling the AGV to dynamically adapt to new obstacles and environmental changes, improving the flexibility and stability of navigation. Through in-depth analysis of the panoramic image, the AGV can identify the three-dimensional structure of the factory environment, such as important features like buildings, walls, doors, and windows. This provides accurate spatial position-based information for subsequent positioning and navigation. Based on the combination of the obstacle collision constraint boxes and the three-dimensional scene structure, the AGV can generate a high-precision panoramic scene map. This map not only provides the spatial information of all obstacles in the environment but also includes detailed information on the factory layout and structure, helping the AGV perform global positioning in a complex environment. The panoramic scene map incorporates all the important features in the factory, providing rich environmental data for the path planning system and enabling it to better guide the AGV to select the optimal path. By combining with the panoramic scene structure map, the AGV can calculate its accurate position in the factory in real time, thus ensuring high-precision positioning. This is crucial for the precise navigation of heavy AGVs in complex environments. Based on the current position and the panoramic scene map, the AGV can perform intelligent navigation planning. By applying advanced path planning algorithms (such as A*, D*, RRT, etc.), the AGV can plan the optimal driving path according to the real-time environment and task requirements, avoiding obstacles and other potential risks. The navigation planning system can perform real-time path adjustment according to changes in the factory (such as sudden obstacles, adjustments in production areas, etc.), thereby improving the flexibility and safety of the AGV. The lidar can provide high-precision environmental feedback data, including the distance, size, and shape of obstacles. By real-time collecting the feedback parameters of the laser sensor, the AGV can promptly identify new obstacles or environmental changes. Based on the lidar data, the AGV can perform local synchronization and optimization of the panoramic scene map to ensure that the map always remains up-to-date and accurate. This is crucial for ensuring the accuracy and stability of path planning, especially in a dynamic environment. The combination of lidar and the vision system provides a more comprehensive environmental perception ability, enabling the AGV to more accurately identify and avoid obstacles in dynamic and complex environments. Through detailed analysis of the AGV's form and the environment, the AGV can analyze the passage restrictions in the path according to its own size and shape. This enables the AGV to select a path that meets its size requirements and avoid narrow spaces or areas that are not suitable for passage.The AGV can identify possible scene structure protrusions (such as platform edges, brackets, etc.) on the synchronized and optimized scene map and avoid them. This not only prevents collisions but also ensures the smooth operation of the AGV in complex terrains. Based on morphological constraints and protrusion avoidance, the AGV can generate an optimized path. This path avoids areas that are not suitable for passage and takes into account the movement characteristics of the AGV, providing a safer and more efficient navigation solution. By continuously monitoring and tracking dynamic obstacles (such as moving AGVs, staff, or material handling equipment, etc.), the AGV can predict the future trajectories of these obstacles. Through time series analysis and trajectory prediction, the AGV can make preparations for avoidance in advance and avoid collisions with these dynamic obstacles. Based on the real-time feedback of dynamic obstacles, the AGV can adaptively adjust the path. As the environment changes (new obstacles appear, the target position changes, etc.), the path planning of the AGV will be optimized in real time, making the navigation more intelligent and flexible. By continuously iterating and optimizing the path, the AGV can build a highly adaptive intelligent navigation model. This enables the AGV to continuously operate in a changing environment and adjust its behavior according to new environmental data to ensure the successful completion of tasks.

[0014] In the embodiments of the present invention, refer to Figure 1 , which is a schematic diagram of the step flow of a high-precision laser SLAM navigation method for a heavy-duty AGV of the present invention. In this example, the steps of the high-precision laser SLAM navigation method for the heavy-duty AGV include: Step S1: Continuously collect panoramic monitoring images of the factory based on a high-definition camera, and perform terrain obstacle segmentation to generate multiple obstacle collision bounding boxes; In this embodiment, a suitable high-definition camera is selected, usually a camera with a high resolution (such as 4K or higher) to ensure image quality. A wide-angle lens is selected to monitor a larger area simultaneously. A camera with a 4K resolution and a 120-degree viewing angle is selected to ensure coverage of the main areas of the entire factory. Such devices can provide images that are clear enough to facilitate subsequent obstacle recognition. The camera is installed at a high position in the factory to ensure that the field of view can cover the ground and the obstacles around it. During installation, it is necessary to avoid occlusion and blind spots, and it is recommended to use multiple cameras to achieve overlapping coverage. The camera height is set to 4 meters, and the installation location is selected on the ceiling in the center of the factory to ensure that the entire working area can be overlooked and any obstacle occlusion is avoided. Adjust the camera parameters according to the lighting conditions in the factory, including exposure time, white balance, and ISO value, etc., to ensure clear images can be obtained under different lighting conditions. The exposure time is set to 1 / 60 second, and the ISO value is set to 400 to adapt to the light changes in the factory while ensuring image clarity. The camera is set to collect panoramic monitoring images in continuous mode, usually at a frequency of 30 frames per second for image acquisition. This frequency can ensure the capture of dynamic environmental information. The monitoring period is set to collect one frame of image every 5 seconds, and the images are stored in a local server or the cloud to ensure data security and accessibility. The collected images are preprocessed, including denoising, enhancing contrast, and adjusting brightness, to improve the effect of subsequent obstacle segmentation. Image processing techniques (such as median filtering) are used to remove noise. The window size of the denoising algorithm is set to 3x3 pixels to effectively remove random noise in the image and ensure the accuracy of the segmentation effect. A semantic segmentation algorithm (such as FCN, U-Net, or Mask R-CNN) is used to segment topographic obstacles in the image. These algorithms can effectively identify and label different objects in the image. The U-Net model is selected for training, and a labeled dataset containing common obstacles (such as machines, goods, etc.) in the factory environment is used for model training to ensure that the model can accurately identify different types of obstacles. The trained segmentation model is applied to the collected images to generate a binary segmentation map of the obstacles, identifying the areas of all obstacles in the image. Ensure the accuracy and integrity of the segmentation results. In the generated binary map, the background is 0, and the obstacle area is 1. Small artifacts are removed through post-processing to ensure that only actual obstacles are retained. For the segmented obstacle areas, the minimum bounding rectangle or bounding box method is used to generate collision constraint boxes. These constraint boxes will be used for subsequent path planning and collision detection. The minimum bounding rectangle algorithm is selected to ensure that the constraint box can tightly enclose each obstacle as much as possible, improving the safety of path planning. Traverse the segmented obstacle areas, generate corresponding collision constraint boxes for each obstacle, and record the coordinate and size information of the boxes. Ensure that each constraint box can effectively reflect the spatial occupancy of the obstacle.If the segmentation area of a certain obstacle is (50, 50, 100, 100), the coordinates of the generated bounding box are the upper left corner (50, 50) and the lower right corner (100, 100).

[0015] Step S2: Perform three-dimensional scene structure recognition based on the panoramic monitoring image of the factory, and construct a panoramic scene map based on multiple obstacle collision bounding boxes to build a panoramic scene structure map; In this embodiment, collect the previously acquired panoramic monitoring images of the factory, ensuring that the image quality is high enough for subsequent 3D structure recognition. The images should be in a standardized format (such as JPEG or PNG) and have been preprocessed to remove noise. Select at least 30 panoramic images from different angles to obtain sufficient perspective information and ensure the accuracy of subsequent 3D reconstruction. Use feature point detection algorithms (such as SIFT, SURF, or ORB) to extract features from each monitoring image. The extracted feature points are used for subsequent 3D reconstruction to ensure the repeatability and stability of the feature points. Set at least 500 feature points to be extracted from each image, and use the FLANN (Fast Library for Approximate Nearest Neighbors) algorithm for feature matching to improve the efficiency and accuracy of the matching. Adopt multi-view stereo technology (MVS) or structure from motion (SfM) algorithms for 3D reconstruction. These algorithms can reconstruct the 3D structure of the factory based on the camera positions and the matching information of the feature points. Select to use COLMAP as the 3D reconstruction tool and set the minimum reprojection error to 1 pixel to ensure the accuracy of the reconstruction result. Input the matched feature points and camera position information into the 3D reconstruction algorithm to generate a 3D point cloud model of the factory. This process requires optimizing the reconstruction result, removing outliers, and improving the quality of the model. The generated point cloud should contain at least 50,000 points, and use processing methods such as filtering and downsampling to ensure the clarity and usability of the point cloud model. Collect the information of multiple previously generated obstacle collision bounding boxes, including the position information, size, and type of each obstacle. This information will be used for subsequent scene mapping to ensure the accuracy of the model. Record the coordinates, width, and height of the bounding boxes of each obstacle for accurate geometric matching during scene mapping. Combine the obstacle collision bounding box information with the generated 3D point cloud model. By comparing the point cloud data with the bounding boxes, identify the static obstacles in the scene to ensure that the 3D model can accurately reflect the actual situation of the factory environment. Ensure that the bounding box of each obstacle can cover the corresponding point cloud area in the 3D model to verify the position and shape of the obstacle. Adopt a graph-based panoramic scene construction method to combine the 3D point cloud with the obstacle bounding boxes to generate a complete panoramic scene map. This method can effectively fuse data from different sources to ensure the accuracy and consistency of the map. Use 3D mapping frameworks such as Octomap, and set the voxel size to 0.1 meters to ensure that the resolution of the map meets the actual requirements. Construct a complete panoramic scene map based on the information of the 3D point cloud and the obstacle bounding boxes. This process requires optimizing the map to reduce noise and redundant data to ensure the clarity of the map. The generated panoramic scene map should contain all key feature points and ensure that each static obstacle can be accurately represented in the map.

[0016] Step S3: Calculate the current position coordinates of the AGV, perform intelligent AGV navigation planning based on the panoramic scene structure map, and construct an intelligent AGV planned path; In this embodiment, sensors installed on the AGV (such as lidar, IMU, and wheel speedometer) are used to collect the motion state and surrounding environment information of the AGV in real time. The sensor data should include the position, speed, acceleration of the AGV, and the angle information of the sensors. Set the data collection frequency of the lidar to 10 Hz and the sampling frequency of the IMU to 100 Hz to obtain the dynamic state of the AGV and the positions of surrounding obstacles in real time. A positioning method based on sensor fusion technology, such as Extended Kalman Filter (EKF) or Particle Filter (PF), is adopted to calculate the current position of the AGV by combining multiple sensor data. Using the EKF method, the ranging data of the lidar is fused with the angular velocity and acceleration data of the IMU to improve the accuracy and reliability of position estimation. The position information of the AGV is updated in real time through the sensor fusion algorithm. Set each time step to 0.1 seconds, and use the sensor data to calculate the current position coordinates (x, y, θ) of the AGV, where θ is the orientation angle of the AGV. After a period of time, assuming the current calculated position of the AGV is (4.5, 2.3, 90°), it should be ensured that this coordinate can accurately reflect the position of the AGV in the panoramic scene structure map. Ensure that the latest version of the panoramic scene structure map is available. The map should include the collision bounding boxes of all obstacles and their relative positions. At the same time, the map should be optimized to remove redundant information. Use a grid map with a resolution of 0.1 meters to ensure that the map accuracy is high enough to accurately reflect the factory environment. Select a suitable path planning algorithm, such as the A* algorithm, Dijkstra algorithm, or RRT (Rapidly-Exploring Random Tree) algorithm. The selected algorithm should be able to effectively handle obstacles in complex environments to ensure that the AGV can reach the target position safely and efficiently. Select the A* algorithm and set the heuristic function to the Euclidean distance to improve the efficiency and accuracy of path search. On the panoramic scene structure map, perform path planning based on the current position and target position of the AGV. First, set the coordinates of the target position. For example, the target position is (8.0, 5.0). Then, calculate the optimal path from the current position of the AGV to the target position through the path planning algorithm. Use the A* algorithm to calculate the path from (4.5, 2.3) to (8.0, 5.0). The generated path should avoid all obstacles and ensure a safe distance. Further optimize the generated preliminary navigation path to ensure the smoothness and feasibility of the path. A path smoothing algorithm (such as Bezier curve smoothing) can be used to reduce sharp turns in the path. Check the generated path to ensure that the total path length does not exceed 1.2 times the target distance, and each turning angle in the path does not exceed 30 degrees to ensure the stability of the AGV's travel.

[0017] Step S4: Based on the lidar, collect the laser sensing feedback parameters in real time, and perform local building structure synchronization processing on the panoramic scene structure map to construct a synchronized and optimized scene map; In this embodiment, a lidar sensor is installed on the AGV to ensure that it can scan the surrounding environment 360 degrees. The lidar should have high resolution and a large ranging scope to accurately collect data of surrounding buildings and obstacles. Select a lidar with a ranging scope of 0.2 meters to 12 meters and set its operating frequency to 10 Hz to ensure that 10 frames of data can be collected per second to capture rapidly changing environmental information. Start the lidar for real-time data collection and record the point cloud data returned by the lidar sensor. This point cloud data will provide basic information for subsequent building structure synchronization processing. Set the point cloud density of each frame of data to 1000 points per square meter to ensure that the data has sufficient details to support the accurate reconstruction of local building structures. Preprocess the collected point cloud data, including denoising, filtering, and downsampling, to improve the data quality. Remove environmental noise and redundant points to ensure the accuracy of subsequent analysis. Apply the Statistical Outlier Removal method to remove points whose distance exceeds 2 times the standard deviation from the mean to ensure the reliability of the point cloud data. Extract the geometric features of local buildings from the preprocessed point cloud data, including walls, columns, doors, and windows. These features will be used to match and synchronize with the panoramic scene structure map. Use the RANSAC algorithm to detect plane features and set the maximum distance for plane fitting to 0.01 meters to ensure that the extracted building features have high precision. Match the extracted local building features with the panoramic scene structure map to identify changes and new structures in the area. This step will ensure that the map can reflect the latest state of the environment in real time. Use the ICP (Iterative Closest Point) algorithm for point cloud alignment, compare the point cloud of the local building with the corresponding part in the panoramic map, set the maximum number of iterations to 50 times, and the convergence threshold to 0.01 meters to ensure the accuracy of the matching. Synchronize the changes found after matching and update the building structure information in the panoramic scene structure map. This may include adding newly identified building elements or adjusting the positions and shapes of existing elements. If a new wall is detected by the local feature, then update the information and position of the wall to the panoramic map to ensure the accuracy and integrity of the map. Integrate the updated building structure information into the panoramic scene structure map to generate a synchronized and optimized scene map. Ensure that the map contains all the latest building features and obstacle information. The generated map should accurately reflect the building layout after synchronization processing and ensure consistency with the real-time lidar data. Verify the synchronized and optimized scene map to ensure the accuracy and consistency of the new information. The effectiveness of the map can be verified by on-site inspection or comparison with previous map data. Conduct on-site inspection to ensure that the newly added walls or obstacles match the actual environment and ensure that the accuracy of the map is within the allowable error range (such as ±5 cm).

[0018] Step S5: Perform maximum passing form limit analysis and scene structure protrusion collision avoidance on the planned path of the intelligent AGV according to the synchronized optimized scene map, and generate a collision avoidance optimized path; In this embodiment, the key form parameters of the AGV are defined, including its length, width, height, turning radius, and center of gravity position. These parameters will be used to evaluate the passing ability of the AGV on a specific path. Set the length of the AGV to 1.2 meters, the width to 0.8 meters, and the turning radius to 0.6 meters to ensure that these parameters can accurately reflect the physical characteristics of the AGV in different environments. Based on the form parameters of the AGV, establish a passing form constraint model to calculate the minimum passing width and height required by the AGV during movement. The passing width should include the width of the AGV itself and the additional space required for its turning. On the synchronized optimized scene map, analyze each key point of the planned path of the AGV and check whether the surrounding space meets the requirements of the minimum passing form. Compare the distance between the path and the obstacles to ensure that the AGV will not collide with the obstacles during movement. At each key point on the path, calculate the distance between the AGV and the adjacent obstacles. If the distance is less than the set safety distance (such as 0.5 meters), mark this path point as high risk. By analyzing the terrain features in the synchronized optimized scene map, identify potential scene structure protrusions. These protrusions may affect the passing of the AGV and need special treatment. Set the protrusion height threshold to 0.1 meters. If the height change in a certain area of the map exceeds this value, mark it as a protrusion. For the identified protrusions and high-risk path points, formulate corresponding collision avoidance strategies. Potential collisions can be avoided by adjusting the geometric shape of the path or selecting an alternative path. Set that if the risk assessment value of a path point is high, or there are protrusions around it, adjust the path to ensure that the AGV can pass safely. Use path optimization algorithms (such as A* algorithm, Dijkstra algorithm, or RRT algorithm) for path adjustment to ensure that the generated path can avoid all high-risk areas and scene protrusions. Select the A* algorithm and set the heuristic function to the Euclidean distance to improve the efficiency of path search while ensuring the safety of the path. On the panoramic scene structure map, plan the collision avoidance path according to the current position and target position of the AGV. The path should avoid all identified obstacles and protrusions while meeting the passing form limit of the AGV. If the target position of the AGV is (8.0, 5.0) and there is a protrusion in a certain area of the path, the algorithm will recalculate the path to ensure that the AGV can safely bypass the protrusion.

[0019] Step S6: Perform dynamic obstacle time-series movement tracking according to the laser sensing feedback parameters, and perform adaptive path iteration optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

[0020] In this embodiment, the lidar installed on the AGV is used to collect the point cloud data of the surrounding environment in real time to identify dynamic obstacles. These dynamic obstacles may be moving devices, personnel, or other AGVs, etc. During the collection process, ensure that the scanning frequency of the lidar is high enough to capture fast-moving objects. Set the working frequency of the lidar to 10 Hz to ensure that 10 frames of point cloud data can be obtained per second. By analyzing each frame of data, information such as distance and speed is extracted. A dynamic obstacle recognition algorithm based on point cloud is adopted, and the time series analysis method is used to compare the point cloud data at different time points to identify objects with significant position changes. The optical flow method or object detection algorithms based on deep learning (such as YOLO or SSD) can be used for the detection of dynamic objects. The optical flow method is used to calculate the motion vector between adjacent frames, and by setting a threshold (such as 5 cm / s), it is determined whether an object is a dynamic obstacle. The temporal trajectory of the identified dynamic obstacles is recorded, and the position and speed information of each dynamic obstacle at different time points are recorded. This information will be used for subsequent path planning and dynamic avoidance. If the positions of a certain dynamic obstacle in 5 consecutive frames are (5.0, 3.0), (5.1, 3.1), (5.3, 3.3), (5.5, 3.5), (5.7, 3.8) respectively, then its trajectory is recorded, and its moving speed and direction are calculated. Based on the position information and motion trajectory of the dynamic obstacles, an adaptive path iterative optimization model is constructed. This model needs to update the path of the AGV in real time to adapt to the changes in the dynamic environment. Set the time window for path optimization to 5 seconds, and within this time, the current position of the AGV, the target position, and the position of the dynamic obstacles are evaluated in real time. When the planned path of the AGV intersects with the predicted trajectory of the dynamic obstacle, a path replanning strategy is implemented. By recalculating the path of the AGV, ensure that it can safely avoid the dynamic obstacle. If the current path of the AGV intersects with the predicted path of the dynamic obstacle at (6.0, 4.0), use the A* algorithm to replan the path to ensure that the AGV can find a safe alternative path. During the driving process of the AGV, the position and speed of the dynamic obstacles are monitored in real time, and the path is iteratively optimized according to the latest data. Path optimization should consider the motion model and dynamic constraints of the AGV to ensure the feasibility and safety of the path. Set the frequency of path optimization to once every 0.5 seconds. In each iteration, calculate the optimal path of the AGV to the target position and update the path information in real time. Smooth the replanned path to ensure the continuity and naturalness of the path. Use path smoothing algorithms (such as Bezier curves or spline curves) to reduce sharp turns in the path and improve the driving stability of the AGV. Set the number of control points for path smoothing to 5 to ensure that the generated smooth path can reduce unnecessary turns and accelerations during actual driving.

[0021] In this embodiment, refer to Figure 2, which is a schematic diagram of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Continuously collect panoramic monitoring images of the factory based on a high-definition camera; Calculate the internal and external calibration parameters of the high-definition camera, and perform image distortion correction on the panoramic monitoring images of the factory according to the internal and external calibration parameters to obtain distortion-corrected monitoring images; Perform super-resolution enhancement on the distortion-corrected monitoring images to obtain super-resolution optimized monitoring images; Perform omnidirectional visual semantic recognition on the super-resolution optimized monitoring images and mark the static terrain obstacles in the scene; Calculate the geometric shape parameters of the static terrain obstacles in the scene; Perform adaptive bounding box segmentation according to the geometric shape parameters to generate multiple obstacle collision bounding boxes.

[0022] In this embodiment, a suitable high-definition camera is selected, usually a camera with high resolution (such as 4K or higher) and wide viewing angle, to ensure that the entire factory area can be covered. The installation location should consider the field of view coverage of the camera to avoid blind spots. Multiple cameras are installed at high places in the factory to facilitate obtaining panoramic images from different angles and ensure that every corner can be monitored. Set the image acquisition parameters, including frame rate, exposure time, and white balance, etc., to adapt to the lighting conditions in the factory. Usually, set the frame rate to 30 FPS and the exposure time to 1 / 60 second to ensure clear and stable images. Use automatic white balance settings to automatically adjust under different lighting conditions and ensure image quality. Adopt continuous acquisition mode, and store the monitoring images in real time to a local server or the cloud. Use a compression algorithm (such as H.264) to reduce storage requirements while ensuring that the image quality remains within an acceptable range. Set to acquire and save one frame of image every 5 seconds, and record the timestamp for subsequent analysis and processing. Adopt the checkerboard calibration method or Zhang Zhengyou calibration method to calculate the internal and external calibration parameters of the camera. This process includes obtaining multiple checkerboard images at different angles and using these images for calibration. Use a standard 9x6 checkerboard, and each image should contain at least 20 valid corner points to ensure the accuracy of calibration. Calculate the internal parameters (focal length, principal point position, and distortion coefficients) and external parameters (rotation matrix and translation vector) through a calibration tool (such as OpenCV). Ensure that the reprojection error of the calculation result is less than 1 pixel to guarantee the accuracy of calibration. The calculated focal length is 2000 pixels, the principal point position is (960, 540), and the distortion coefficients are (-0.2, 0.1) for subsequent image correction. By verifying the calibration results, use an object of known size for testing to ensure the consistency between the actual measurement and the calculated value. Usually, the measurement error is required to be less than 2%. Use a cube of known size to measure its size in the image and ensure it matches the actual size. According to the calculated internal parameters, select a suitable distortion model (such as radial distortion and tangential distortion models) for image correction. Use the distortion correction function in OpenCV. Set the distortion coefficients to (-0.2, 0.1) and calculate the corrected image coordinates. Use the calibration parameters to perform distortion correction on the acquired monitoring images to generate distortion-corrected monitoring images. Ensure that the color value of each pixel remains consistent and avoid information loss during the correction process. When processing each image, set the reprojection error to be less than 1 pixel to ensure the accuracy of the correction result. Conduct a visual inspection on the corrected images to ensure that the geometric shapes of all images meet the expectations and there are no obvious distortion traces. Save the corrected images for subsequent use. Save the corrected images in JPEG format, named "corrected_image_YYYYMMDD_HHMM.jpg" for subsequent processing. Select a suitable super-resolution algorithm, such as SRCNN, ESPCN, or a deep learning model (such as GAN).These methods can effectively improve the resolution of images and enhance details. The ESPCN model is used, which has performed well on multiple datasets and can improve the resolution while maintaining image details. If a deep learning model is used, high-resolution and low-resolution image pairs are required to train the model. After training, the model is applied to the distorted correction monitoring images to generate super-resolution optimized monitoring images. During the training process, the batch size is set to 16 and the number of training epochs is 50 to ensure good convergence of the model. The quality of the generated super-resolution optimized monitoring images is evaluated, and quantitative analysis is carried out using indicators such as PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index). Ensure that the PSNR value is higher than 30dB and the SSIM value is close to 1. Deep learning models (such as YOLO, Mask R-CNN) are used for visual semantic recognition. These models can efficiently identify static objects in images and mark them. The YOLOv5 model is selected because of its good performance in terms of real-time and accuracy, which is suitable for obstacle recognition in the factory environment. The model is trained using the labeled training dataset, including common static obstacles in the factory (such as machines, goods, etc.). Ensure the diversity of the training data to improve the generalization ability of the model. Prepare a training set containing 5000 labeled images, set the learning rate to 0.001, and the number of training epochs to 100. The trained model is applied to the super-resolution optimized monitoring images for obstacle recognition and marking. The recognition results should include the category and location box of each obstacle. The model identifies 3 machines and 5 goods in the factory and outputs the corresponding bounding boxes and confidence levels. Image processing techniques are used to extract the geometric shape parameters of the obstacles, including height, width, depth, and shape features (such as aspect ratio, area, etc.). By calculating the bounding box of the obstacle, its width and height are extracted to obtain basic geometric information. The geometric parameters of each identified obstacle are calculated and recorded in the database. Ensure that the geometric features of each obstacle can be accurately captured. The width of a certain machine is calculated as 2m and the height is 1.5m, and it is recorded as "Machine_A: Width = 2m, Height = 1.5m". The extracted geometric parameters are verified to ensure they are consistent with the actual dimensions of the obstacles. This can be done through on-site measurement or comparison with known dimensions. Adaptive collision constraint boxes are generated based on the geometric shape parameters of the obstacles. An algorithm based on geometric features (such as the minimum bounding rectangle or bounding box) is used to calculate the constraint boxes. The minimum bounding rectangle algorithm is selected to generate suitable constraint boxes according to the position and geometric shape of each obstacle. The collision constraint box for each obstacle is calculated to ensure that the size and position of the box can effectively avoid collisions. The size of the box is dynamically adjusted according to the geometric parameters. For each obstacle, the margin of the collision box is set to 0.5m to ensure that the AGV will not collide with the obstacle during movement. The generated collision constraint boxes are recorded in the database and visualized on the monitoring images.Ensure that each constraint box can be clearly displayed in the image.

[0023] In this embodiment, refer to Figure 3 , which is a schematic diagram of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Detect scene structure feature points based on the super-resolution optimized monitoring image, and extract multiple scene structure feature points; Conduct building space layout analysis on multiple scene structure feature points to obtain the factory building space layout features; Identify the three-dimensional scene structure according to the factory building space layout features, and extract the three-dimensional scene structure features; Build a panoramic scene map based on multiple obstacle collision constraint boxes and three-dimensional scene structure features to construct a panoramic scene structure map.

[0024] In this embodiment, a suitable feature point detection algorithm is selected, such as SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), or ORB (Oriented FAST and Rotated BRIEF). These algorithms can effectively detect and describe the key structural feature points in the image. The ORB algorithm is selected because it performs well in terms of computational efficiency and feature description and is suitable for real-time applications. The super-resolution optimized monitoring image is used as the input, and the selected feature point detection algorithm is applied for processing. The algorithm will identify the key feature points in the image and extract the corresponding feature descriptors. The monitoring image is processed, and the number of feature points is set to 500. The ORB algorithm is used to extract the feature points and their descriptors for subsequent analysis. The detected feature points are screened to remove low-contrast or redundant feature points to ensure that high-quality feature points are retained for subsequent analysis. Non-maximum suppression technology can be used to further improve the quality of the feature points. After screening, 300 high-quality feature points are retained to ensure that these feature points have good repeatability and stability. Spatial layout analysis techniques, such as graph-based analysis, Voronoi diagrams, or Delaunay triangulation, are used to analyze the spatial relationships of the extracted feature points to identify the spatial layout features of the building. The Delaunay triangulation method is selected to generate a triangular network by connecting the feature points and analyze the spatial structure. By constructing a connection relationship graph between the feature points, the relative positions of the feature points are analyzed. The adjacency relationship of each feature point is calculated to establish a spatial layout model. The distances between the feature points are calculated, and the adjacency matrix is output to ensure that the spatial positions of each feature point can be accurately recorded. According to the connected feature points and spatial relationships, the spatial layout features of the factory building are extracted, including the main channels, room divisions, and regional functions, etc. Clustering algorithms can be used to classify the feature points. The main channels and functional areas in the factory are identified, recorded as "production area", "storage area", etc., and their spatial positions are marked. A suitable 3D reconstruction algorithm is selected, such as structured light reconstruction, stereo vision, or multi-view stereo (MVS) technology. These methods can use the extracted feature point data to reconstruct the 3D scene. The stereo vision-based method is selected to reconstruct the 3D scene using the monitoring images from two or more perspectives. By matching the feature points in the images from different perspectives, their depth information is calculated. The RANSAC (Random Sample Consensus) algorithm can be used to remove the mismatched feature points to improve the matching accuracy. During the matching process, the matching accuracy is set to 0.5 pixels to ensure the reliability of the matching results. Based on the calculated depth information, a 3D scene model is constructed. Point cloud processing tools (such as PCL) can be used to convert the feature points into 3D point clouds and perform post-processing to optimize the model quality. The generated 3D point cloud contains approximately 50,000 points, which are filtered and reconstructed to form a clear 3D model. A suitable scene mapping technology is selected, such as SLAM (Simultaneous Localization and Mapping) or graph-based mapping methods.These methods can combine the obstacle collision bounding boxes with the three-dimensional scene structure features to generate a panoramic scene map. By using a graph-based SLAM method, a panoramic map is formed by constructing the relationship between the feature map and the bounding boxes. The obstacle collision bounding boxes are combined with the extracted three-dimensional scene structure features to map the scene. Ensure that the map contains information about all static obstacles to facilitate the navigation of the AGV. Combine the geometric shape and position of the obstacles to generate a scene map that includes all key features. Generate a panoramic scene map and perform optimization to ensure the accuracy and consistency of the map. Graph optimization algorithms (such as g2o) can be used for global optimization. Through the optimization algorithm, ensure that the error of the feature points in the map is less than 5 cm to guarantee the high precision of the map.

[0025] In this embodiment, refer to Figure 4 , which is a schematic diagram of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Obtain the AGV operation log and identify multiple staged targets; Calculate the spatial position information of the staged targets according to the panoramic scene structure map, and extract the spatial position coordinates of each target; Calculate the distance intervals of the spatial position coordinates, perform sequential navigation sorting, and generate a multi-target navigation sequence; Calculate the current position coordinates of the AGV and mark them on the panoramic scene structure map; According to the panoramic scene structure map, perform intelligent AGV navigation planning on the multi-target navigation sequence to construct an intelligent AGV planned path.

[0026] In this embodiment, operation logs are extracted from the control system of the AGV to record the status information of the AGV at different time periods, including the current position, speed, acceleration, sensor data, etc. Ensure the integrity and accuracy of the log data. Set the log recording frequency to once per second to ensure that the collected logs cover all operation stages of the AGV, including startup, driving, docking, etc. Analyze the operation logs to identify the goals that the AGV needs to achieve at different stages. These goals may include specific workstations, material storage areas, or charging stations, etc. Using the timestamp and location data, identify multiple target points on the past path of the AGV, record them as "Target A", "Target B", etc., and extract their corresponding timestamp and location information. Based on the panoramic scene structure map, use geometric calculation methods to convert the location information of the staged goals into three-dimensional space coordinates. This can be achieved through reverse projection technology or map registration methods. Select the reverse projection method to convert the two-dimensional coordinates of the target on the map into three-dimensional space coordinates for subsequent processing. For each identified staged goal, use the location information stored in the panoramic scene structure map to calculate its spatial coordinates. Ensure the accuracy of the calculation results to support subsequent navigation planning. Assume that the coordinates of Target A on the map are (x1, y1) and Target B is (x2, y2), then the calculated spatial position coordinates may be Target A (3.5, 2.0, 0.0) and Target B (6.0, 4.5, 0.0). Record the calculated spatial position coordinates in the database and mark them on the panoramic scene structure map to ensure that the spatial position of each target can be clearly identified. Generate a marked image to show the spatial positions of Target A and Target B and save it as "Target Location Marking Map_YYYYMMDD.jpg". Use the Euclidean distance formula to calculate the distances between each staged goal. This formula can be used to determine the actual distance between goals, facilitating subsequent navigation planning. Calculate the pairwise distances for all identified goals to obtain a distance matrix between goals. This will provide basic data for subsequent sequential navigation sorting. Assume that the calculated distance between Target A and Target B is 3.6 meters, recorded as "Distance A-B = 3.6m". According to the calculated distance information, use the greedy algorithm or the nearest neighbor algorithm to perform sequential navigation sorting of the goals. The goals should be visited in sequence along the shortest path to improve navigation efficiency. Assume the goal sorting is Target A → Target B → Target C to minimize the total travel distance and generate a sequential navigation sequence. Use the AGV's positioning system (such as lidar, IMU, or GPS) to obtain the current spatial coordinates. Ensure that the positioning accuracy is within an acceptable range to support subsequent navigation planning. Set the positioning accuracy to ±5 cm to ensure that the current position of the AGV can be accurately reflected on the map. According to the data of the positioning system, calculate the current position coordinates of the AGV in real time and match this coordinate with the panoramic scene structure map to ensure the accuracy of the position.Assume that the current position of the AGV is (4.0, 3.0, 0.0), and this position should be clearly visible on the map. Mark the current position of the AGV on the panoramic scene structure map and record this information. Ensure that the current state of the AGV can be intuitively displayed on the map. Adopt path planning algorithms such as the A* algorithm, Dijkstra algorithm, or RRT (Rapidly-Exploring Random Tree) algorithm to perform path planning for the multi-objective navigation sequence. Select a suitable algorithm to ensure the efficiency and safety of path planning. Select the A* algorithm because it can quickly find the optimal path in a complex environment. Take the current AGV position and the multi-objective navigation sequence as inputs, and calculate the optimal paths from the current AGV position to each target through the path planning algorithm. The paths should consider the presence of obstacles to avoid collisions. Set the starting point as the current AGV position, and the targets are Target A and Target B in sequence, and calculate the path length and the estimated travel time. Optimize the planned paths to ensure the smoothness and feasibility of the paths. The sharp turns and unnecessary detours in the paths can be reduced through path smoothing algorithms. The generated optimized paths should avoid obstacles, ensure the minimum path length, and record the path information.

[0027] In this embodiment, step S4 includes the following steps: Perform real-time navigation control on the AGV based on the intelligent AGV planned path, and collect laser sensing feedback parameters in real time based on the lidar; Identify the tiny abnormal protrusions on the terrain from the laser sensing feedback parameters, and mark the tiny abnormal protrusion points on the terrain; Register the building positions of the tiny abnormal protrusion points to generate the matching information of the building positions of the protrusion points; Perform local building structure synchronization processing on the panoramic scene structure map according to the matching information of the building positions of the protrusion points to construct a synchronized and optimized scene map.

[0028] In this embodiment, a real-time navigation control system is designed, which combines the planned path and real-time status (such as position, speed, direction) of the AGV. This system needs to ensure that the AGV can dynamically adjust its driving trajectory and perform the navigation task according to the predetermined path. The control frequency is set to 10 Hz, which means that the status of the AGV is updated and the path is adjusted every second. The system should be able to calculate the required steering angle and speed in real time according to the deviation between the current position of the AGV and the target path. The PID (Proportional-Integral-Derivative) control algorithm is adopted for path tracking. By calculating the deviation between the current path of the AGV and the target path, the speed and steering of the AGV are adjusted to ensure that it travels along the planned path. The proportional coefficient of the PID controller is set to 1.2, the integral coefficient is set to 0.01, and the derivative coefficient is set to 0.1, so as to quickly respond to the path deviation and maintain the stability and accuracy of the AGV. A real-time feedback mechanism is implemented to monitor the operating status of the AGV and compare it with the path tracking accuracy. If the actual path deviation exceeds the set threshold (for example, 5 cm), the speed and direction of the AGV are automatically adjusted to correct its driving trajectory. If the AGV deviates from the planned path by more than 5 cm, the control system will immediately adjust its steering to make the AGV return to the target trajectory to ensure the safety and effectiveness of driving. A lidar sensor is installed on the AGV to ensure that it can scan the surrounding environment 360 degrees. The working frequency of the lidar is set to 10 Hz to obtain environmental data in real time. The measurement range of the lidar is set to 0.2 m to 10 m, and the resolution is 1 degree to improve the perception ability of the surrounding environment and ensure that obstacle and terrain information can be obtained in time. Through the lidar data collected in real time, point cloud data is generated and processed to extract terrain information, including the position and height change of obstacles. Filtering algorithms are used to remove noise points to ensure the accuracy of the data. The statistical outlier removal method (such as RANSAC) is applied to clean the point cloud, extract effective terrain features, and generate a three-dimensional model containing obstacle information. The real-time feedback parameters of the laser sensor are recorded in the database, and the data is transmitted to the central control system through the wireless network for subsequent analysis and decision support. Ensure that the lidar data recorded once per second includes point cloud information, obstacle position, and its height change for subsequent processing. A statistical-based method or machine learning algorithm (such as Support Vector Machine SVM) is used to detect terrain anomalies in the lidar feedback data to identify small protrusions. Preliminary screening can be carried out by setting a height threshold. The height change threshold is set to 5 cm. When the height change detected in the point cloud data exceeds this value, it is determined as a terrain anomaly protrusion. The identified terrain anomaly small protrusion points are marked, and their spatial coordinates are recorded. Ensure that the information of each protrusion point (such as position, size, and height) can be clearly identified and accessed. If a protrusion point coordinate (5.2, 3.1, 0.15) is detected, it is recorded as "Protrusion Point 1", and its relevant information, including the protrusion height and other features, is saved.Use feature point matching and registration algorithms (such as the ICP algorithm) to register the protruding points with the buildings in the panoramic scene structure map to ensure the accurate position of the protruding points in the map. Use the ICP algorithm for point cloud alignment, set the maximum number of iterations to 50 times, and the convergence threshold to 0.01 meters to ensure the registration accuracy. Perform building position registration on the marked protruding points, calculate their corresponding building positions in the panoramic scene structure map, optimize the matching results, and ensure that the protruding point information is accurately integrated into the map. Calculate the distance between the protruding points and the building features, use the least squares method to optimize the matching results, and ensure that the matching error is less than 5 centimeters. Record the matching information between the protruding points and the buildings in the database, including the matching coordinates, errors, and matching status, for subsequent analysis and optimization. Record the matching coordinates and errors of protruding point 1 and building A, and name it "matching information_protruding point 1_building A" to ensure the traceability of the data. Select a suitable synchronous processing algorithm, such as a graph optimization-based synchronous processing method or a local map update method, to synchronously update the local part of the panoramic scene structure map. Use a graph optimization algorithm (such as g2o) to update the local map to integrate the newly recognized terrain protrusion information. Integrate the registration information of the protruding points with the panoramic scene structure map, update the relevant information of the building structure, and ensure that the synchronized map can reflect the latest environmental changes. During the update process, ensure that the protruding point information is consistent with the height and shape information of the corresponding buildings in the map to avoid information conflicts. Verify the synchronized and optimized scene map to ensure the accuracy and consistency of the new information. Save the updated panoramic scene structure map for subsequent use.

[0029] In this embodiment, step S4 includes the following steps: Conduct a maximum passing form limit analysis on the intelligent AGV planned path according to the synchronized and optimized scene map to obtain the maximum safe passing form parameters on the path; Obtain the preset AGV form parameters, and conduct a collision risk assessment based on the maximum safe passing form parameters on the path to obtain the collision risk assessment value; Based on the collision risk assessment value, conduct a scene structure protrusion collision avoidance for the intelligent AGV planned path to generate a collision avoidance optimized path.

[0030] In this embodiment, the maximum passing form parameters of the AGV are defined, including the size (length, width, height), turning radius, and maximum tilt angle of the AGV, etc. These parameters will be used to analyze the passability of the path. The length of the AGV is set to 1.2 meters, the width is 0.8 meters, the height is 0.5 meters, the turning radius is 0.6 meters, and the maximum tilt angle is 10 degrees to ensure that the AGV can drive safely in different environments. Extract the data of the AGV planned path from the synchronized and optimized scene map, including the geometric shape of the path, the positions of obstacles, and the characteristic information. Ensure the accuracy of the path data for subsequent analysis. Extract the coordinates of the key points on the path and the distance information between them and the surrounding obstacles, and record them as a list of path points for subsequent use. Use geometric analysis methods, combined with the form parameters of the AGV and the path characteristics, to analyze the passability of each point on the path. By calculating the minimum distance between the AGV and the obstacles, judge the feasibility of the path. Calculate the circumscribed rectangle of the AGV at each key point of the path and perform collision detection with the geometric shape of the obstacles. If the detected distance is less than the preset safety distance (such as 0.2 meters), this path point will be marked as impassable. Obtain the preset form parameters of the AGV for collision risk assessment. Ensure that these parameters can accurately reflect the state of the AGV during actual operation. Record the existing form parameters of the AGV, including its length, width, and height, and set these parameters as the standard for the AGV during operation. Adopt a collision risk assessment model, such as a probability-based risk assessment or a distance-based risk assessment method, to evaluate each key point on the path. Ensure that the model can accurately reflect the potential collision risks. Select the distance-based risk assessment method and evaluate the collision risk by calculating the distance between the AGV and the obstacles. Set the risk assessment threshold. If the distance is less than 0.5 meters, the assessed risk is high. Conduct a collision risk assessment on each key point on the AGV planned path, calculate the distance to the surrounding obstacles, and calculate the risk assessment value according to the distance. If the distance from a certain path point to the nearest obstacle is 0.4 meters, the calculated collision risk assessment value is high; if the distance is 1.0 meters, the assessment value is low. According to the collision risk assessment results, formulate a collision avoidance strategy for the AGV. For high-risk path points, consider adjusting the path or choosing an alternative route to avoid potential collisions. Set that if the risk assessment value of a path point is high, the path needs to be replanned to ensure that the AGV can pass safely. Adopt a path optimization algorithm, such as the A* algorithm, Dijkstra algorithm, or RRT (Rapidly-Exploring Random Tree) algorithm, to generate an optimized collision avoidance path. Ensure that the algorithm can quickly find the optimal path. Select the A* algorithm and set the heuristic function as the Euclidean distance to improve the efficiency of path search. Apply the collision avoidance strategy in the AGV planned path and generate a new collision avoidance path through algorithm calculation. Ensure that the new path can avoid all high-risk areas and still meet the requirements of minimizing time and distance.Calculate a new path through the path optimization algorithm to ensure that the AGV can still complete the task while bypassing obstacles, and record the key point coordinates of the optimized path.

[0031] In this embodiment, the steps of obtaining the preset AGV form parameters and performing a collision risk assessment based on the maximum safe passage form parameters on the path to obtain a collision risk assessment value include the following: Obtain the preset AGV form parameters; Conduct a turning radius constraint analysis based on the AGV form parameters and extract the AGV turning radius; Calculate the minimum passage path plane width and the minimum scene height limit according to the AGV form parameters; Perform a collision risk assessment on the maximum safe passage form parameters based on the minimum passage path plane width and the minimum scene height limit to obtain a collision risk assessment value.

[0032] In this embodiment, first, define the basic form parameters of the AGV, including length, width, height, turning radius, and center of gravity position, etc. These parameters will be used for subsequent turning radius constraint analysis and passage path calculation. Set the length of the AGV to 1.2 meters, the width to 0.8 meters, the height to 0.5 meters, the turning radius to 0.6 meters, and the center of gravity height to 0.3 meters. These parameters should accurately reflect the physical characteristics of the AGV during actual operation. Record the form parameters of the AGV in the database to ensure that this information can be conveniently accessed during subsequent path planning and collision assessment. The data should include all key parameters and their units. Create a data table to record the AGV form parameters, including "Length (m)", "Width (m)", "Height (m)", "Turning Radius (m)", etc., and update the record regularly. Verify the obtained AGV form parameters to ensure their accuracy. Verification can be carried out through on-site measurement or comparison with the data provided by the manufacturer to ensure reliability in subsequent analysis. Use measurement tools to actually measure the AGV to ensure that the recorded length, width, and height match the actual values, and the error should be within ±2 centimeters. Adopt a geometric analysis method to calculate the turning radius of the AGV. This process needs to consider the form parameters of the AGV, especially the influence of the width and center of gravity position on the turning radius. Set the calculation formula for the turning radius as: R = , where R is the turning radius, W is the width of the AGV, and L is the length of the AGV. Based on the morphological parameters of the AGV, the actual turning radius is calculated. If the width of the AGV is 0.8 meters and the length is 1.2 meters, then the turning radius is: R = 0.4 + 0.6 = 1.0 meter. Define the minimum width of the passage path plane and the minimum height limit of the scene. These parameters will directly affect the passing ability of the AGV. The minimum width of the passage path should consider the width and turning radius of the AGV, while the minimum height limit should consider the height of the AGV and possible obstacles. Set the minimum width of the passage path as the width of the AGV plus twice the turning radius, that is: W(min) = W + 2×R, where R is the turning radius calculated previously. According to the morphological parameters of the AGV, if the width of the AGV is 0.8 meters and the turning radius is 1.0 meter, then the minimum width of the passage path is: 0.8 + 2×1.0 = 2.8 meters. The minimum height limit of the scene can be simply set as the height of the AGV plus a safety margin. For example, set it as 0.5 meters (the height of the AGV) plus a margin of 0.2 meters, totaling 0.7 meters. Select a suitable collision risk assessment model, such as a distance-based risk assessment model, and use the minimum width of the passage path and the height limit of the scene to evaluate the path. Set that if the path width is not sufficient to accommodate the width of the AGV during turning, the collision risk assessment value of this path segment is high. Conduct a collision risk assessment for each key point on the path, calculate its distance from surrounding obstacles, and combine the minimum width of the passage path and the height limit to obtain the specific risk assessment value. If the width of a certain path segment is 2.5 meters, while the width required for the AGV to turn is 2.8 meters, then the risk assessment value of this path segment is high, otherwise it is low. Record the collision risk assessment value of each path segment in the database and generate an assessment report for subsequent analysis and decision support.

[0033] In this embodiment, step S6 includes the following steps: Identify the transient change characteristics of the laser sensing feedback parameters; Conduct dynamic obstacle analysis based on the transient change characteristics and mark the dynamic obstacles; Perform time-series movement tracking on the dynamic obstacles to generate a time-series movement trajectory of the dynamic obstacles; Conduct multi-time-point movement prediction based on the time-series movement trajectory of the dynamic obstacles to obtain a multi-time-point movement prediction trajectory; Based on the multi-time-point movement prediction trajectory, perform adaptive path iterative optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

[0034] In this embodiment, real-time feedback parameters from lidar are collected, including point cloud data, the distance and reflection intensity of obstacles, etc. The data acquisition frequency is set to 10 Hz to ensure sufficient dynamic information is obtained. The point cloud data collected per second is recorded to generate a high-frequency environmental model for subsequent analysis. Signal processing techniques (such as the Fast Fourier Transform FFT) are used to perform frequency-domain analysis on the collected point cloud data to identify the transient change characteristics of the feedback parameters. These characteristics may include rapid changes in the distance of obstacles or sudden changes in reflection intensity. A threshold of 5 cm is set, and when the distance change of an obstacle in the point cloud data exceeds this threshold, it is marked as a transient change. This can help identify dynamic obstacles. A rule-based dynamic obstacle analysis algorithm is used, combining the transient change characteristics and spatial position data to mark dynamic obstacles. This process will utilize the previously extracted transient change characteristics. The marking condition for dynamic obstacles is set as follows: within the same time window, if the distance change of an obstacle exceeds 5 cm and the obstacle persists at multiple time points, it is marked as a dynamic obstacle. According to the identified transient change characteristics, dynamic obstacles are marked and their spatial positions are recorded. Ensure that the information of each dynamic obstacle (such as position, speed, change direction) can be clearly identified. If the coordinates of an obstacle are (5.2, 3.1, 0.15) and it is marked as dynamic during the transient change, it is recorded as "Dynamic Obstacle 1" and its relevant information is saved. Temporal movement tracking of dynamic obstacles is performed to collect their position information at different time points. The data collection frequency is set to 5 Hz to ensure that the movement trajectories of dynamic obstacles can be captured. The position changes of each dynamic obstacle within 5 consecutive seconds are recorded for subsequent analysis. Using timestamps and position information, the temporal movement trajectories of dynamic obstacles are generated. Interpolation methods can be used to smooth the trajectories to improve the visualization effect of the trajectories. If the position of a dynamic obstacle at the 1st second is (5.2, 3.1) and the position at the 2nd second is (5.4, 3.3), trajectory points are generated through linear interpolation to provide data support for subsequent prediction. The temporal movement trajectories of dynamic obstacles are recorded in the database and visual charts are generated to facilitate the observation of the movement patterns of dynamic obstacles. A dynamic obstacle movement trajectory map is generated and saved as "Dynamic Obstacle Trajectory Map_YYYYMMDD.jpg" for subsequent analysis. Polynomial fitting or time series prediction models (such as the ARIMA model) are used to perform multi-time-point movement prediction on the temporal movement trajectories of dynamic obstacles to predict future movement paths. The polynomial fitting method is selected, and the order of the fitted polynomial is set to 2 to capture the acceleration and deceleration trends of dynamic obstacles. Based on the already generated temporal movement trajectories, the selected model is used to predict future positions. The prediction time range can be set to 5 seconds, and the prediction step size is 1 second.If the current speed of the dynamic obstacle is 0.5 m / s, predict the trajectory points for the next 5 seconds and record them as (5.6, 3.5), (5.8, 3.7), (6.0, 3.9), etc. Design an adaptive path iterative optimization model to adjust the driving path of the AGV according to the predicted trajectory of the dynamic obstacle. The model needs to consider the current state of the AGV, the predicted position of the dynamic obstacle, and its moving speed. Set the iteration frequency of the model to 5 Hz to quickly adjust the path according to the new dynamic information. Use a path optimization algorithm (such as A* or Dijkstra) combined with the predicted trajectory of the dynamic obstacle to adjust the path, ensuring that the AGV can avoid dynamic obstacles in real time. Set the safety distance for avoidance to 0.5 m. If the predicted trajectory intersects with the current path of the AGV, calculate a new path to ensure that the AGV can pass safely without colliding with the dynamic obstacle. Record the optimized path in the database and generate a visualization chart to show the relationship between the optimized path and the trajectory of the dynamic obstacle.

[0035] In this embodiment, a high-precision laser SLAM navigation system for a heavy-duty AGV is provided, which is used to execute the high-precision laser SLAM navigation method for the heavy-duty AGV as described above, and includes: A terrain recognition module, which is used to continuously collect panoramic monitoring images of the factory based on a high-definition camera, perform terrain obstacle segmentation, and generate multiple obstacle collision constraint boxes; A panoramic mapping module, which is used to perform three-dimensional scene structure recognition based on the panoramic monitoring images of the factory, and perform panoramic scene mapping based on multiple obstacle collision constraint boxes to construct a panoramic scene structure map; A path planning module, which is used to calculate the current position coordinates of the AGV, perform intelligent AGV navigation planning based on the panoramic scene structure map, and construct an intelligent AGV planning path; A local synchronization optimization module, which is used to collect laser sensing feedback parameters in real time based on a lidar, and perform local building structure synchronization processing on the panoramic scene structure map to construct a synchronized optimization scene map; A passage restriction analysis module, which is used to perform maximum passage form restriction analysis and scene structure protrusion collision avoidance on the intelligent AGV planning path according to the synchronized optimization scene map, and generate a collision avoidance optimized path; A dynamic obstacle avoidance module, which is used to perform dynamic obstacle time-series movement tracking according to the laser sensing feedback parameters, and perform adaptive path iterative optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

[0036] The present invention can provide high-resolution image data through a high-definition camera, enabling the AGV to more clearly identify static and dynamic obstacles in the surrounding environment. This enables the AGV to maintain a high level of stability and accuracy when facing complex or changing environments. By segmenting terrain obstacles from the image data, the AGV can accurately identify potential collision obstacles and clearly define the spatial scope of these obstacles by generating collision bounding boxes for multiple obstacles. This provides an effective reference for subsequent path planning and collision avoidance. By identifying and reconstructing the three-dimensional structure of the scene, the AGV can obtain a more comprehensive and detailed environmental model. This three-dimensional mapping can not only show the spatial layout of all obstacles and building structures in the factory, but also improve the navigation accuracy of the AGV in complex environments. The panoramic mapping based on the obstacle collision bounding boxes can not only depict the static environment, but also provide a complete view containing information such as the position, shape, and size of environmental obstacles. In this way, the AGV can clearly understand the distribution of obstacles in the entire environment during path planning and provide an accurate reference for intelligent navigation. By combining the panoramic scene structure map and the current position of the AGV, the system can accurately calculate the coordinates of the AGV and generate an optimal planned path according to the real-time map. This ensures that the AGV can move forward stably in complex environments and avoid problems such as incorrect positioning or inability to pass through narrow paths. The intelligent navigation planning based on the panoramic scene map can dynamically adjust the path to adapt to new obstacles or changes in the environment. This intelligent planning can significantly improve the autonomous navigation ability of the AGV and enhance its flexibility and adaptability in changing environments. The high-precision distance information provided by the lidar combined with the visual data of the high-definition image can accurately correct the deviation in environmental mapping. By synchronously processing the lidar data with the panoramic scene map, high-precision optimization of the local area building structure can be achieved, ensuring that the AGV can obtain the latest scene information. As the AGV moves, the obstacles and structures in the environment may change. By real-time synchronously optimizing the scene map, the AGV can ensure that the navigation path always remains consistent with the actual environment, thus avoiding navigation errors caused by outdated map information. By analyzing the passing restrictions of the AGV (such as turning radius, load capacity, etc.), a path that can maximize the use of its driving space can be generated for the AGV, thereby optimizing the driving efficiency and safety. This enables the AGV to select the best path in different environments according to its specific physical form and work tasks. By analyzing the protrusions or obstacles in the synchronously optimized scene map, the AGV can intelligently avoid collisions. The collision avoidance optimized path can be adjusted in real time according to the obstacle layout around the AGV, providing a safer driving route for the AGV, which is particularly important in narrow spaces or complex factory environments. By real-time tracking the movement trajectory of dynamic obstacles, the AGV can predict their future positions and make avoidance decisions in advance.This is particularly important for dealing with occasional dynamic obstacles in the factory environment, such as the movement of personnel or equipment. Based on the laser sensing feedback parameters, the AGV can adjust its path in real time according to the dynamic changes of obstacles during driving. Adaptive path optimization can significantly improve the navigation ability of the AGV, ensuring that it can still operate efficiently in the face of unforeseen obstacles or environmental changes. Combining the above dynamic feedback and path optimization strategies, the AGV can continuously learn and improve its navigation decisions, generating more efficient and safe navigation strategies. This optimization model can not only improve the navigation accuracy of the AGV, but also make it more adaptable to variable and complex environments.

[0037] Therefore, in any aspect, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be included in the present invention.

[0038] As described above, these are only the specific implementation manners of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A high-precision laser SLAM navigation method for heavy-duty AGV, characterized in that, The heavy-duty AGV device has a high-definition camera and a lidar, including the following steps: Step S1: Continuously collect factory panoramic monitoring images based on the high-definition camera, and perform terrain obstacle segmentation to generate multiple obstacle collision bounding boxes; Step S2: Identify the three-dimensional scene structure based on the factory panoramic monitoring images, and perform panoramic scene mapping based on multiple obstacle collision bounding boxes to construct a panoramic scene structure map; Step S3: Calculate the current position coordinates of the AGV, perform intelligent AGV navigation planning based on the panoramic scene structure map, and construct an intelligent AGV planning path; Step S4: Continuously collect laser sensing feedback parameters based on the lidar, and perform local building structure synchronization processing on the panoramic scene structure map to construct a synchronized and optimized scene map; Step S5: Perform maximum passage form limit analysis and scene structure protrusion collision avoidance on the intelligent AGV planning path according to the synchronized and optimized scene map to generate a collision avoidance optimized path; Step S6: Perform dynamic obstacle time-series movement tracking based on the laser sensing feedback parameters, and perform adaptive path iterative optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

2. The high-precision laser SLAM navigation method for the heavy-duty AGV according to claim 1, characterized in that The specific steps of Step S1 are as follows: Continuously collect factory panoramic monitoring images based on the high-definition camera; Calculate the internal and external calibration parameters of the high-definition camera, and perform image distortion correction on the factory panoramic monitoring images according to the internal and external calibration parameters to obtain a distortion-corrected monitoring image; Perform super-resolution enhancement on the distortion-corrected monitoring image to obtain a super-resolution optimized monitoring image; Perform omnidirectional visual semantic recognition on the super-resolution optimized monitoring image to mark the static terrain obstacles in the scene; Calculate the geometric shape parameters of the static terrain obstacles in the scene; Perform adaptive bounding box segmentation according to the geometric shape parameters to generate multiple obstacle collision bounding boxes.

3. The high-precision laser SLAM navigation method for a heavy-duty AGV according to claim 1, wherein The specific steps of Step S2 are as follows: Detect scene structure feature points based on the super-resolution optimized monitoring image, and extract multiple scene structure feature points; Perform building space layout analysis on multiple scene structure feature points to obtain the factory building space layout features; Identify the three-dimensional scene structure according to the factory building space layout features, and extract the three-dimensional scene structure features; Perform panoramic scene mapping based on multiple obstacle collision bounding boxes and three-dimensional scene structure features to construct a panoramic scene structure map.

4. The high-precision laser SLAM navigation method for the heavy-duty AGV according to claim 1, characterized in that, The specific steps of Step S3 are as follows: Obtain the AGV operation log and identify multiple staged targets; Calculate the spatial position information of the staged targets according to the panoramic scene structure map, and extract the spatial position coordinates of each target; Calculate the distance intervals of the spatial position coordinates, and perform sequential navigation sorting to generate a multi-target navigation sequence; Calculate the current position coordinates of the AGV and mark them on the panoramic scene structure map; Perform intelligent AGV navigation planning on the multi-target navigation sequence according to the panoramic scene structure map to construct an intelligent AGV planning path.

5. The high-precision laser SLAM navigation method for a heavy-duty AGV according to claim 1, characterized in that, The specific steps of Step S4 are as follows: Perform real-time navigation control on the AGV based on the intelligent AGV planning path, and continuously collect laser sensing feedback parameters based on the lidar; Identify terrain abnormally small protrusions in the laser sensing feedback parameters and mark the terrain abnormally small protrusion points; Perform building position registration for tiny abnormal protrusions on the terrain to generate building position matching information for the protrusions; Perform local building structure synchronization processing on the panoramic scene structure map according to the building position matching information for the protrusions to construct a synchronized and optimized scene map.

6. The high-precision laser SLAM navigation method for the heavy-duty AGV according to claim 1, characterized in that, The specific steps of step S5 are as follows: Perform maximum passing form limit analysis on the planned path of the intelligent AGV according to the synchronized and optimized scene map to obtain the maximum safe passing form parameters on the path; Obtain the preset AGV form parameters, and perform collision risk assessment according to the maximum safe passing form parameters on the path to obtain a collision risk assessment value; Perform scene structure protrusion collision avoidance on the planned path of the intelligent AGV based on the collision risk assessment value to generate a collision avoidance optimized path.

7. The high-precision laser SLAM navigation method for the heavy-duty AGV according to claim 6, characterized in that, The specific steps of obtaining the preset AGV form parameters and performing collision risk assessment according to the maximum safe passing form parameters on the path to obtain a collision risk assessment value are as follows: Obtain the preset AGV form parameters; Perform turning radius constraint analysis based on the AGV form parameters and extract the AGV turning radius; Calculate the minimum passing path plane width and the minimum scene height limit according to the AGV form parameters; Perform collision risk assessment on the maximum safe passing form parameters based on the minimum passing path plane width and the minimum scene height limit to obtain a collision risk assessment value.

8. The high-precision laser SLAM navigation method for the heavy-duty AGV according to claim 1, characterized in that, The specific steps of step S6 are as follows: Identify the transient change characteristics of the laser sensing feedback parameters; Perform dynamic obstacle analysis according to the transient change characteristics and mark the dynamic obstacles; Perform time-series movement tracking on the dynamic obstacles to generate a time-series movement trajectory of the dynamic obstacles; Perform multi-time-point movement prediction according to the time-series movement trajectory of the dynamic obstacles to obtain a multi-time-point movement prediction trajectory; Perform adaptive path iterative optimization on the collision avoidance optimized path based on the multi-time-point movement prediction trajectory to construct an intelligent navigation optimization model.

9. A high-precision laser SLAM navigation system for a heavy-duty AGV, characterized in that, A high-precision laser SLAM navigation method for a heavy AGV as claimed in claim 1, comprising: A terrain recognition module for continuously collecting factory panoramic monitoring images based on a high-definition camera and performing terrain obstacle segmentation to generate a plurality of obstacle collision constraint frames; A panoramic mapping module for performing three-dimensional scene structure recognition according to the factory panoramic monitoring images and performing panoramic scene mapping based on a plurality of obstacle collision constraint frames to construct a panoramic scene structure map; A path planning module for calculating the current position coordinates of the AGV and performing intelligent AGV navigation planning according to the panoramic scene structure map to construct an intelligent AGV planned path; A local synchronization optimization module for continuously collecting laser sensing feedback parameters based on a lidar and performing local building structure synchronization processing on the panoramic scene structure map to construct a synchronized and optimized scene map; A passing limit analysis module for performing maximum passing form limit analysis and scene structure protrusion collision avoidance on the intelligent AGV planned path according to the synchronized and optimized scene map to generate a collision avoidance optimized path; A dynamic obstacle avoidance module for performing time-series movement tracking of dynamic obstacles according to the laser sensing feedback parameters and performing adaptive path iterative optimization on the collision avoidance optimized path to construct an intelligent navigation optimization model.

Images