Target point exploration method and system for robot automatic mapping

Abstract

The invention belongs to the technical field of robots, and relates to a target point exploration method and system for robot automatic mapping, and the method comprises the steps: obtaining data collected by a depth camera at the current moment, converting the data into a grid map, and preprocessing the grid map into a gray image; performing multi-level edge detection and moment analysis on the image, and extracting and screening a leading edge point to be explored; designing and carrying out data management and dynamic radiation radius strategy optimization based on an accessed navigation point data list, a failed navigation point data list and an effective access point data list, obtaining an effective access front point, constructing a multi-factor comprehensive evaluation system, carrying out quantitative scoring on the effective access front point, and selecting an optimal access point; exploring by taking the optimal access point as a target, and monitoring environment information of a leading edge point on an exploring path in real time, so as to judge whether to trigger a re-evaluation process or not. According to the invention, boundary point selection can be optimized, exploration efficiency and map integrity can be improved, and the autonomous navigation capability of the robot in an unknown environment can be enhanced.

Description

Technical Field

[0001] This invention belongs to the field of robotics technology and relates to a method and system for automatic mapping and target point exploration in robots. Background Technology

[0002] With the rapid development of robotics technology, robots are increasingly being used in numerous fields such as industrial manufacturing, commercial services, and home life. However, indoor robots still face many serious challenges in autonomous navigation in unknown environments. In terms of map building, traditional methods are not only inefficient but also lack versatility, making it difficult to quickly and accurately construct complete and precise indoor environment maps. This results in a slow process of environmental cognition for robots, making it unable to quickly adapt to complex and ever-changing new environments.

[0003] Existing automated mapping solutions mostly employ boundary point selection strategies, which identify boundary points representing the intersection of unknown and traversable areas, then drive the robot to those points, converting sensor data into map data during the process. However, such solutions only stop at boundary point detection and lack optimization mechanisms for these points. In practical applications, problems often arise due to unreasonable boundary point selection: as the robot moves forward, the system may detect a new boundary point behind it, causing the robot to change direction and move backward; upon reaching the rear, it may then detect a new boundary point again in the original forward direction, resulting in frequent back-and-forth movements, planning numerous redundant paths, and significantly reducing mapping efficiency.

[0004] Furthermore, existing solutions only consider empty and unknown areas when defining boundary points, which often leads to the omission of areas behind obstacles. Since these areas are often the boundaries between obstacles and unknown regions, they cannot be effectively identified using existing detection logic. In actual robot tasks, such as emergency rescue and warehouse inspection, this can result in missing information about critical areas, posing significant application drawbacks. Summary of the Invention

[0005] To address the aforementioned technical problems in the existing technology, this invention proposes a target point exploration method and system for automatic robot mapping, the specific technical solution of which is as follows:

[0006] A method for target point exploration in automatic robot mapping includes:

[0007] Step 1: Acquire the data collected by the depth camera at the current moment, convert it into a raster map, and preprocess it into a grayscale image;

[0008] Step 2: Use OpenCV to perform multi-level edge detection on the image, extract leading edge points, and then use moment analysis to quantify and evaluate the extracted leading edge points to select the leading edge points to be explored.

[0009] Step 3: Design and optimize the data management and dynamic radiation radius strategy based on the list of visited navigation points, the list of failed navigation points, and the list of valid access points. Obtain valid access frontier points, construct a multi-factor comprehensive evaluation system including unknown area density, known area distribution, obstacle area density, exploration distance, robot size, and current heading, and combine the three designed lists to quantify and score the valid access frontier points to select the best access points.

[0010] Step 4: Explore towards the best access point and monitor the environmental information of the leading points on the exploration path in real time to determine whether to trigger the re-evaluation process of Steps 1 to 3 and re-explore.

[0011] Furthermore, in step 1, the OpenCV library functions are used to perform image denoising and grayscale conversion operations. The data values ​​in the raster map are defined as follows: data values ​​of -1 are defined as unknown regions, data values ​​of 0-50 are defined as free regions, and data values ​​greater than 50 are defined as obstacle regions. The unknown regions, free regions, and obstacle regions are mapped to 128, 255, and 0 in the grayscale image, respectively.

[0012] Further, in step 2, the multi-level edge detection specifically includes: first, calling the cv::Canny function in the detectFrontiersWithOpenCV detection function, that is, using the Canny operator to perform preliminary edge detection on the grayscale image to obtain image edge information; then, using the cv::dilate function and an elliptical morphological structuring element to perform morphological dilation, wherein the size of the structuring element is determined according to the resolution of the raster map according to a preset ratio; using the cv::erode function to perform erosion operation on the dilated image to remove noise interference, connect broken edges, and create mask images of unknown and empty regions; after dilation, the intersection is taken; and combined with the Canny detection results, pixels that are adjacent to both the unknown and empty regions are selected as leading edge points.

[0013] Furthermore, in step 2, in the `detectFrontiersWithOpenCV` detection function, the moments of the leading edge point contour are calculated by calling the `cv::Moments` function to obtain the moment characteristics of the center of gravity and moment of inertia. Combining the position of the leading edge point, the distance and direction from the robot's current position, and other factors, the moment characteristics of each leading edge point are comprehensively analyzed through a preset evaluation model to select the leading edge points to be explored. The preset evaluation model is a regression model trained on a large amount of historical mapping data using a machine learning algorithm.

[0014] Furthermore, in step 3, the list of visited navigation points records: the target point successfully navigated to by the robot, the location information of the point, the range defined by the point and the robot's radiation radius, as well as the surrounding area's idle rate and obstacle distribution density calculated by traversing map data;

[0015] The list of failed navigation points records: the leading edge point of the robot navigation failure, the location information of the point, the robot's radiation radius, the idle rate of the surrounding area, the distribution density of obstacles, and the reason for the failure.

[0016] The list of valid access points records the following: the front points remaining after excluding the front points in the list of accessed navigation points, the list of failed navigation points, and the front points located in narrow passages;

[0017] The strategy for determining whether the leading edge point is in a narrow passage is as follows: starting from the current point, find the nearest obstacle points in its upper, lower, left, and right directions, and calculate the 1-norm of its left and right points:

[0018]

[0019] The 1-norm of the upper and lower points is:

[0020]

[0021] If the robot's physical size is greater than the left-right 1-norm and up-down 1-norm of the leading edge point, then the robot is determined to be in a narrow passage and needs to be excluded.

[0022] Furthermore, in step 3, the multi-factor comprehensive evaluation system quantifies and scores each factor by setting different scoring coefficients, as shown in the following expression:

[0023]

[0024] Where n1 is the number of unknown regions near the leading edge, w1 is the score coefficient for unknown regions (default setting is 5); n2 is the number of free regions near the leading edge, w2 is the score coefficient for free regions (default setting is 1); n3 is the number of obstacle regions near the leading edge, w3 is the score coefficient for obstacles (default setting is -2); d is the Euclidean distance between the leading edge and the robot, w... d The distance score coefficient is set to 1 by default; h is the difference between the heading at the leading edge and the current heading angle; w h This is the heading score coefficient, set to -5 by default; d s r is the distance from the leading edge point to the points in the reached list. s The dynamic radiation radius corresponding to the points in the visited list is d, where n is the number of points in the visited list and d is the number of points in the visited list. f r is the distance from the leading edge to the points in the failure list.f is the dynamic radiation radius corresponding to the point in the failure list, and m is the number of points in the failure list.

[0025] Furthermore, the multi-factor comprehensive evaluation system introduces a distance penalty mechanism. For forward points that are close to old navigation points (i.e., visited or failed points), quantitative deductions are made based on their distance from the old navigation points, with greater deductions for closer points.

[0026] Furthermore, in step 3, the dynamic radiation radius strategy optimization is specifically as follows: for visited points in the visited navigation point data list, if their surrounding environment is complex and obstacles are densely distributed, the robot system automatically expands their radiation radius; while for points in the failed navigation point data list that have failed to be explored multiple times, their radiation radius is significantly increased according to the number of failures and the severity.

[0027] Furthermore, in step 4, when the robot is exploring towards the target leading edge point, once it detects that the state of the area surrounding the leading edge point changes from unknown to free or an obstacle is detected, that is, the information of the current point is known and it has lost its exploration value, the re-evaluation process is immediately triggered to re-select and determine the best access point.

[0028] A target point exploration system for automatic mapping by a robot includes:

[0029] The data acquisition and preprocessing module is used to acquire environmental information from the depth camera, convert and preprocess it into a grid map, and perform further preprocessing.

[0030] The leading edge detection module is used to perform multi-level edge detection and moment analysis to acquire and filter leading edge points.

[0031] The comprehensive scoring module is used to construct a multi-dimensional comprehensive scoring system to quantitatively score cutting-edge points;

[0032] The database management module is used to manage the access navigation point data list, the failed navigation point data list, and the valid access point data list, and to perform frontier point filtering and score calculation.

[0033] The dynamic evaluation module is used to monitor changes in the environmental status of frontier points in real time, triggering re-evaluation and selection of the best access points.

[0034] Beneficial effects: This invention can optimize the selection of frontier points, improve exploration efficiency and map integrity, and has the advantage of enhancing the robot's autonomous navigation ability in unknown environments. Attached Figure Description

[0035] Figure 1 This is a flowchart of a target point exploration method for automatic mapping of robots in this embodiment. Detailed Implementation

[0036] To make the objectives, technical solutions, and technical effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0037] like Figure 1 As shown in the figure, an embodiment of the present invention provides a target point exploration method for automatic robot mapping, comprising: selecting a leading edge point based on map information constructed by a depth camera. This leading edge point selection takes into account factors such as unknown areas, feasible areas, obstacle density, reached points, unreachable points, distance to the leading edge point, robot physical dimensions, and robot heading. Specifically, it includes the following steps:

[0038] Step 1: Obtain the environmental information collected by the depth camera at the current moment, convert it into raster map data, and perform preprocessing.

[0039] In this context, the data value corresponding to a grid is the probability of an obstacle. For example, if the data value in a certain grid map is 51, it means that the probability of an obstacle in that grid is 51% and the probability of being idle is 49%. In this embodiment, a data value of -1 is defined as an unknown area, a data value of 0-50 is defined as an idle area, and a data value greater than 50 is defined as an obstacle area.

[0040] Unknown regions, free regions, and obstacle regions are mapped to 128, 255, and 0 in a grayscale image, respectively. OpenCV library functions are then used to perform image denoising and grayscale conversion, accurately distinguishing the boundaries of each region and providing accurate foundational data for subsequent exploration. Simultaneously, the data is updated in real-time, synchronously updating the robot's current position information to accurately distinguish the boundaries of each region, ensuring the real-time performance and accuracy of region recognition.

[0041] Step 2 employs a multi-layered edge detection strategy. First, the `cv::Canny` function is called within the `detectFrontiersWithOpenCV` function to perform preliminary edge detection on the grayscale image using the Canny operator, obtaining image edge information. Then, the `cv::dilate` function and an elliptical morphological structuring element are used for morphological dilation. The size of the structuring element is determined according to a preset ratio based on the resolution of the raster map. Next, the `cv::erode` function is used to erode the dilated image, removing noise interference and connecting broken edges. Specifically, mask images of unknown and empty regions are created, dilated separately, and their intersection is taken. Combined with the Canny detection results, pixels adjacent to both the unknown and empty regions are selected as leading edges, achieving high-precision extraction of leading edge information.

[0042] A moment analysis method is then introduced to quantitatively evaluate the extracted frontier points. In the `detectFrontiersWithOpenCV` function, the moments of the frontier point contours are calculated by calling the `cv::Moments` function, obtaining moment features such as the center of gravity and moment of inertia. Combining the position of the frontier point, its distance from the robot's current position, and its orientation, a pre-defined evaluation model is used to comprehensively analyze the moment features of each frontier point. This comprehensive assessment considers the importance and exploratory value of the frontier points, selecting the most significant frontier points to guide the robot in its next exploration action. The pre-defined evaluation model is a regression model trained on a large amount of historical mapping data using machine learning algorithms.

[0043] Step 3 involves designing and selecting the optimal exploration evaluation function at the crucial stage of issuing robot exploration instructions. This function bears the core screening responsibility.

[0044] The system design includes three data lists: is_visited, is_failed, and is_valid, with corresponding databases for visited navigation points, failed navigation points, and valid navigation points. The exploration process is optimized through refined data management and dynamic strategies. A multi-dimensional comprehensive scoring system is also constructed, considering not only the three lists but also other factors, including unknown area density, known area distribution, obstacle area density, exploration distance, robot size, and current heading. The three data lists serve two purposes: 1. Accelerating the leading edge selection logic; if a subsequently detected leading edge is in its corresponding list, it is skipped. 2. Providing evaluation data for the three-dimensional comprehensive assessment, primarily determining whether a newly selected leading edge falls within the radiation radius recorded in its corresponding list; if so, calculating the corresponding score or deduction.

[0045] First, all detected leading points are filtered, excluding those in the `is_visited` and `is_failed` lists to avoid duplicate calculations. Then, leading points located in narrow passages are identified and excluded, and the remaining points are added to the `is_valid` list. Next, leading points are scored based on multiple factors, including the radiation radius of visited and failed points, the density of unknown areas, the density of obstacle areas, the exploration distance, the robot's size, and the current heading, resulting in a score for each leading point. This process selects truly valuable and effective visiting points, i.e., the optimal visiting points.

[0046] Specifically, at the data structure level, each data list stores custom data units containing rich environmental information.

[0047] When the robot successfully reaches the target point and completes its exploration, the robot system extracts the precise location information of that point and delineates a specific area centered on that point. It then calculates the idle rate and obstacle density of the surrounding area by traversing the map data, while recording a preset radiation radius. The idle rate is the proportion of idle grid cells to the total number of grid cells, and the obstacle density is the proportion of obstacle grid cells to the total number of grid cells. This information is encapsulated into data units and added to the `is_visited` list to complete the real-time update of visited point data.

[0048] If the robot encounters a navigation failure, the system will also record the location of the failure point, the robot's radius of influence, the surrounding idle rate, the density of obstacles, and a detailed record of the reason for the failure, such as the path planning being unsolvable or encountering an insurmountable narrow area, forming a complete failure point data unit, storing it in the is_failed list, and building a failure navigation point database.

[0049] To further improve exploration efficiency, this invention employs a dynamic radiation radius design strategy. This strategy intelligently adjusts based on historical data and environmental characteristics: for visited points, if the surrounding environment is complex and obstacles are densely distributed, the robot's radiation radius is automatically expanded to avoid repeated exploration in similar complex areas; while for points that have failed to be explored multiple times, the radiation radius is significantly increased according to the number and severity of failures, reducing the risk of the robot getting stuck again.

[0050] Through the synergistic operation of the aforementioned dynamic database management and radiation radius optimization strategies, the system effectively reduces ineffective exploration paths and significantly improves overall exploration efficiency.

[0051] The strategy for determining whether it is at the forefront point of a narrow passage is as follows:

[0052] After excluding reached and failed points, among the remaining leading points, each point is traversed to determine if it is in a narrow passage. The rule for determination is to find the nearest obstacle point in all directions (up, down, left, and right) starting from the current point. Here, left is defined as point 1, right as point 2, up as point 3, and down as point 4. The 1-norm of its left and right points is then calculated.

[0053]

[0054] The 1-norm of the upper and lower points is:

[0055]

[0056] If the robot's physical size is greater than the left-right 1-norm and up-down 1-norm of the leading edge point, then the robot is determined to be in a narrow passage and needs to be excluded.

[0057] A strict distance penalty mechanism is introduced. For forward points that are close to old navigation points (i.e., visited or failed points), points are deducted quantitatively based on their distance from the old points. The closer the distance, the more points are deducted. This forces the robot to prioritize exploring more valuable unknown areas.

[0058] The `scoreFrontier` scoring function calculates the distance and angle between the leading edge point and the robot's current position. By assigning different weights, each factor is quantified and scored. For example, a leading edge point with higher unknown area density, moderate distance, and a small angle between its direction and the current heading receives a higher score, thus guiding the robot to prioritize exploring high-value and easily accessible areas. In the scoring function, the evaluation radius of the leading edge point is set to 10, resulting in a total of 317 pixels. The following formula is used to quantify and score each factor:

[0059]

[0060] Where n1 is the number of unknown regions near the leading edge, w1 is the score coefficient for unknown regions (default setting is 5); n2 is the number of free regions near the leading edge, w2 is the score coefficient for free regions (default setting is 1); n3 is the number of obstacle regions near the leading edge, w3 is the score coefficient for obstacles (default setting is -2); d is the Euclidean distance between the leading edge and the robot, w... d The distance score coefficient is set to 1 by default; h is the difference between the heading at the leading edge and the current heading angle; w h This is the heading score coefficient, set to -5 by default; d s r is the distance from the leading edge point to the points in the reached list. s The dynamic radiation radius corresponding to the points in the visited list is d, where n is the number of points in the visited list and d is the number of points in the visited list. f r is the distance from the leading edge to the points in the failure list. f is the dynamic radiation radius corresponding to the point in the failure list, and m is the number of points in the failure list.

[0061] Step 4: Design an effectiveness dynamic evaluation system to monitor changes in the environmental state of the leading edge points along the robot's exploration path in real time. As the robot moves towards the target leading edge point, if it detects that the surrounding area changes from unknown to free or an obstacle is detected (meaning the information about the current point is known and it has lost its exploration value), a re-evaluation process is immediately triggered. The system quickly calls upon the leading edge point detection system and the comprehensive scoring system, combining updated map data and environmental information to re-select and determine the best access point, effectively avoiding ineffective exploration and significantly reducing redundant path planning.

[0062] Based on the above method, this embodiment also provides a target point exploration system for automatic robot mapping, including:

[0063] The data acquisition and preprocessing module is used to acquire environmental information from the depth camera, convert and preprocess it into a grid map, and perform further preprocessing.

[0064] The leading edge detection module is used to perform multi-level edge detection and moment analysis to acquire and filter leading edge points.

[0065] The comprehensive scoring module is used to construct a multi-dimensional comprehensive scoring system to quantitatively score cutting-edge points;

[0066] The database management module is used to manage the access navigation point data list, the failed navigation point data list, and the valid access point data list, and to perform frontier point filtering and score calculation.

[0067] The dynamic evaluation module is used to monitor changes in the environmental status of frontier points in real time, triggering re-evaluation and selection of the best access points.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for target point exploration in automatic robot mapping, characterized in that, include: Step 1: Acquire the data collected by the depth camera at the current moment, convert it into a raster map, and preprocess it into a grayscale image; Step 2: Use OpenCV to perform multi-level edge detection on the image, extract leading edge points, and then use moment analysis to quantify and evaluate the extracted leading edge points to select the leading edge points to be explored. Step 3: Design and optimize the data management and dynamic radiation radius strategy based on the list of visited navigation points, the list of failed navigation points, and the list of valid access points. Obtain valid access frontier points, construct a multi-factor comprehensive evaluation system including unknown area density, known area distribution, obstacle area density, exploration distance, robot size, and current heading, and combine the three designed lists to quantify and score the valid access frontier points to select the best access points. Step 4: Explore towards the best access point and monitor the environmental information of the leading points on the exploration path in real time to determine whether to trigger the re-evaluation process of Steps 1 to 3 and re-explore.

2. The target point exploration method for automatic robot mapping as described in claim 1, characterized in that, In step 1, OpenCV library functions are used to perform image denoising and grayscale conversion. The data values ​​in the raster map are defined as follows: data values ​​of -1 are defined as unknown regions, data values ​​of 0-50 are defined as free regions, and data values ​​greater than 50 are defined as obstacle regions. The unknown regions, free regions, and obstacle regions are mapped to 128, 255, and 0 in the grayscale image, respectively.

3. The target point exploration method for automatic mapping of robots as described in claim 1, characterized in that, In step 2, the multi-level edge detection specifically includes: first, calling the cv::Canny function in the detectFrontiersWithOpenCV detection function, that is, using the Canny operator to perform preliminary edge detection on the grayscale image to obtain image edge information; then, using the cv::dilate function and an elliptical morphological structuring element to perform morphological dilation, wherein the size of the structuring element is determined according to the resolution of the raster map according to a preset ratio; using the cv::erode function to perform erosion operation on the dilated image to remove noise interference, connect broken edges, and create mask images of unknown and empty regions; after dilation, the intersection is taken; combined with the Canny detection results, pixels that are adjacent to both the unknown and empty regions are selected as leading edge points.

4. The target point exploration method for automatic robot mapping as described in claim 1, characterized in that, In step 2, the `detectFrontiersWithOpenCV` function calculates the moments of the leading edge profile by calling the `cv::Moments` function, and obtains the moment characteristics of the center of gravity and moment of inertia. Combining the position of the leading edge, its distance from the robot's current position, and direction, the moment characteristics of each leading edge are comprehensively analyzed using a preset evaluation model to select the leading edge points to be explored. The preset evaluation model is a regression model trained on a large amount of historical mapping data using a machine learning algorithm.

5. The target point exploration method for automatic robot mapping as described in claim 1, characterized in that, In step 3, the list of visited navigation points records: the target point successfully navigated to by the robot, the location information of the point, the range and the robot's radiation radius centered on the point, and the surrounding area's idle rate and obstacle distribution density calculated by traversing map data; The list of failed navigation points records: the leading edge point of the robot navigation failure, the location information of the point, the robot's radiation radius, the idle rate of the surrounding area, the distribution density of obstacles, and the reason for the failure. The list of valid access points records the following: the front points remaining after excluding the front points in the list of accessed navigation points, the list of failed navigation points, and the front points located in narrow passages; The strategy for determining whether the leading edge point is in a narrow passage is as follows: starting from the current point, find the nearest obstacle points in its upper, lower, left, and right directions, and calculate the 1-norm of its left and right points: The 1-norm of the upper and lower points is: If the robot's physical size is greater than the left-right 1-norm and up-down 1-norm of the leading edge point, then the robot is determined to be in a narrow passage and needs to be excluded.

6. The target point exploration method for automatic mapping of robots as described in claim 5, characterized in that, In step 3, the multi-factor comprehensive evaluation system quantifies and scores each factor by setting different scoring coefficients, as shown in the following expression: Where n1 is the number of unknown regions near the leading edge, w1 is the score coefficient for unknown regions (default setting is 5); n2 is the number of free regions near the leading edge, w2 is the score coefficient for free regions (default setting is 1); n3 is the number of obstacle regions near the leading edge, w3 is the score coefficient for obstacles (default setting is -2); d is the Euclidean distance between the leading edge and the robot, w... d The distance score coefficient is set to 1 by default; h is the difference between the heading at the leading edge and the current heading angle; w h This is the heading score coefficient, set to -5 by default; d s r is the distance from the leading edge point to the points in the reached list. s The dynamic radiation radius corresponding to the points in the visited list is d, where n is the number of points in the visited list and d is the number of points in the visited list. f r is the distance from the leading edge to the points in the failure list. f is the dynamic radiation radius corresponding to the point in the failure list, and m is the number of points in the failure list.

7. The target point exploration method for automatic mapping of robots as described in claim 6, characterized in that, The multi-factor comprehensive evaluation system introduces a distance penalty mechanism. For frontier points that are close to old navigation points (i.e., visited or failed points), quantitative points are deducted based on their distance from the old navigation points, with more points deducted for closer points.

8. The target point exploration method for automatic mapping of robots as described in claim 1, characterized in that, In step 3, the dynamic radiation radius strategy optimization is specifically as follows: for visited points in the visited navigation point data list, if the surrounding environment is complex and obstacles are densely distributed, the robot system automatically expands its radiation radius; while for points in the failed navigation point data list that have failed to be explored multiple times, the radiation radius is significantly increased according to the number of failures and the severity.

9. The target point exploration method for automatic mapping of robots as described in claim 1, characterized in that, In step 4, when the robot is exploring towards the target leading edge point, once it detects that the state of the area surrounding the leading edge point changes from unknown to free or an obstacle is detected, that is, the information of the current point is known and it has lost its exploration value, the re-evaluation process is immediately triggered to re-select and determine the best access point.

10. A system employing the target point exploration method for automatic robot mapping according to any one of claims 1 to 9, characterized in that, include: The data acquisition and preprocessing module is used to acquire environmental information from the depth camera, convert and preprocess it into a grid map, and perform further preprocessing. The leading edge detection module is used to perform multi-level edge detection and moment analysis to acquire and filter leading edge points. The comprehensive scoring module is used to construct a multi-dimensional comprehensive scoring system to quantitatively score cutting-edge points; The database management module is used to manage the access navigation point data list, the failed navigation point data list, and the valid access point data list, and to perform frontier point filtering and score calculation. The dynamic evaluation module is used to monitor changes in the environmental status of frontier points in real time, triggering re-evaluation and selection of the best access points.

Images