Underground carry-scraper autonomous navigation path generation system based on interaction of front end and rear end

Abstract

The invention belongs to the technical field of path planning, and discloses an underground carry-scraper autonomous navigation path generation system based on front-end and rear-end interaction, which comprises the following steps: fusing multi-source map data, and constructing a space coordinate bidirectional mapping engine coupling physical space and pixel space association; detecting space conflicts between the user-labeled path points and the roadway structure in real time and generating a dynamic path skeleton; road section types are divided, then paths are dynamically optimized, redundant data are synchronously compressed, and an optimized path packet is generated; constructing a route version control tree, dynamically allocating storage resources according to roadway topology, and generating a simulation ready path file package; the transverse error and the minimum roadway wall distance are solved in real time, a verification report is generated, then an overrun event in the simulation trajectory tracking process is captured, a double-channel tracing report is generated, and a path section needing to be corrected is reversely marked; and performing hot update on the optimization rule, quantifying the path adaptability and performing reverse iteration on the core parameter library.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and more specifically, to an autonomous navigation path generation system for underground shovel loader based on front-end and back-end interaction. Background Technology

[0002] In the process of intelligent transformation of mines, the demand for fixed route navigation of underground loaders is becoming increasingly urgent. However, traditional dynamic programming algorithms suffer from serious problems of repeated calculations and poor path stability in fixed roadway environments. Manual trajectory recording is costly and lacks accuracy. In addition, existing technologies cannot effectively integrate roadway constraints (such as narrow bends and dust interference) with equipment dynamics, resulting in low path generation efficiency and significant fluctuations in driving trajectory, which cannot meet the safety and efficiency requirements of high-precision autonomous driving.

[0003] Current mainstream solutions have significant limitations: while dynamic programming algorithms (such as RRT / A*) can cope with dynamic environments, repeated planning in fixed lanes generates a large amount of redundant calculations, and the path results fluctuate greatly; although manual driving records can generate fixed routes, adjustments require re-operation of the equipment, which is time-consuming and labor-intensive, and the lack of digital editing capabilities means that optimization relies entirely on driving experience; more importantly, existing technologies do not fully consider the impact of dusty environments on perception, the dynamic constraints of narrow lanes on the path, and the tracking adaptability of the control layer to the planned path, resulting in poor path feasibility and severe control jitter in actual deployment. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an autonomous navigation path generation system for underground shovel loader based on front-end and back-end interaction, comprising: Multi-source fusion and path construction unit: It integrates multi-source map data to build a two-way spatial coordinate mapping engine that couples physical space and pixel space; it detects spatial conflicts between user-annotated path points and alleyway structures in real time and generates dynamic path skeletons.

[0005] Path segmentation and differentiated optimization unit: Based on the dynamic path skeleton, the path segment type is divided, and then the path is dynamically optimized, while redundant data is compressed in a synchronous manner to generate an optimized path package; Multi-route management and simulation-ready module: Constructs a route version control tree based on the optimized route package, dynamically allocates and encapsulates storage resources according to the tunnel topology, and generates a simulation-ready route file package; Simulation verification and traceability correction unit: Based on the simulation-ready path file package, it calculates the lateral error and minimum lane wall distance in real time and generates a verification report; then it captures out-of-limit events during the simulation trajectory tracking process, generates a dual-channel traceability report and reverse-marks the path segments that need to be corrected; Actual operating condition closed-loop evolution unit: Based on the hot update optimization rules of the dual-channel traceability report, quantify the path adaptability and iterate the core parameter library in reverse.

[0006] Furthermore, the method by which the bidirectional spatial coordinate mapping engine couples the physical space and pixel space includes: Acquire underground space GIS maps, ROS environment configuration files, and multi-sensor data streams as multi-source map data; Multimodal sensor fusion is performed on multi-sensor data streams to generate fused point cloud data; By combining GIS maps, point cloud data is rasterized, feasible areas and obstacle areas are marked, and semantic maps are generated. The ROS environment configuration file is parsed to obtain ROS environment parameters. Based on the ROS environment parameters, a two-way spatial coordinate mapping engine between the ROS world coordinate system and the map pixel coordinate system is established to perform bidirectional transformation between physical space and pixel space and generate a coordinate transformation rule set.

[0007] Furthermore, the dynamic path skeleton is generated in the following ways: Based on the pixel coordinates of the path points marked by the user and the semantic map, the coordinate transformation rule set is called to convert the pixel coordinates into ROS world coordinates, and the path points are checked to see if they are located in the feasible area. If the area is feasible, allow the addition of waypoints; if it is not feasible, trigger a conflict alarm and prevent the addition of waypoints; synchronously update the sequence of waypoints marked by the user. Compare the current semantic map with the historical semantic map. If a difference is detected, add dynamic obstacle markers to the changed area and trigger an incremental update of the semantic map. Set up an adaptive recognition mechanism for lane width to obtain the effective passage width; If the effective passage width does not meet expectations, mark it as a narrow area and add it to the semantic map as a lane width annotation simultaneously; Based on the updated path point sequence and the dynamic obstacle markers and lane width annotations in the semantic map, the curvature between every three consecutive path points in the path point sequence is obtained. If the curvature does not meet expectations, mark it as a turning point; if the waypoint is located in a dynamic obstacle identification area or a narrow area, forcibly mark it as a turning point. Integrate the tagged waypoint sequence and semantic map to generate a dynamic path skeleton; Based on the dynamic path skeleton, a dynamic topology map is constructed and visualized, generating a map rendering interface.

[0008] Furthermore, the method for classifying road segment types includes: Based on the dynamic path skeleton, safety margin constraints are constructed and added to the road segments traversed by the path points. Based on safety margin constraints, straight sections and turning sections are adaptively divided to obtain road segment classification with width constraints.

[0009] Furthermore, the method for generating the optimized path package includes: Based on the road segment classification results, differentiated optimization is performed on straight-ahead segments and turning segments; For straight sections, increase the density of waypoints and optimize the shape of the path curve; For the turning section, curvature constraint treatment is applied. Traverse the optimized path point sequence, delete redundant path points whose distances between adjacent path points do not meet expectations, obtain the compressed path point sequence, and generate the optimized path package.

[0010] Furthermore, the method for generating the simulation-ready path file package includes: Based on the optimized path package, create an independent branch for each alleyway path and add a version tag to each path; Extract the alleyway topology from the dynamic topology map and dynamically allocate storage paths; Define the original coordinate system and target coordinate system for all path points in the optimized path package, and call the coordinate transformation rule set to perform coordinate transformation; The converted pathpoint data is integrated, packaged in ROS, and a simulation-ready path file package is generated.

[0011] Furthermore, the verification report is generated in the following ways: Based on the simulation-ready path file package, the optimized path is deployed in the ROS environment, the lateral error is calculated in real time and the heading angle convergence performance is verified, and the heading angle convergence time is recorded synchronously. Simulate laser point cloud scanning and obtain the nearest alleyway wall point to the equipment to obtain the minimum alleyway wall distance; The system integrates lateral error, heading angle convergence time, and minimum tunnel wall distance to generate a verification report, which is then visualized.

[0012] Furthermore, the method for generating a dual-channel traceability report and reverse-marking the path segments that need correction includes: Based on the visualization-processed verification report, extract the horizontal error time series data, construct the time-error curve, and then identify and mark the horizontal error exceedance points; Extract the tunnel map from the simulation-ready path file package, perform rasterization, and then create a two-dimensional array; For each grid cell in the two-dimensional array, the grid risk is updated based on the minimum alleyway distance of the most recent simulated laser point cloud scan, and a spatial conflict heat map marking high-risk areas is generated. Integrate time-error curves and spatial conflict heatmaps to generate dual-channel traceability reports; Based on the dual-channel traceability report, the high-risk areas of the lateral error exceeding the limit are combined with the high-risk areas of the spatial conflict heat map to locate high-risk road segments, generate a list of high-risk road segments and perform visualization processing.

[0013] Furthermore, the hot update optimization rules are implemented in the following ways: According to the dual-channel tracing report, the differentiated optimization parameters for the straight and turning segments are dynamically adjusted based on the convergence time of the lateral error direction and heading angle.

[0014] Furthermore, the method of quantizing the path adaptively and iteratively back through the core parameter library includes: Based on the results of the hot update optimization rules, the path fit rate is quantified by the time-error curve, and the division accuracy of straight and turning segments and the identification accuracy of redundant path points are updated in reverse. All updated parameters are integrated, written into a new generation of core parameter library and encapsulated for adaptive iteration.

[0015] The technical effects and advantages of this invention, a front-end and back-end interactive autonomous navigation path generation system for underground loaders, are as follows: This invention achieves precise correlation between physical space and pixel space through multi-source map fusion and bidirectional coordinate mapping engine, combines multi-sensor data to compensate for environmental interference in real time, and dynamically embeds lane conflict detection and turning marks in interactive annotation to ensure the accuracy and environmental adaptability of the path skeleton from the source. Secondly, the system intelligently divides straight sections and turning sections, performs differentiated optimization, and simultaneously introduces adaptive constraints on lane width to enhance curvature control, ensuring a smooth path that strictly conforms to the dynamic characteristics of the articulated vehicle body. Next, a topology-aware version control tree is constructed, and a multi-route anti-overwrite management is achieved through a dynamic storage preemption strategy. Simulation-ready files are generated to support efficient batch deployment. Then, in the simulation environment, the feasibility of the path is verified through a dual-loop process, and a dual-channel traceability report is generated that integrates the time-error curve and the spatial conflict heatmap to accurately locate high-risk road sections. Finally, based on the hot update optimization rules of the traceability report, the weights and interpolation densities of the control points are dynamically adjusted, and the core parameter library is iterated in reverse through the quantification of the fitting rate to achieve the self-optimization closed-loop evolution of the system.

[0016] This invention systematically addresses four major pain points—planning redundancy, manual inefficiency, environmental disconnect, and control jitter—through a five-part technical closed loop, realizing a complete path generation chain in mining scenarios: "precise construction → intelligent optimization → autonomous management → full-dimensional verification → dynamic evolution." Attached Figure Description

[0017] Figure 1This is a schematic diagram of the autonomous navigation path generation system for underground loaders based on front-end and back-end interaction according to the present invention; Figure 2 This is a schematic diagram of the path optimization and differentiation optimization process in the autonomous navigation path generation system for underground loaders based on front-end and back-end interaction of the present invention. Figure 3 This is a schematic diagram of the autonomous navigation path generation method for underground shovels based on front-end and back-end interaction according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1:

[0019] Please see Figure 1 As shown in this embodiment, the autonomous navigation path generation system for underground loaders based on front-end and back-end interaction includes: Multi-source fusion and path construction unit: It integrates multi-source map data to build a two-way spatial coordinate mapping engine that couples physical space and pixel space; it detects spatial conflicts between user-annotated path points and alleyway structures in real time and generates dynamic path skeletons.

[0020] Path segmentation and differentiated optimization unit: Based on the dynamic path skeleton, the path segment type is divided, and then the path is dynamically optimized, while redundant data is compressed in a synchronous manner to generate an optimized path package; Multi-route management and simulation-ready module: Constructs a route version control tree based on the optimized route package, dynamically allocates and encapsulates storage resources according to the tunnel topology, and generates a simulation-ready route file package; Simulation verification and traceability correction unit: Based on the simulation-ready path file package, it calculates the lateral error and minimum lane wall distance in real time and generates a verification report; then it captures out-of-limit events during the simulation trajectory tracking process, generates a dual-channel traceability report and reverse-marks the path segments that need to be corrected; Actual operating condition closed-loop evolution unit: Based on the hot update optimization rules of the dual-channel traceability report, quantify the path adaptability and iterate the core parameter library in reverse.

[0021] Acquire underground space GIS maps (raster maps stored in .pgm format, with grayscale values ​​representing spatial information (white for feasible areas, black for obstacles, and gray for unexplored areas)), ROS environment configuration files, and multi-sensor data streams (including 32-line LiDAR point clouds (providing high-precision 3D structure data of tunnels), thermal imaging frames (used to compensate for missing areas of LiDAR point clouds in dusty environments), and UWB positioning signals (assisting RFID tags in locating key points (such as ore passes and mining points)) as multi-source map data; Multimodal sensor fusion is performed on multi-sensor data streams to generate fused point cloud data; Specifically, 3D point cloud data of the tunnel is acquired using a 32-line lidar, and the point cloud data is preprocessed, including noise reduction (such as statistical filtering) and removal of outliers (such as dust interference points). Then, thermal imaging frames of the underground space are scanned by a thermal imaging camera, and the presence of high-temperature areas (e.g., >60°C) is detected. These areas are identified as suspected dust interference areas. The missing high-temperature areas in the point cloud data are then filled in using the neighborhood point interpolation method, thereby improving the integrity of the point cloud. Finally, by loading the coordinate table of the pre-embedded RFID tags (such as the preset ROS coordinates of the ore pass and mining point), when the UWB positioning signal strength is too high (such as >-70dBm), the pixel coordinates are calculated in reverse, and then the calibration mark is displayed on the front end. Sub-meter level key point positioning can be achieved through RFID tags, which can calibrate the deviation between the map and the actual physical coordinates; By combining GIS maps to rasterize point cloud data, feasible areas (white) and obstacle areas are marked (it should be noted that due to unknown risks in unexplored areas, direct passage is not recommended, so unexplored areas are included in obstacle areas, but there should be a distinction in the visualization interface, i.e. obstacle areas (black) and unexplored areas (gray)). Then, by combining thermal imaging frames in multi-sensor data streams, unexplored areas (such as landslide areas) are dynamically updated to generate real-time semantic maps. The ROS environment configuration file is parsed to obtain ROS environment parameters (including map resolution, ROS coordinates of the map origin (format [x,y,z], only the first two coordinates are used) and map pixel size). Based on the ROS environment parameters, a two-way spatial coordinate mapping engine between the ROS world coordinate system and the map pixel coordinate system is established to perform two-way transformation between physical space and pixel space and generate a coordinate transformation rule set (including the ros_to_pixel function and the pixel_to_ros function). Specifically, the spatial coordinate bidirectional mapping engine uses coordinate transformation functions in the coordinate transformation rule set to convert pixels into ROS coordinates and vice versa, ensuring that the physical location of the navigation path and the image presentation maintain a precise correspondence. The principle of ROS to pixel coordinate transformation is to establish a mathematical model by parsing map parameters. Its core calculation is completed by the ros_to_pixel function: this function obtains the corresponding pixel position by quantizing the physical space coordinates based on the map resolution and the origin ROS coordinates (origin_x, origin_y). The specific coordinate transformation function for ROS to pixel coordinates is ros_to_pixel(x_ros, y_ros):

[0022] x-axis transformation: ;

[0023] y-axis transformation: ; It should be noted that the y-axis of the ROS map is usually opposite to that of the image coordinate system (the ROS y-axis points upwards, while the image y-axis points downwards), so the y-axis needs to be flipped. In practical use, it is necessary to ensure that the resolution is small enough (e.g., 0.05m / pixel) to avoid coordinate quantization errors. Boundary checks are also required, that is, to verify whether it is within the map range (0≤x_pixel<map width, 0≤y_pixel<map height). )); The `pixel_to_ros` function performs coordinate inversion, converting pixels to ROS coordinates. This function performs an inverse coordinate transformation based on map data, and its transformation formula is as follows: The coordinate transformation function from pixel to ROS coordinates is pixel_to_ros(x_pixel, y_pixel): x_ros=origin_x+x_pixel×resolution; y_ros=origin_y+(map_height-1-y_pixel)×resolution; The y-axis direction in the ROS coordinate system needs to be restored using map_height-1-y_pixel. The conversion result may be a floating-point number, and it is necessary to decide whether to round it based on the requirements. Where x_ros and y_ros are the x and y coordinates in the ROS world coordinate system, respectively, representing the device's position in the actual physical world; origin_x and origin_y are the x and y coordinates of the map's origin in the ROS coordinate system, respectively, representing the map's starting position in the ROS world coordinate system; x_pixel and y_pixel are the x and y coordinates in the map's pixel coordinate system, respectively, representing the pixel's position in the map image; resolution (map resolution) represents the actual physical length represented by each pixel in the map; and map_height is the map's height, representing the total number of pixels in the vertical direction of the map image. It should be noted that in the ROS system, maps are usually stored in raster form (such as .pgm files), and their coordinate system differs from the physical world coordinate system. In order to achieve map rendering, path planning, or sensor data fusion, it is necessary to perform bidirectional conversion between ROS world coordinates (physical spatial coordinates in meters) and pixel coordinates (pixel positions in map images). Differences between the two coordinate systems: ROS world coordinate system: X-axis to the right, Y-axis to the top, and the origin is usually located at the bottom left corner of the map; This is a standard physical space coordinate system (similar to the Cartesian coordinate system in mathematics); Image / pixel coordinate system: X-axis to the right, Y-axis downwards, and the origin is usually located at the top left corner of the image.

[0024] This is a common convention in computer image processing (such as image libraries like OpenCV and PNG); The ros_to_pixel function converts ROS world coordinates to pixel coordinates for map rendering or image processing. The pixel_to_ros function converts pixel coordinates back to ROS world coordinates, which can be used for path point annotation or sensor data alignment. The core design principle of these two functions is based on the standard specifications of ROS raster maps and the coordinate system conventions of computer image processing. Their core logic is to perform a two-way mapping between pixel coordinates and physical space coordinates through linear transformation. Based on the pixel coordinates of the path points marked by the user and the real-time semantic map, the pixel coordinates are converted into ROS world coordinates by calling the coordinate transformation rule set, and the path points are checked to see if they are located in the feasible zone. If the area is feasible, add waypoints; if it is not feasible, trigger a conflict alarm and prevent waypoints from being added; synchronously update the user-annotated waypoint sequence (the updated waypoint sequence includes obstacle conflict detection results). By comparing the differences between the current semantic map and the historical semantic map (i.e., the difference in grayscale values ​​between the rasterized current semantic map and the historical semantic map; the historical time point of the historical semantic map can be manually set according to actual needs, such as the current time and the historical semantic map 10 minutes ago, or 1 day or 1 week ago), if a difference is detected (a threshold can be used to determine whether a difference exists, such as setting a threshold of 20% to determine areas where the difference between the current and historical semantic maps exceeds 20%), dynamic obstacle markers are added to the areas with differences (such as landslides), and incremental updates of the semantic map are triggered (to reflect environmental changes). It should be noted that this method can detect changes in underground tunnels in real time and flexibly, thus facilitating subsequent adjustments to route planning. Set up an adaptive recognition mechanism for lane width to obtain the effective passage width; The adaptive roadway width recognition mechanism calculates the effective passage width by subtracting the actual roadway width extracted from the point cloud data from twice the safety margin. The safety margin is a preset value, which can be a fixed value or half the width of the loader plus 0.3 meters as the safety margin, with 0.3 meters as an additional margin. The specific value can be set freely according to actual needs. This effective passage width is used to prevent insufficient passage capacity in narrow alleyways, that is, due to the dynamic changes in alleyway width, path planning fails in narrow areas (such as <2.5 meters); If the effective passage width does not meet expectations (e.g., <2.5 meters), it is marked as a narrow area and simultaneously added to the semantic map as a lane width label; Based on the updated path point sequence and the dynamic obstacle markers and lane width annotations in the semantic map, the curvature between every three consecutive path points in the path point sequence is obtained (which can be calculated using the curvature formula). If the curvature does not meet expectations (set according to actual needs, such as greater than 0.8 rad / m), mark it as a turning point; if the waypoint is located in a dynamic obstacle marking area or a narrow area, forcibly mark it as a turning point and add a turning direction prompt (such as a blue arrow). Integrate the tagged path point sequence and semantic map to generate a dynamic path skeleton, which includes path type (i.e., the status of path points, including normal points, turning points, and obstacle points), lane width data (effective passage width and lane width threshold), and dynamic obstacle markers (including the coordinates and radius of dynamic obstacle areas (such as landslide areas) and narrow areas). Based on the dynamic path skeleton, a dynamic topology map is constructed by building a tunnel structure model, thereby supporting the width constraint of path planning; and a visualization rendering is performed by a visualization engine (such as OpenCV or ROS Rviz) to generate a map rendering interface (displaying the visualization effect of the underground space map (divided into white / black / gray three colors, and path point types distinguished by different colors); Specifically, the color mapping rules for visualization rendering are as follows: White: Feasible area (lanes, open areas); Black: Obstacles (rocks, walls); Gray: Unexplored area (unknown area); Green - Normal point, Blue - Turning point, Red - Obstacle point; In the visual map interface, users can drag waypoints to adjust their positions. During the dragging process, the ROS coordinates of the waypoints are updated and the display is refreshed in real time. This function is implemented through the on_press, on_motion, and on_release functions. The on_press function searches for the nearest waypoint when the mouse is pressed, and if found, dragging begins. The on_motion function updates the position of the selected waypoint when the mouse moves. The on_release function ends the dragging when the mouse is released. Extract the path point sequence and lane width data from the dynamic path skeleton, align them, and ensure that each path point is associated with the effective passage width of the lane it belongs to; Based on the dynamic path skeleton, safety margin constraints are constructed and added to the road segments traversed by the path points. If the roadway where the path point is located is a narrow area, it is marked as a narrow road segment. If the path point is located in a dynamic obstacle area, it is forcibly marked as a narrow road segment. The minimum radius of curvature is obtained by adaptive adjustment based on the roadway width. Minimum radius of curvature = max(minimum safety threshold, adjustment coefficient × effective passage width); For example, if the effective passage width is 3m, the adjustment coefficient is 0.8, and the minimum safety threshold is 2.0, then the minimum radius of curvature is max(2.0, 2.4) = 2.4m; if the effective passage width is 2.0m, then the minimum radius of curvature is max(2.0, 1.6) = 2.0m. The advantage of this design is that the adjustment coefficient × effective passage width can ensure that the radius of curvature is dynamically adjusted with the effective passage width of the roadway, and max(2.0,...) can guarantee the minimum safety threshold (e.g., 2.0m); the minimum safety threshold here is the turning limit of the loader hinge; Based on safety margin constraints, straight-ahead segments and turning segments are adaptively divided to obtain road segment classifications with width constraints (straight-ahead segment: includes waypoint sequence, effective traffic width, minimum radius of curvature, and safety margin constraint status; turning segment: includes waypoint sequence, effective traffic width, minimum radius of curvature, whether it is narrow, and alarm information). Specifically, the vector angle between every three consecutive path points in the path point sequence is obtained. If the road segment where the path point is located is not a narrow road segment and the vector angle meets the expectation (i.e., it is greater than the preset angle threshold, such as >25°), it is marked as a straight segment. The straight segment must be greater than or equal to the minimum radius of curvature; otherwise, an alarm is triggered. If a path point is located in a narrow road segment or the vector angle does not meet expectations, it is marked as a turning segment. If the turning segment is smaller than the minimum radius of curvature, an alarm is triggered and a prompt is made to adjust the road segment. Limiting the minimum radius of curvature ensures that the minimum turning radius in narrow tunnels always meets the safety requirements of the loader. If the path point is located in a dynamic obstacle area such as a landslide area, it will still be forcibly marked as a narrow section even if the width meets the requirements, indicating safety. By associating the vector angle with the turning point, and by identifying whether the vector angle meets the expectations, combined with the effective passage width information of the tunnel, misjudgments caused by relying solely on the angle (such as slight turns in wide areas) can be avoided. In the visualization rendering interface, the straight sections and turning sections are marked (e.g., the straight sections are marked with green dashed lines and the turning sections are marked with red solid lines), and the narrow areas are marked (e.g., highlighted in yellow) and the minimum radius of curvature is displayed. Based on the road segment classification results, differentiated optimization is performed on straight-ahead segments and turning segments; For straight sections, increase the density of waypoints and optimize the shape of the path curve; Specifically, the number of intermediate points between the first and last points of a straight segment is counted. After identifying these intermediate points, linear interpolation is performed on the first and last points of the straight segment to generate intermediate path points, thereby increasing the path point density and improving navigation accuracy. Then, polynomial fitting or spline interpolation (such as B-spline smoothing) is used to fit the interpolated path points to generate a smooth trajectory curve. Adjust the lateral offset of the path according to the effective passage width of the lane to ensure that the path is always within the feasible area and avoid the risk of turning due to the path being too close to the lane boundary. For the turning section, curvature constraint treatment is applied. Specifically, the turning segment is fitted with B-spline three times to generate a smooth trajectory. During the B-spline fitting, the first and last points of the path segment are given higher weights (e.g., the weight of the turning segment in a narrow area is 3 times the original weight, and the weight of the turning segment in a non-narrow area is 2 times), thereby reducing the deviation of the starting point and the ending point during the fitting process. For example, the weight of the middle point in the turning segment is 1, and the weight of the starting point and the ending point is set to 2, thereby strengthening the starting point constraint and the ending point constraint. It should be noted that the weight parameters of the B-spline control the degree to which each path point affects the curve. The higher the weight of the first and last points, the closer the curve is to these points, thus avoiding the trajectory from deviating from the roadway boundary due to the degree of freedom in the fitting process. Based on the optimization results of the path points, curvature safety verification and iterative adjustment are performed. The curvature of the fitted path is calculated. If the path curvature is greater than the reciprocal of the minimum curvature radius (i.e., >1 / minimum curvature radius), iterative adjustment of the path point position is triggered until the safety requirements are met (i.e., the path curvature is less than or equal to the reciprocal of the minimum curvature radius). Specifically, in the vicinity of the region where curvature exceeds the limit, the position of the path point is finely adjusted to reduce the curvature value. If this local adjustment is ineffective, the path point sequence is recalculated to ensure that all path points satisfy the condition that the path curvature is less than or equal to the reciprocal of the minimum curvature radius. Traverse the optimized path point sequence, delete redundant path points whose distance between adjacent path points does not meet the expectations (e.g., calculate the Euclidean distance between adjacent path points, and if the distance between adjacent points is less than the distance threshold (e.g., 0.1m), delete the latter to reduce the path point density), obtain the compressed path point sequence, and generate the optimized path package (including coordinates, effective passage width, minimum radius of curvature, and security check status). Based on the optimized path package, a lightweight version control system is used to create independent branches for each tunnel path (such as branch route_001, branch route_002), and a version tag is added to each optimized version of the path, including a timestamp and version number (such as v1.0.0_20200101) to ensure that historical versions are traceable; Extract the lane topology (such as branch nodes and intersections) from the dynamic topology map, and dynamically allocate storage paths. For example, store multi-path data of the same lane in the same subdirectory; then, for lanes that are frequently updated (such as high dynamic obstacle areas with an update frequency greater than the update threshold), allocate higher storage priority (such as allocating SSD cache). When allocating storage paths, check if a file with the same name exists under the target storage path. If it does, automatically generate a unique file name. Add a file locking mechanism: Use the fcntl module (Linux) or Redis locks (distributed environment) to ensure the atomicity of concurrent writes; if a write fails, automatically roll back the updated path to the most recent stable version; Define the original coordinate system (pixel coordinates) and target coordinate system (ROS coordinates) of all path points in the optimized path package, and call the coordinate transformation rule set to perform coordinate transformation (pixel → ROS coordinates), synchronously record the correspondence between path points and device coordinate system, and then generate a coordinate system mapping table for each path; The integrated and converted path point data is exported in the format required by the simulation environment (such as CSV, YAML), ROS encapsulated, and the path data is encapsulated into ROS messages (such as nav_msgs / Path) using the rosbag tool to generate a simulation-ready path file package. Then, the generated simulation-ready path file is loaded in Gazebo or RViz to verify the matching with the tunnel topology. Based on the simulation-ready path file package, an optimized path is deployed in the ROS environment, and the map and control nodes are initialized to provide a data foundation for dual-loop verification. Specifically, the tunnel map in the simulation-ready path file package is loaded by the map server to ensure that the map coordinate system is consistent with the path data. The path tracing tool in the map server is run and the optimized path file is subscribed to. Then, control parameters such as aiming distance and maximum speed are set through the ROS parameter server. After deploying the optimized path, perform control layer verification and security layer verification. Control layer verification: The lateral error is calculated in real time using the Pure Pursuit algorithm, and the heading angle convergence performance is verified, while the heading angle convergence time is recorded synchronously. Specifically, the current pose of the equipment (loader) is obtained, and the shortest path point on the path is searched as the aiming point based on the aiming distance; the lateral error is calculated by trigonometric functions based on the geometric relationship between the loader center and the aiming point. After obtaining the lateral error, the initial heading angle deviation is recorded in the ROS callback function. Then, the heading angle deviation between the current heading angle and the optimal heading angle is calculated in real time and adjusted in real time. When the heading angle deviation is less than the preset convergence threshold (e.g., <0.1rad), it is determined that the heading angle has converged, and the heading angle convergence time (the time from the initial heading angle deviation to the heading angle convergence) is recorded. Safety layer verification: Based on the tunnel map, simulated laser point cloud scanning data, covering a 360° range, and denoising processing of the simulated point cloud is performed, including filtering invalid points and converting them into two-dimensional coordinates. The KD-tree algorithm is used to quickly find and obtain the nearest lane wall point to the equipment (i.e., the lane wall position closest to the loader), obtain the minimum lane wall distance, and then verify the path safety. One way to verify the safety of a path is through threshold analysis. That is, a safety distance threshold is set. If the minimum distance from the alley wall is greater than or equal to the safety distance threshold, the path safety verification is considered successful, and the subsequent process can continue. If the minimum distance to the roadway wall is less than the safe distance threshold, the roadway wall point is determined to be a dangerous area of ​​the roadway wall, a safety alarm is triggered and the abnormal event is recorded. The lateral error, heading angle convergence time, and minimum tunnel wall distance are integrated and written to a log file. Then, the average value and standard deviation of multiple simulation verifications are calculated to evaluate the path tracking accuracy and safety. Finally, a verification report is generated and visualized. One method of visualization processing is to perform visualization analysis using the RViz tool, comparing the planned path with the actual trajectory of the loader in the simulation, adding markers (such as red circles) in RViz to mark the convergence point of the heading angle and the dangerous area of ​​the tunnel wall, and displaying the minimum tunnel wall distance value in the visualization interface. Based on the visualization-processed verification report, extract the lateral error time series data (i.e., the lateral error of each roadway wall point passed by the loader during the simulation, organized into time series data according to time sequence), use Matplotlib or RViz visualization tools to plot the curve of lateral error changing with time, construct the time-error curve, and then identify and mark (e.g., mark in red) the path points where the lateral error exceeds the limit (i.e., the lateral error is greater than the preset lateral error threshold, such as >0.23m), as the lateral error exceedance points; The tunnel map is extracted from the simulation-ready path file package, rasterized, and then a two-dimensional array is created; that is, the map is divided into 0.1m×0.1m grids and a two-dimensional array is created. For each grid cell in the two-dimensional array, based on the minimum alleyway distance of the most recent simulated laser point cloud scan, the grid cell risk is updated according to the heat map algorithm (i.e., the weight value of each grid cell is obtained by inversely weighting by distance). Different colors (such as red-yellow-green gradient) are used to map the weight values ​​(such as red indicating the highest risk). Grid cells with weight values ​​higher than the preset risk threshold are marked as high-risk areas (red areas), and a spatial conflict heat map marked with high-risk areas is generated. Integrate time-error curves and spatial conflict heatmaps to generate dual-channel traceability reports; Based on the dual-channel traceability report, the high-risk areas of the lateral error exceedance point are combined with the high-risk areas of the spatial conflict heat map to locate high-risk path segments and generate a list of high-risk road segments. Specifically, the coordinates of all points where the lateral error exceeds the limit are extracted from the time-error curve, and the points where the lateral error exceeds the limit are associated with the high-risk grids in the heat map. The association method is as follows: map the points where the lateral error exceeds the limit to the spatial conflict heat map grid, and filter out the areas that meet the conditions that the points where the lateral error exceeds the limit are located in the current grid and the weight value of the current grid is higher than the risk threshold, and associate them as high-risk grids. The optimized path point sequence is used as the optimized path and divided into predefined path segments (such as path segment 1, path segment 2, ...). For each high-risk grid, the path segment it belongs to is matched and the path segment ID is marked (such as path segment 3) as a high-risk path segment. Display risk conflict heatmaps and optimized paths in RViz, with high-risk areas marked in red and text labels added to indicate high-risk path segments; Based on the dual-channel tracing report, the differentiated optimization parameters for the straight and turning segments are dynamically adjusted based on the convergence time of the lateral error direction and heading angle. Specifically, the weights of the control points of the B-spline curve are locally adjusted according to the direction of the lateral error to optimize the path's ability to compensate for lateral errors; if the loader is detected to yaw to the left, the weights of the control points on the right are increased, and vice versa. Then, the effect of the weight update was verified, including: recalculating the B-spline curve using the updated weights to obtain the corrected path points; comparing the time-error curves before and after the adjustment to verify the effect of the weight adjustment on improving the path tracking accuracy. The controller response speed is optimized by using adaptive interpolation density based on the heading angle convergence time. If the heading angle convergence time exceeds the expectation (e.g., >5 seconds), it is determined to be too slow. It is necessary to increase the number of interpolation points and dynamically insert additional interpolation points within the path segment to optimize path smoothness and controller tracking capability. The interpolation effect was verified by running the Pure Pursuit algorithm with the adjusted interpolation density and recording the convergence time of the new heading angle. The convergence time and lateral error before and after the adjustment were compared to ensure that the interpolation density adjustment effectively reduced the response latency. Based on the results of the hot update optimization rules, the path fit rate is quantified by the time-error curve, and the division accuracy of straight and turning segments and the identification accuracy of redundant path points are updated in reverse. Specifically, the area under the time-error curve is calculated by integrating the time-error curve, which is used as the path fit rate. If the path fit rate is low, the angle threshold of the turning segment vector is reduced to reduce the risk of sharp turns. The deletion strategy is adjusted according to the distribution of redundant path points (such as increasing or decreasing the distance threshold) to indicate the path simplicity. All updated parameters are integrated, written into a new generation of core parameter library, and encapsulated through ROS for adaptive iteration; the new parameter library is tested in narrow alleyway scenarios to ensure the accuracy and safety of path planning.

[0025] Example 2: Please see Figure 2 As shown, parts not described in detail in this embodiment are described in Embodiment 1. A method for generating autonomous navigation paths for underground loader operators based on front-end and back-end interaction is provided, including: S1: Integrates multi-source map data to build a two-way spatial coordinate mapping engine that couples physical space with pixel space; detects spatial conflicts between user-annotated path points and alleyway structures in real time and generates dynamic path skeletons; S2: Based on the dynamic path skeleton, road segment types are divided, and then the path is dynamically optimized, redundant data is compressed in a synchronous manner, and an optimized path package is generated. S3: Construct a route version control tree based on the optimized path package, dynamically allocate and encapsulate storage resources according to the tunnel topology, and generate a simulation-ready path file package; S4: Based on the simulation-ready path file package, calculate the lateral error and minimum lane wall distance in real time and generate a verification report; then capture the out-of-limit events in the simulation trajectory tracking process, generate a dual-channel traceability report and reverse-mark the path segments that need to be corrected; S5: Based on the dual-channel traceability report hot update optimization rules, quantify path adaptability and iterate the core parameter library in reverse.

[0026] Example 3: This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-described method for generating autonomous navigation paths for underground shovels based on front-end and back-end interaction.

[0027] Since the electronic device described in this embodiment is the one used to implement the autonomous navigation path generation method for underground loader based on front-end and back-end interaction in this application embodiment, those skilled in the art can understand the specific implementation and various variations of the electronic device in this embodiment based on the autonomous navigation path generation method for underground loader based on front-end and back-end interaction described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any electronic device used by those skilled in the art to implement the autonomous navigation path generation method for underground loader based on front-end and back-end interaction in this application embodiment falls within the scope of protection of this application.

[0028] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0029] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An autonomous navigation path generation system for underground loaders based on front-end and back-end interaction, characterized in that, include: Multi-source fusion and path construction unit: Fusion of multi-source map data to build a two-way spatial coordinate mapping engine that couples physical space and pixel space; Real-time detection of spatial conflicts between user-annotated path points and the tunnel structure, and generation of dynamic path skeletons.

2. Path segmentation and differentiated optimization unit: Based on the dynamic path skeleton, the road segment types are divided, and then the path is dynamically optimized, while simultaneously compressing redundant data and generating optimized path packages; Multi-route management and simulation-ready module: Constructs a route version control tree based on the optimized route package, dynamically allocates and encapsulates storage resources according to the tunnel topology, and generates a simulation-ready route file package; Simulation verification and traceability correction unit: Based on the simulation-ready path file package, it calculates the lateral error and minimum lane wall distance in real time and generates a verification report; Furthermore, it captures out-of-limit events during the simulation trajectory tracking process, generates a dual-channel traceability report, and reverse-marks the path segments that need to be corrected; Actual operating condition closed-loop evolution unit: Based on the hot update optimization rules of the dual-channel traceability report, quantify the path adaptability and iterate the core parameter library in reverse.

3. The autonomous navigation path generation system for underground loaders based on front-end and back-end interaction as described in claim 1, characterized in that, The methods for constructing a bidirectional spatial coordinate mapping engine to couple physical space and pixel space include: Acquire underground space GIS maps, ROS environment configuration files, and multi-sensor data streams as multi-source map data; Multimodal sensor fusion is performed on multi-sensor data streams to generate fused point cloud data; By combining GIS maps, point cloud data is rasterized, feasible areas and obstacle areas are marked, and semantic maps are generated. The ROS environment configuration file is parsed to obtain ROS environment parameters. Based on the ROS environment parameters, a two-way spatial coordinate mapping engine between the ROS world coordinate system and the map pixel coordinate system is established to perform bidirectional transformation between physical space and pixel space and generate a coordinate transformation rule set.

4. The underground loader autonomous navigation path generation system based on front-end and back-end interaction as described in claim 2, characterized in that, The methods for generating the dynamic path skeleton include: Based on the pixel coordinates of the path points marked by the user and the semantic map, the coordinate transformation rule set is called to convert the pixel coordinates into ROS world coordinates, and the path points are checked to see if they are located in the feasible area. If the area is feasible, allow the addition of waypoints; if it is not feasible, trigger a conflict alarm and prevent the addition of waypoints; synchronously update the sequence of waypoints marked by the user. Compare the current semantic map with the historical semantic map. If a difference is detected, add dynamic obstacle markers to the changed area and trigger an incremental update of the semantic map. Set up an adaptive recognition mechanism for lane width to obtain the effective passage width; If the effective passage width does not meet expectations, mark it as a narrow area and add it to the semantic map as a lane width annotation simultaneously; Based on the updated path point sequence and the dynamic obstacle markers and lane width annotations in the semantic map, the curvature between every three consecutive path points in the path point sequence is obtained. If the curvature does not meet expectations, mark it as a turning point; if the waypoint is located in a dynamic obstacle identification area or a narrow area, forcibly mark it as a turning point. Integrate the tagged waypoint sequence and semantic map to generate a dynamic path skeleton; Based on the dynamic path skeleton, a dynamic topology map is constructed and visualized, generating a map rendering interface.

5. The underground loader autonomous navigation path generation system based on front-end and back-end interaction as described in claim 3, characterized in that, The methods for classifying road segment types include: Based on the dynamic path skeleton, safety margin constraints are constructed and added to the road segments traversed by the path points. Based on safety margin constraints, straight sections and turning sections are adaptively divided to obtain road segment classification with width constraints.

6. The underground loader autonomous navigation path generation system based on front-end and back-end interaction according to claim 4, characterized in that, The methods for generating the optimized path package include: Based on the road segment classification results, differentiated optimization is performed on straight-ahead segments and turning segments; For straight sections, increase the density of waypoints and optimize the shape of the path curve; For the turning section, curvature constraint treatment is applied. Traverse the optimized path point sequence, delete redundant path points whose distances between adjacent path points do not meet expectations, obtain the compressed path point sequence, and generate the optimized path package.

7. The autonomous navigation path generation system for underground loader based on front-end and back-end interaction as described in claim 5, characterized in that, The generation methods for the simulation-ready path file package include: Based on the optimized path package, create an independent branch for each alleyway path and add a version tag to each path; Extract the alleyway topology from the dynamic topology map and dynamically allocate storage paths; Define the original coordinate system and target coordinate system for all path points in the optimized path package, and call the coordinate transformation rule set to perform coordinate transformation; The converted pathpoint data is integrated, packaged in ROS, and a simulation-ready path file package is generated.

8. The autonomous navigation path generation system for underground loaders based on front-end and back-end interaction according to claim 6, characterized in that, The verification report is generated in the following ways: Based on the simulation-ready path file package, the optimized path is deployed in the ROS environment, the lateral error is calculated in real time and the heading angle convergence performance is verified, and the heading angle convergence time is recorded synchronously. Simulate laser point cloud scanning and obtain the nearest alleyway wall point to the equipment to obtain the minimum alleyway wall distance; The system integrates lateral error, heading angle convergence time, and minimum tunnel wall distance to generate a verification report, which is then visualized.

9. The autonomous navigation path generation system for underground loader based on front-end and back-end interaction according to claim 7, characterized in that, The methods for generating dual-channel traceability reports and back-marking the path segments that need correction include: Based on the visualization-processed verification report, extract the horizontal error time series data, construct the time-error curve, and then identify and mark the horizontal error exceedance points; Extract the tunnel map from the simulation-ready path file package, perform rasterization, and then create a two-dimensional array; For each grid cell in the two-dimensional array, the grid risk is updated based on the minimum alleyway distance of the most recent simulated laser point cloud scan, and a spatial conflict heat map marking high-risk areas is generated. Integrate time-error curves and spatial conflict heatmaps to generate dual-channel traceability reports; Based on the dual-channel traceability report, the high-risk areas of the lateral error exceeding the limit are combined with the high-risk areas of the spatial conflict heat map to locate high-risk road segments, generate a list of high-risk road segments and perform visualization processing.

10. The autonomous navigation path generation system for underground loaders based on front-end and back-end interaction according to claim 8, characterized in that, The hot update optimization rules include the following methods: According to the dual-channel tracing report, the differentiated optimization parameters for the straight and turning segments are dynamically adjusted based on the convergence time of the lateral error direction and heading angle.

11. The autonomous navigation path generation system for underground loaders based on front-end and back-end interaction according to claim 9, characterized in that, The method of quantizing the path adaptively and iterating backward through the core parameter library includes: Based on the results of the hot update optimization rules, the path fit rate is quantified by the time-error curve, and the division accuracy of straight and turning segments and the identification accuracy of redundant path points are updated in reverse. All updated parameters are integrated, written into a new generation of core parameter library and encapsulated for adaptive iteration.

Images