Unmanned sweeper missing sweeping path planning method and system based on semantic topology factors

Abstract

The invention discloses a method and system for planning a sweeping missing path of an unmanned sweeper based on a semantic topology factor, and relates to the technical field of planning and control of a sweeping path of an automatic driving vehicle, and the method comprises the following steps: 1, constructing an initial sweeping path capable of covering a whole operation area; step 2, carrying out real-time semantic recognition on a missing scanning area, judging that the sweeper deviates according to a deviation threshold value, sequentially obtaining a deviation point set, a convex hull point set and an obstacle target set, determining the boundary of the missing scanning area and setting a semantic missing scanning area block set; step 3, constructing a semantic topology model of the missing scanning area, packaging semantic missing scanning area blocks into nodes, and then constructing a semantic topological graph and a node distance matrix according to requirements; and step 4, full-coverage complementary scanning path planning: calculating the distance of each node connection edge, and planning a shortest area complementary scanning path sequence. In this way, in the sweeping operation of the sweeping vehicle, the sweeping missing area is efficiently recognized, and the sweeping supplementing path is efficiently planned.

Description

Technical Field

[0001] This invention relates to the field of cleaning path planning and control technology for autonomous vehicles, and in particular to a method and system for planning missed cleaning paths for unmanned cleaning vehicles based on semantic topology factors. Background Technology

[0002] In unstructured road environments, such as parks, streets, residential areas, and industrial parks, the application of autonomous vehicle cleaning is becoming increasingly common. Currently, the static full-coverage path planning method is commonly used for cleaning vehicle path planning. This method relies on a pre-built environmental map to plan a path that can cover the entire work area, and the cleaning vehicle can clean the work area according to the predetermined route.

[0003] However, in actual cleaning operations, unstructured road environments exhibit high complexity and dynamism. For example, dynamic obstacles such as pedestrians, pets, and moving vehicles, as well as temporary obstacles such as suddenly appearing construction areas and randomly placed debris, frequently and randomly appear on the cleaning path. These dynamic and static obstacles severely interfere with the normal operation of autonomous cleaning vehicles, making it difficult for them to strictly follow the pre-planned static path to complete the cleaning task, ultimately leading to missed areas.

[0004] In existing technologies, when missed areas are detected, there is a lack of efficient missed area identification mechanisms, making it difficult to quickly and accurately locate missed areas in complex and ever-changing environments. In addition, there is a lack of efficient path replanning methods, making it impossible to quickly and accurately generate new paths that can cover missed areas. This greatly affects the overall efficiency and quality of vehicle cleaning operations, making it difficult to achieve the ideal cleaning results. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, one of the objectives of this invention is to provide a method for planning the missed sweeping path of an unmanned sweeping vehicle based on semantic topological factors.

[0006] One of the objectives of this invention is achieved through the following technical solution: a method for planning missed sweeping paths for unmanned sweeping vehicles based on semantic topological factors, comprising the following method steps:

[0007] Step 1: Construct the initial cleaning path. Based on the pre-built environmental planning map and vehicle operation requirements, construct an initial cleaning path that can cover the entire work area.

[0008] Step 2: Real-time semantic recognition of missed areas, real-time acquisition of vehicle pose information when the sweeper is traveling along the initial sweeping path, when the distance between the sweeper's position and the nearest point on the initial sweeping path is greater than the preset offset threshold, it is determined that the vehicle has deviated, and the set of deviation points is obtained based on the vehicle's deviation trajectory.

[0009] Based on the set of deviation points and the initial cleaning path, convex hull calculation is performed to obtain the set of convex hull points, which is used to determine the boundary of the missed area. Then, target data within the missed area is collected and calculated to obtain the set of obstacle targets. The semantic features of the area are statistically obtained according to their target type information. At least the target type, target quantity, and target area of ​​the semantic features of the area are combined to set the semantic missed area block set.

[0010] Step 3: Construct a semantic topology model of the missed areas. Encapsulate each semantic missed area block set in Step 2 into a node, and assume that there is connectivity between each node. Construct a semantic topology graph based on the connectivity of each node and the cleaning business attributes of the sweeper. Then add the starting point of the sweeper to the semantic topology graph and construct a node distance matrix including the starting point of the sweeper.

[0011] Step 4: Full-coverage supplementary cleaning path planning. Calculate the distance between the connecting edges of every pair of nodes, including the starting point of the sweeper, in Step 3. This distance is used to plan the shortest supplementary cleaning path for missed areas after the initial cleaning path is completed, starting from the starting point of the sweeper and traversing all missed areas before returning to the initial position.

[0012] Furthermore, in step 2, after the initial cleaning path is completed and the supplementary cleaning begins, the deviation point is translated from the center of the rear axle of the sweeper to the left boundary point or the right boundary point of the sweeper.

[0013] At the same time, the initial cleaning path is translated and transformed into an initial cleaning path boundary point that is a certain distance to the left or right, which is used to enclose and calculate the boundary of the missed area with the left or right boundary point of the sweeper.

[0014] Furthermore, in step 2, let the real-time pose of the sweeper in the two-dimensional plane coordinate system be... , This indicates the location coordinates of the sweeper. Indicates the heading angle of the sweeper truck;

[0015] The formula for calculating the distance of a sweeper truck is as follows:

[0016] ;

[0017] Where, d i x represents the distance between the sweeper's position and the nearest point on the initial sweeping path; x, y represent the sweeper's real-time position coordinates; x i y i This represents the position coordinates of the i-th point on the initial cleaning path;

[0018] Based on the vehicle offset trajectory formed by the real-time position coordinates of the sweeper, the set of deviation points is defined as P. n, ,in, , represents the pose information of the j-th deviation point, j=0,1,…,n.

[0019] Furthermore, in step 2, let the set of convex hull points be... , ,in, , represents the pose information of the i-th convex hull point, i=0,1,2,…,h-1.

[0020] Furthermore, in step 2, sensors are used to collect target data outside the area, and deep learning algorithms are used to process the target data to obtain the set of obstacle targets;

[0021] The set of obstacle targets is represented as , ,in, This represents the first objective, and each objective has the following attributes:

[0022] Target location information: ;

[0023] Target type information: , ,

[0024] These respectively represent dynamically moving targets, static fixed targets, temporary obstacle targets, and targets with uncleaned debris;

[0025] Dynamic target speed information: ;

[0026] Based on the boundary of the missed scan region and the target type information of the target set, the semantic missed scan region block set is represented as follows: , ,in, This represents the i-th semantically missed scan region block.

[0027] Furthermore, each semantic missed scan region block contains a region boundary and region semantic features, wherein the region boundary is explicitly defined by the set of convex hull points;

[0028] Regional semantic features are obtained statistically based on the target type information. These regional semantic features can be represented as follows: ,in,

[0029] , represents the semantic feature information of dynamically movable targets in the missed scan area, including at least the target type, the area occupied, and the number;

[0030] , represents the semantic feature information of static fixed targets in the missed scan area, which includes at least the target type, occupied area, and quantity;

[0031] This represents the semantic feature information of temporary obstacle targets within the missed scan area, which includes at least the target type, the area it occupies, and the number of targets.

[0032] This indicates the garbage target information in the missed area.

[0033] Furthermore, in step 3, the cleaning service attributes include garbage cleaning service and water flushing cleaning service;

[0034] When carrying out garbage cleaning operations, if the area of ​​the missed area is occupied by dynamic moving targets that occupy more than 80% of the total area of ​​all non-garbage obstacle targets, the number of garbage targets is more than 3, and the area of ​​the missed area is more than 10 square meters, then it will be included in the effective cleaning area.

[0035] When performing flush cleaning and sweeping operations, areas with a missed sweeping area exceeding 10 square meters (excluding the target area with temporary obstacles) are included as valid areas for additional sweeping.

[0036] Furthermore, in step 3, the semantic topology graph expression corresponding to the garbage cleaning business is as follows: The semantic topology graph expression corresponding to the flushing and cleaning business is as follows: Where V is the node of the effective cleaning area under the corresponding cleaning business attribute, and Egarbage and Ewash are the sets of edges under the corresponding cleaning business attributes.

[0037] Furthermore, in step 3, based on the semantic topology graph, edges are connected between every pair of nodes, including the starting point of the sweeper, to construct the node distance matrix, where every pair of nodes is a group.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] The missed cleaning path planning method in this embodiment includes four core steps: constructing an initial cleaning path, real-time semantic recognition of missed cleaning areas, constructing a semantic topology model of the missed cleaning areas, and full-coverage supplementary cleaning path planning. These steps form a logical closed loop that is progressive, interlocking, and functionally independent.

[0040] Specifically, by performing semantic recognition and setting semantic missing scan region blocks for obstacles within the defined missing scan region boundary, compared with the traditional method of judging missing scans based solely on location information, this invention can more comprehensively understand the essence of the missing scan region, obtain rich semantic information, and quickly and accurately locate the missing scan region in complex and ever-changing environments.

[0041] Based on the determined missed scan area, the semantic topology graph and node distance matrix of the missed scan area are constructed to quantify the semantic relationship between the missed scan area and the surrounding environment and the topology graph. For example, the quantification of obstacle information and garbage information in the environment, as well as the quantification of the connection relationship between nodes and edges in the topology graph, are combined to fully consider the semantic information in the environment of the missed scan area.

[0042] Based on the semantic information of the missed scan areas, the distance to each node, garbage, and business attribute variables are considered during the node traversal search process. Thus, according to the semantic features and topological relationships of the missed scan areas, the direction and order of the new re-scanning path are automatically adjusted, and the optimal solution for path distance sorting is obtained. This allows for fast and accurate planning of re-scanning paths, prioritizing areas with more garbage in close proximity for re-scanning, avoiding unnecessary repeated trips, improving re-scanning efficiency, and ensuring that all missed scan areas are effectively covered, with a re-scanning coverage rate of over 98% for a single missed scan area.

[0043] To overcome the shortcomings of the prior art, the second objective of this invention is to provide a path planning system for unmanned sweeping vehicles that misses sweeping spots based on semantic topological factors.

[0044] The second objective of this invention is achieved through the following technical solution: a path planning system for unmanned sweeping vehicles that miss areas based on semantic topological factors, comprising:

[0045] The map generation module is used to construct an initial cleaning path that can cover the entire work area based on a pre-built environmental planning map and vehicle operation requirements;

[0046] The positioning module is used to acquire the vehicle's position and pose information in real time as the sweeper travels along the initial sweeping path, and send the acquired vehicle coordinates to the terminal; when the distance between the sweeper's position and the nearest point on the initial sweeping path is greater than a set offset threshold, it is determined that the vehicle has deviated, and the set of deviation points is obtained based on the vehicle's deviation trajectory.

[0047] The target calculation module is used to perform convex hull calculation based on the set of deviation points and the initial cleaning path to obtain the set of convex hull points, which is used to determine the boundary of the missed area; then, it collects and calculates the target data in the missed area to obtain the set of obstacle targets, and obtains semantic features according to their target type information. At least the target type, target quantity, and target area of ​​the semantic features are combined to set the set of semantic missed area blocks.

[0048] The topology model generation module is used to encapsulate each set semantic missed area block into several nodes, and assumes that there is connectivity between the nodes. Based on the connectivity of each node and the cleaning business attributes of the sweeper, a semantic topology graph is constructed. Then, the starting point of the sweeper is added to the semantic topology graph to construct a node distance matrix including the starting point of the sweeper.

[0049] The matrix calculation module is used to calculate the distance between the edges connecting any two nodes, including the starting point of the sweeper, and to sort the shortest area cleaning paths after traversing all areas from the starting point of the sweeper back to the initial position. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the missed scan path planning method in the first embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of the missed scan region in the real-time semantic recognition of the missed scan path planning method in the first embodiment of the present invention.

[0052] Figure 3 This refers to the pre-constructed, full-coverage environmental planning map in the first embodiment;

[0053] Figure 4 This is a schematic diagram of the obstacles in the missed scan area of ​​the missed scan path planning method in the first embodiment of the present invention;

[0054] Figure 5 This is a schematic diagram of the missed scan area nodes, the supplementary scan area, and the topology network of the missed scan path planning method in the first embodiment of the present invention;

[0055] Figure 6 This is a semantic topology map of the missed scan region in the missed scan path planning method of the first embodiment of the present invention;

[0056] Figure 7 This is a schematic diagram of the node distance matrix of the missed scan path planning method in the first embodiment of the present invention;

[0057] Figure 8 This is a schematic diagram illustrating the effect of the missed area re-scanning planning route in the missed scan area of ​​the missed scan path planning method in the first embodiment of the present invention;

[0058] Figure 9 This is a schematic diagram illustrating the effect of increasing the density of the planned route for supplementing the missed area in the missed area of ​​the missed area in the first embodiment of the present invention.

[0059] Figure 10 This is a block diagram of the module principle of the missed scan path planning system in the second embodiment of the present invention. Detailed Implementation

[0060] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0061] First Embodiment

[0062] like Figures 1 to 9As shown, a method for planning missed sweeping paths for unmanned sweeping vehicles based on semantic topological factors is applied to path planning during sweeping operations of autonomous vehicles, and is particularly suitable for sweeping path planning in unstructured road environments such as park streets, residential areas, and industrial zones.

[0063] This embodiment presents a method for planning missed sweeping paths for unmanned sweeping vehicles based on semantic topological factors, including the following steps:

[0064] Step 1: Construct the initial cleaning path. Based on the pre-built environmental planning map and vehicle operation requirements, construct an initial cleaning path that can cover the entire work area.

[0065] like Figures 2 to 3 As shown, the environmental planning map used in this embodiment is a high-precision map of the autonomous sweeper. It is a high-precision vector planning map created through automated production or manual deployment, based on a point cloud map constructed using multi-sensor fusion. Vehicle operation requirements can be understood as the cleaning tasks to be performed, such as sweeping and rinsing. By combining these two aspects, an initial sweeping path covering the entire work area is planned, providing a foundation for subsequent missed sweeping detection and re-sweeping planning. The initial sweeping path consists of a series of discrete points, which can be represented as... ,in This represents the coordinates of the i-th point on the route.

[0066] For route coverage planning, one can choose loop-shaped area coverage route planning or other conventional static full coverage route planning.

[0067] Step 2: Real-time semantic recognition of missed areas, real-time acquisition of vehicle pose information when the sweeper is traveling along the initial sweeping path, when the distance between the sweeper's position and the nearest point on the initial sweeping path is greater than the set offset threshold, it is determined that the vehicle has deviated, and the set of deviation points is obtained based on the vehicle's deviation trajectory.

[0068] Based on the set of deviation points and the initial cleaning path, convex hull calculation is performed to obtain a set of convex hull points, which is used to determine the boundary of the missed area. Then, target data within the missed area is collected and calculated to obtain a set of obstacle targets. Semantic features are statistically obtained according to their target type information. At least the target type, target quantity, and target area of ​​the semantic features are combined to set a set of semantic missed area blocks.

[0069] like Figure 2 , 4As shown, in this embodiment, when the sweeper travels along the set initial sweeping path, the positioning module monitors and obtains the current vehicle pose information in real time, which is reflected in a two-dimensional plane pose, including the X and Y axis coordinates of the sweeper and the heading angle of the sweeper. The heading angle of the sweeper can be understood as the angle between the direction of the vehicle's center of mass velocity and the horizontal axis of the ground coordinate system (usually referring to the due north direction), which is used to describe the absolute travel direction of the vehicle.

[0070] As the sweeper travels along the initial cleaning path, it may deviate from the path when encountering obstacles. Therefore, by monitoring the sweeper's position and orientation in real time, when the distance between the sweeper's position and the nearest point on the initial cleaning path exceeds a preset offset threshold, it is determined that the sweeper has deviated. At this point, the offset points are recorded based on the sweeper's offset trajectory and organized into a set of deviation points to represent the position and orientation information of the sweeper at each deviation point on the offset trajectory.

[0071] In practical applications, the deviation point is the control point for vehicle operation, defined at the center of the sweeper's rear axle. When the sweeper travels along the initial sweeping path, it can sweep both sides of the initial sweeping path. However, when the sweeper encounters an obstacle and needs to detour to the left or right to avoid it, it will cause the left or right sides of the original sweeping path to be missed.

[0072] like Figure 2 As shown in the example, this embodiment takes the sweeper going around the obstacle to the left as an example, which will cause a missed area on the right side of the initial sweeping path.

[0073] Therefore, in order to obtain more accurate boundary points of the missed areas, after completing the initial cleaning path and when starting the supplementary cleaning, the deviation point needs to be translated from the center of the sweeper's rear axle to the left or right boundary point of the sweeper; at the same time, the initial cleaning path is translated to the initial cleaning path boundary point a certain distance to the left or right, so that the boundary of the missed area is formed by combining the left or right boundary point of the sweeper with the initial cleaning path boundary point.

[0074] After converting the deviation point to the left or right boundary point of the sweeper, it is easier to define the missed area on the offset side using the left or right boundary point of the sweeper. This also avoids the problem of dividing the side that the sweeper has already cleaned into the missed area, which would lead to duplicate and overlapping divisions.

[0075] The initial cleaning path is translated and transformed into the boundary point of the initial cleaning path a certain distance to the left or right, so that the side that was missed by the original initial cleaning path can be divided into the missed area, thus avoiding the problem of missing or incomplete division of the missed area.

[0076] Then, based on the set of deviation points on the sweeper's offset trajectory and the boundary points of the initial sweeping path, convex hull calculation is performed to obtain the convex hull point set. The boundary of the missed area formed by the enclosed convex hull point set is also the missed area constituting the convex hull method. During convex hull calculation, the existing Graham Scan algorithm can be used to generate the sequential region point set envelope.

[0077] Based on the defined boundaries of the missed scan area, obstacle target data within the missed scan area is collected and calculated. Target data outside the area is collected using instruments such as LiDAR and cameras. Then, existing deep learning algorithms such as YOLO and Faster R-CNN are used to process the target data to obtain an obstacle target set. Each target in the obstacle target set has information attributes such as position, type, and velocity.

[0078] For the set of obstacle targets within the boundary of the missed scan area, semantic features of the region are set based on their target type information, such as the target type information of the obstacle target set, including dynamically movable, statically fixed, temporary obstacles, and garbage targets, as well as the corresponding number of targets. Then, a semantic missed scan area block set is set based on the regional semantic features of the obstacle target set, such as the target type, number of targets, and occupied area. In addition, information on garbage targets within the missed scan area is also included. By combining these semantic features, the missed scan area is set and calculated, thereby gaining a more comprehensive and richer understanding of the essence of the missed scan area and locating the missed scan area in complex environments more quickly and accurately.

[0079] Therefore, compared with the traditional method of judging missed scans by relying solely on location information, this embodiment performs semantic recognition on the internal obstacles within the defined missed scan area boundary and sets semantic missed scan area blocks, thereby gaining a more comprehensive understanding of the essence of the missed scan area, obtaining rich semantic information, and being able to quickly and accurately locate the missed scan area in a complex and ever-changing environment, providing support for subsequent supplementary scan path planning.

[0080] Step 3: Construct a semantic topology model of the missed areas. Encapsulate each semantic missed area block set in Step 2 into several nodes, and assume that there is connectivity between each node. Construct a semantic topology graph based on the connectivity of each node and the cleaning business attributes of the sweeper. Then add the starting point of the sweeper to the semantic topology graph and construct a node distance matrix including the starting point of the sweeper.

[0081] like Figures 5 to 7 As shown, when constructing the semantic topology graph, each semantic missed scan area block needs to be encapsulated into a node. Since the missed scan areas are selected based on the pre-constructed initial cleaning path, it can be assumed that each missed scan area is connected, that is, all missed scan areas are interconnected and reachable, and there is no need to make a connectivity judgment.

[0082] Then, based on the cleaning service requirements of the sweeper, the nodes are connected to form a topology network, such as... Figure 6 As shown. The sweeper's sweeping service attributes in this embodiment include at least garbage sweeping and water flushing sweeping services. Different sweeping services will affect the selection of effective cleaning areas.

[0083] Based on the semantic topology graph, the starting point of the sweeper is treated as a region and transformed into a node in the topology graph, which is then added to the semantic topology graph, connecting the starting point of the sweeper to other nodes.

[0084] like Figure 7 As shown, by connecting every pair of nodes, including the starting point of the sweeper, a node distance matrix is ​​constructed, with each pair of nodes forming a group.

[0085] Therefore, by constructing a semantic topology graph and node distance matrix of the missed scan area, the semantic relationship between the missed scan area and the surrounding environment and the topology graph are quantitatively expressed. For example, the quantification of obstacle information and garbage information in the environment, as well as the quantification of the connection relationship between nodes and edges in the topology graph are combined. Thus, by fully considering the semantic information in the environment of the missed scan area, a more comprehensive and accurate environmental model is provided for the re-scanning path planning, making the re-scanning path planning more reasonable and efficient.

[0086] Step 4: Full-coverage supplementary cleaning path planning. Calculate the distance between the connecting edges of every pair of nodes, including the starting point of the sweeper, in Step 3. This distance is used to plan the shortest supplementary cleaning path for missed areas after the initial cleaning path is completed, starting from the starting point of the sweeper and traversing all missed areas before returning to the initial position.

[0087] like Figures 6 to 7 As shown, based on the node distance matrix, the distance between the connecting edges of the node group including the starting point of the sweeper is calculated. For example, existing TSP traversal algorithms (such as genetic algorithm, simulated degradation algorithm, nearest neighbor algorithm, etc.) are used to calculate the shortest closed supplementary sweeping path to visit a group of nodes and to carry out traversal coverage sweeping planning for each missed area.

[0088] In this embodiment, the sweeper's supplementary sweeping operation occurs after the initial sweeping path has been executed, i.e., after the optimal supplementary sweeping path has been planned. The sweeper then travels along and performs supplementary sweeping according to this path. Specifically, starting from the node vstart associated with the sweeper's starting point, the sweeper sequentially traverses other nodes in the node distance matrix, eventually returning to the starting point, performing a traversal search. The system obtains the shortest region ranking after traversing all regions from the sweeper's starting region back to its starting region. For example, if the sweeper's starting region is AS, the search proceeds to the range As--A1--A2--A0--As.

[0089] Therefore, during the node traversal search process, the distance to each node, garbage, and business attribute variables are considered. Based on the semantic features and topological relationships of the missed scan areas, the direction and order of the new re-scanning path are automatically adjusted to find the optimal solution for path distance sorting. This allows for fast and accurate planning of re-scanning paths, prioritizing areas with more garbage in close proximity for re-scanning, avoiding unnecessary repeated trips, improving re-scanning efficiency, and ensuring that all missed scan areas are effectively covered, with a re-scanning coverage rate of over 98% for a single missed scan area.

[0090] The calculation of the sweeper's offset distance is further described in step 2, where the real-time pose of the sweeper in the two-dimensional plane coordinate system is assumed to be... , This indicates the location coordinates of the sweeper. This indicates the heading angle of the sweeper.

[0091] The formula for calculating the distance of a sweeper truck is as follows:

[0092] ;

[0093] Where, d i This indicates the distance between the sweeper's position and the nearest point on the initial sweeping path; This indicates the real-time location coordinates of the sweeper. This represents the coordinates of the i-th point on the initial cleaning path.

[0094] In practical applications, an offset threshold is defined. Offset threshold The value range is 0.25 to 0.35m, because there is an overlapping area of ​​about 0.3m in the full coverage route of the vehicle. When it exceeds this value, it means that there are missed areas in the overlapping area.

[0095] Therefore, when the distance between the sweeper's position and the nearest point on the initial sweeping path If the deviation is greater than 0.3m, the vehicle is considered to have deviated.

[0096] Further describing the deviation point set collaboration, based on the vehicle offset trajectory formed by the real-time position coordinates of the sweeper, the deviation point set is defined as... , ,in, Let represent the pose information of the j-th deviation point, where j = 0, 1, ..., n. By recording the set of deviation points, the actual offset of the sweeper is clarified, which is used to perform convex hull calculation with the initial sweeping path to obtain the set of convex hull points and clarify the boundaries of the missed sweeping areas.

[0097] Further describing the convex hull point set, in step 2, the convex hull point set obtained by calculating the convex hull based on the set of deviation points and the initial cleaning path is denoted as CH(P).n ),

[0098] ,in, Let represent the pose information of the i-th convex hull point, i = 0, 1, 2, ..., h-1. Thus, the final convex hull missed scan region is formed by enclosing multiple convex hull points.

[0099] To further describe the obstacle target set, in step 2, sensor instruments such as lidar are used to collect target data outside the area, and deep learning algorithms are used to process the target data to obtain the obstacle target set.

[0100] The obstacle target set in this embodiment is represented as follows: , ,in, This represents the first objective, and each objective... It has the following attributes:

[0101] Target location information: This includes the three-dimensional coordinates of the obstacle target (x, y, z), which can be obtained through existing sensing modules.

[0102] Target type information: , ;

[0103] These respectively represent dynamically moving targets, static fixed targets, temporary obstacle targets, and targets with uncleaned debris;

[0104] Dynamic target speed information: This includes the target's speed in the x and y directions.

[0105] Based on the boundary of the missed scan region and the target type information of the target set, the semantic missed scan region block set is represented as , ,in, This represents the i-th semantically missed scan region block.

[0106] Each semantically missed scan region block contains a clear region boundary and region semantic features, wherein the region boundary is clearly defined by the aforementioned set of convex hull points.

[0107] ,in,

[0108] , represents the semantic feature information of dynamically movable targets in the missed scan area, including at least the target type, the area occupied, and the number;

[0109] , represents the semantic feature information of static fixed targets in the missed scan area, which includes at least the target type, occupied area, and quantity;

[0110] This represents the semantic feature information of temporary obstacle targets within the missed scan area, which includes at least the target type, the area it occupies, and the number of targets.

[0111] This indicates the garbage target information in the missed area.

[0112] Further describing the cleaning operations, step 3 includes cleaning attributes for both garbage collection and flushing cleaning. Based on the defined set of semantically missed area blocks, the selection requirements for missed areas of obstacle targets are as follows, combined with different operation attributes:

[0113] When carrying out garbage cleaning operations, if the area of ​​the missed area is occupied by dynamic moving targets that occupy more than 80% of the total area of ​​all non-garbage obstacle targets, the number of garbage targets is more than 3, and the area of ​​the missed area is more than 10 square meters, then it will be included in the effective cleaning area.

[0114] When performing flush cleaning and sweeping operations, areas with a missed sweeping area exceeding 10 square meters (excluding the target area with temporary obstacles) are included as valid areas for additional sweeping.

[0115] In step 3, when constructing the topology graph based on the cleaning service attributes, the semantic topology graph expression for the corresponding waste cleaning service is: The semantic topology graph expression corresponding to the flushing and cleaning business is as follows: Where V is the node of the effective cleaning area under the corresponding cleaning business attribute, and Egarbage and Ewash are the sets of edges under the corresponding cleaning business attributes.

[0116] Second Embodiment

[0117] like Figure 10 As shown, a path planning system for unmanned sweeping vehicles that misses areas based on semantic topological factors is also provided, including:

[0118] The map generation module is used to construct an initial cleaning path that can cover the entire work area based on a pre-built environmental planning map and vehicle operation requirements;

[0119] The positioning module is used to acquire the vehicle's position and pose information in real time as the sweeper travels along the initial sweeping path, and send the acquired vehicle coordinates to the terminal; when the distance between the sweeper's position and the nearest point on the initial sweeping path is greater than a preset offset threshold, it is determined that the vehicle has deviated, and the set of deviation points is obtained based on the vehicle's deviation trajectory.

[0120] The target calculation module is used to perform convex hull calculation based on the set of deviation points and the initial cleaning path to obtain the set of convex hull points, which is used to determine the boundary of the missed area; and to collect and calculate the target data in the missed area to obtain the set of obstacle targets, and to obtain the semantic features of the area according to its target type information, and to set the semantic missed area block set by combining at least the target type, target quantity and target area of ​​the semantic features of the area.

[0121] The topology model generation module is used to encapsulate each set semantic missed area block into several nodes, and assumes that there is connectivity between the nodes. Based on the connectivity of each node and the cleaning business attributes of the sweeper, a semantic topology graph is constructed. Then, the starting point of the sweeper is added to the semantic topology graph to construct a node distance matrix including the starting point of the sweeper.

[0122] The matrix calculation module is used to calculate the distance between the edges connecting any two nodes, including the starting point of the sweeper, and to sort the shortest area cleaning paths after traversing all areas from the starting point of the sweeper back to the initial position.

[0123] The functional modules used in the missed scan path planning system of this embodiment can be matched one-to-one with the functional modules actually used in the first embodiment. The actual module names or models will not be described again.

[0124] Therefore, through the mutual support and assistance among the various system modules of this embodiment—map generation module, positioning module, target calculation module, topology model generation module, and matrix calculation module—it is possible to quickly and accurately identify missed areas and generate supplementary sweeping paths in vehicle missed sweeping path planning.

[0125] Specifically, by performing semantic recognition and setting semantic missing scan region blocks for obstacles within the defined missing scan region boundary, compared with the traditional method of judging missing scans based solely on location information, this invention can more comprehensively understand the essence of the missing scan region, obtain rich semantic information, and quickly and accurately locate the missing scan region in complex and ever-changing environments.

[0126] Based on the determined missed scan area, the semantic topology graph and node distance matrix of the missed scan area are constructed to quantify the semantic relationship between the missed scan area and the surrounding environment and the topology graph. For example, the quantification of obstacle information and garbage information in the environment, as well as the quantification of the connection relationship between nodes and edges in the topology graph, are combined to fully consider the semantic information in the environment of the missed scan area.

[0127] Based on the semantic information of the missed scan areas, the distance to each node, garbage, and business attribute variables are considered during the node traversal search process. Thus, according to the semantic features and topological relationships of the missed scan areas, the direction and order of the new re-scanning path are automatically adjusted, and the optimal solution for path distance sorting is obtained. This allows for fast and accurate planning of re-scanning paths, prioritizing areas with more garbage in close proximity for re-scanning, avoiding unnecessary repeated trips, improving re-scanning efficiency, and ensuring that all missed scan areas are effectively covered, with a re-scanning coverage rate of over 98% for a single missed scan area.

[0128] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A method for planning a missed path of an unmanned sweeper based on a semantic topological factor, characterized in that, The method comprises the following steps: Step 1, constructing an initial cleaning path, constructing an initial cleaning path capable of covering the entire working area according to a pre-constructed environment planning map and vehicle working requirements; Step 2, real-time semantic recognition of a missed cleaning area, real-time acquisition of vehicle pose information of the cleaning vehicle when driving along the initial cleaning path, when the distance between the cleaning vehicle position and the nearest point on the initial cleaning path is greater than a preset offset threshold, it is judged that the vehicle has deviated, and a deviation point set is acquired according to the vehicle deviation trajectory; According to the deviation point set and the initial cleaning path, a convex hull calculation is performed to obtain a convex hull point set for determining the boundary of the missed cleaning area; then target data in the missed cleaning area is collected and calculated to obtain an obstacle target set, and regional semantic features are obtained according to the target type information, and a semantic missed cleaning area block set is set at least according to the target type, target quantity and target area of the regional semantic features; Step 3, constructing a semantic topology model of the missed cleaning area, encapsulating each semantic missed cleaning area block set in step 2 into a node, and assuming that there is connectivity between each node, constructing a semantic topology graph according to the connectivity of each node and the cleaning business attribute of the cleaning vehicle; Then the starting point of the cleaning vehicle is added to the semantic topology graph to construct a node distance matrix including the starting point of the cleaning vehicle; Step 4, full coverage re-cleaning path planning, calculating the distance between the connection edges of each two nodes including the starting point of the cleaning vehicle in step 3, for planning the shortest missed cleaning area re-cleaning path from the starting point of the cleaning vehicle to traverse all missed cleaning areas and return to the initial position after completing the initial cleaning path cleaning.

2. The semantic topology factor based unmanned sweeper path planning method of claim 1, wherein, In step 2, when the initial cleaning path cleaning is completed and the re-cleaning is started, the deviation points are converted from the cleaning vehicle rear axle center to the cleaning vehicle left boundary point or the cleaning vehicle right boundary point; At the same time, the initial cleaning path is converted to the initial cleaning path boundary point with a certain distance to the left or right, which is used to enclose and calculate the boundary of the missed cleaning area with the cleaning vehicle left boundary point or right boundary point.

3. The semantic topology factor based unmanned sweeper path planning method of claim 1, wherein, In Step 2, let the real-time pose of the sweeper in the two-dimensional plane coordinate system be , represents the position coordinates of the sweeper, represents the heading angle of the sweeper; The distance calculation formula of the cleaning vehicle is as follows: ； Wherein, d i represents the distance between the sweeper position and the nearest point on the initial cleaning path; x, y represents the real-time position coordinates of the sweeper; x i , y i represents the position coordinates of the i-th point on the initial cleaning path; The vehicle offset trajectory formed based on the real-time position coordinates of the sweeper is set as P n , wherein, , represents the pose information of the jth offset point position, j = 0, 1, …, n.

4. The method of claim 1, wherein the semantic topology factor is determined based on a semantic map of the environment. In step 2, let the set of convex hull points be , where, , denotes the pose information of the i-th convex hull point, i = 0, 1, 2, …, h-1.

5. The method of claim 1, wherein the semantic topology factor is a semantic factor of a semantic map. In step 2, the target data in the deviation area is collected by using a sensor, and the target data is processed by using a deep learning algorithm to obtain the obstacle target set; The set of obstacle goals is represented as , wherein, represents the 1st goal, each goal having the following attributes: Target position information: ; Target type information: ,​ Indicating a dynamic moving target, a static fixed target, a temporary obstacle target and an unswept garbage target, respectively; Dynamic moving target speed information: ; based on the leak scan area boundary and target type information of the target set, the semantic leak scan area block set is represented as , , wherein, the i-th semantic leak scan area block is represented as 6. The semantic topology factor based unmanned sweeper path planning method of claim 5, wherein, Each semantic missed cleaning area block includes a region boundary and a region semantic feature, wherein the region boundary is clearly divided by the convex hull point set; The regional semantic features are obtained according to the target type information, and the regional semantic features can be represented as wherein, , represents the semantic feature information of the dynamically movable target in the missed scanning area, and at least contains target type, occupied area, and quantity; , represents the static fixed target semantic feature information in the leakage scanning area, at least contains target type, occupied area, quantity; , represents the temporary obstacle target semantic feature information in the leakage scanning area, and at least contains a target type, an occupied area, and a number; , represents the garbage target information in the missed scanning area.

7. The semantic topology factor based unmanned sweeper path planning method of claim 6, wherein, In step 3, the cleaning business attribute includes garbage cleaning business and flushing cleaning cleaning business; When performing garbage cleaning business, if the dynamic moving target in the missed cleaning area occupies more than 80% of the non-garbage obstacle target area, the number of garbage targets is more than 3, and the area of the missed cleaning area is more than 10 square meters, it is included in the effective re-cleaning area; When performing flushing cleaning cleaning business, the area of the missed cleaning area excluding the temporary obstacle target area is more than 10 square meters, which is included in the effective re-cleaning area.

8. The semantic topology factor based unmanned sweeper path planning method of claim 7, wherein, The semantic topology graph expression corresponding to the garbage cleaning service in step 3 is The semantic topology graph expression corresponding to the flush cleaning service is wherein V is a valid supplementary cleaning area node corresponding to the cleaning service attribute, and Egarbage and Ewash are edge sets corresponding to the cleaning service attribute.

9. The method of claim 1, wherein the semantic topology factor is based on a semantic map of the environment. In step 3, based on the semantic topology graph, edges are connected between each two nodes including the starting point of the cleaning vehicle to construct the node distance matrix of each two nodes as a group.

10. An unmanned sweeper path leakage planning system based on semantic topology factors, characterized in that, It comprises: The map generation module is configured to construct an initial cleaning path capable of covering the entire cleaning area according to a pre-constructed environment planning map and vehicle operation requirements; The positioning module is configured to acquire vehicle pose information of the cleaning vehicle when driving along the initial cleaning path in real time, and send the acquired vehicle coordinates to the terminal; when the distance between the cleaning vehicle and the nearest point on the initial cleaning path is greater than a preset offset threshold, it is determined that the vehicle has deviated, and a set of deviated points is acquired according to the vehicle deviation trajectory; The target calculation module is configured to perform convex hull calculation on the set of deviated points and the initial cleaning path to acquire a set of convex hull points, so as to determine the boundary of the missed cleaning area; then, target data in the missed cleaning area is collected and calculated to obtain a set of obstacle targets, and semantic features are obtained by counting the target type information; at least in combination with the target type, target quantity and target area of the semantic features, a set of semantic missed cleaning area blocks is set; The topology model generation module is configured to encapsulate each set of semantic missed cleaning area blocks into a plurality of nodes, and by default, the nodes are connected; a semantic topology graph is constructed according to the connectivity of the nodes and the cleaning business attributes of the cleaning vehicle; Then, the starting point of the cleaning vehicle is added to the semantic topology graph to construct a node distance matrix including the starting point of the cleaning vehicle; The matrix calculation module is configured to calculate the distance between the connection edges of two nodes including the starting point of the cleaning vehicle, and sort the shortest area re-cleaning path from the starting point of the cleaning vehicle to all areas and back to the initial position.

Images