Incremental topological map construction method based on pruned Voronoi diagram

Abstract

The present invention relates to an incremental topological map construction method based on pruned Voronoi diagrams. In urban semi-structured environments, off-road and rescue unstructured environments, where obstacle shapes are difficult to fit, topological relationships are complex, and the mapping range is large, a lightweight topological map is incrementally and rapidly autonomously constructed as a ground or low-altitude manned / unmanned mobile platform travels, so as to facilitate robot navigation. Beneficial effects: In large-scale complex semi-structured and unstructured environments, the present invention can incrementally and rapidly generate lightweight topological maps for robot navigation based on the boundary points of the drivable area, avoiding the influence of obstacle fitting errors and complex calculations, and realizing rapid autonomous topological map construction in complex environments with numerous topological relationships, irregular obstacles, and a large mapping range.

Description

Technical Field

[0001] The present invention belongs to the technical field of topological maps, and in particular relates to an incremental topological map construction method based on pruned Voronoi diagrams. Background Art

[0002] As an efficient representation method of the environment, topological maps are an important form of achievement for robot autonomous exploration and SLAM (Simultaneous Localization And Mapping) technology. Topological maps represent the environment in the form of nodes and edges. Compared with grid metric maps and semantic maps, topological maps have a more concise and compact road structure representation method and are easier for humans to understand. They are widely used in robot search and rescue, underground surveys, and other autonomous navigation technologies in unknown environments. In the above environments and tasks, it is difficult for the perception results to quickly and accurately fit the shape of irregular obstacles, which makes it difficult for some topological map generation methods. At present, the methods for automatically constructing topological maps are mainly divided into three categories: methods based on visibility graphs, methods based on random sampling, and methods based on skeletonization.

[0003] The visibility graph-based method fits obstacles into polygons, connects the vertices of the polygons to form edges, and leaves the visible edges as topological edges. When facing irregular obstacles and environments with a large number of obstacles, it is easily affected by fitting errors, and the computational complexity will gradually increase; the random sampling-based method uses the connection of passable sampling points as topological edges by randomly sampling in the environment, such as the Probabilistic Roadmap (PRM) and the Rapidly-exploring Random Tree (RRT). This type of method has probabilistic completeness, but in a narrow environment, it takes a long time to sample enough points to ensure the completeness of the topological relationship. The skeletonization-based method is mainly based on the generalized Voronoi diagram (GVD). The GVD takes the obstacle target in the environment as the minimum representation unit, defines the distance representation method from any point in space to the target, and then constructs it according to the definition of the Voronoi diagram. The traditional GVD construction method is also affected by obstacle shape fitting. Another Voronoi diagram construction method based on Euclidean distance map has the problem of topological relationship redundancy.

[0004] In view of the high cost of manual mapping in large-scale complex semi-structured and unstructured environments, the complex topological relationship structure of automatic mapping, and the difficulty of efficient incremental construction, it is urgent to solve the problem of automatic construction of topological maps based on the driving process of ground or low-altitude manned / unmanned mobile platforms. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned technology and to provide an incremental topological map construction method based on pruned Voronoi diagram to solve the problem of automatic construction of topological maps during the driving process of ground or low-altitude manned / unmanned mobile platforms in large-scale complex semi-structured and unstructured environments.

[0006] To achieve the above purpose, the present invention adopts the following technical scheme: an incremental topological map construction method based on pruned Voronoi diagram, mapping the boundary points of the drivable area to the global grid map; using the maximum empty circle property of Voronoi vertices, selecting the minimum generated area in the global grid map to realize incremental topological map construction; pruning the topological map by boundary point expansion and invalid edge screening to retain the skeleton information of the environment; finally, constructing a fully connected topological map based on a breadth-first search algorithm, the specific steps are as follows:

[0007] (1) Mapping the boundary points of the drivable area to the global grid map;

[0008] (2) Establish a rectangular envelope of the robot's circular perception range, expand the rectangular envelope outward until the vertices on the Voronoi edge that intersects the rectangular envelope meet the following properties: the number of base points in the largest empty circle of the vertex is not less than 3, and none of them is a newly added drivable area boundary point, and generate a traditional Voronoi diagram with all drivable boundary points in the rectangular envelope as base points;

[0009] (3) Expand the drivable boundary points in the generated area and delete the Voronoi edges in the occupied space;

[0010] (4) Construct a local undirected graph based on Voronoi edges and vertices, delete invalid edges with short length and only one connection, retain the skeleton information of the environment, and use the local undirected graph to replace the global undirected graph in the original rectangular frame;

[0011] (5) Starting from the robot's position, the undirected graph is traversed aimlessly based on the breadth-first search algorithm until the open list is empty, and the edges that have not been included in the open list and the closed list are deleted.

[0012] Furthermore, the drivable area boundary point in step (1) is a static obstacle point and an effective perception boundary point in the environment, and when the drivable area boundary point is stored in the global grid map, only one point is retained in the same grid according to the grid map accuracy.

[0013] Furthermore, the maximum empty circle property of the vertex in step (2) refers to that the circumference of the maximum empty circle of the Voronoi vertex contains at least 3 base points, and the base point set on the circumference of the maximum empty circle is obtained by finding the base point closest to the vertex.

[0014] Furthermore, the principle of deleting the invalid edges in step (4) is that the length of the invalid edge is less than the sum of the robot length and the length threshold, and the edge has only one connected relationship in the local undirected graph.

[0015] Furthermore, the local undirected graph established in step (4) is a connected graph after being processed based on the breadth-first search algorithm, and there is a path between any two vertices.

[0016] Beneficial effects: Compared with the prior art, in large-scale complex semi-structured and unstructured environments, the present invention can incrementally and quickly generate lightweight topological maps for robot navigation based on the boundary points of the drivable area, avoiding the influence of obstacle fitting errors and complex calculations, and realizing rapid autonomous topological map construction in complex environments with numerous topological relationships, irregular obstacles, and a large mapping range; the present invention is sensitive to environmental information when constructing a topological map, and the addition and deletion of any boundary / obstacle point can be quickly reflected in the topological map; the present invention has high timeliness, thanks to the high efficiency of the traditional Voronoi diagram generation method, as well as the high efficiency of the pruning algorithm and the incremental method in the present invention, the topological map generation process has high timeliness; the present invention has consistency, and the map result constructed in a fixed area is not affected by the order of environmental exploration; the map structure constructed by the present invention is complete, and the construction result is lighter than other methods, and the topological map construction process is not affected by the narrow environment and the complexity of the topological relationship, but is only related to the number of boundary / obstacle points, and can be applied to any environment to achieve reliable, accurate, and complete mapping. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1A It is a schematic diagram of the information of the historical environmental perception obstacle points;

[0018] Figure 1B It is a schematic diagram of the result after superposition of environmental perception obstacle points;

[0019] Figure 2A This is a schematic diagram of identifying the largest empty circle in the Voronoi diagram generation area during incremental composition.

[0020] Figure 2B It is a schematic diagram of the identification of the region generated by the Voronoi diagram during the incremental composition;

[0021] Figure 3Ais the original Voronoi diagram;

[0022] Figure 3B It is a schematic diagram of the obstacle expansion result;

[0023] Figure 3C is a schematic diagram of pruning results;

[0024] Figure 4A is the historical global Voronoi diagram;

[0025] Figure 4B It is the result diagram of region identification and local Voronoi diagram generation;

[0026] Figure 4C It is an incremental update result graph;

[0027] Figure 5 It is a topological map based on breadth-first search and geometric simplification, where: the left picture is the initial pruning result after expansion pruning, and the right picture is the final pruning result after breadth-first search algorithm and geometric simplification;

[0028] Figure 6 It is a topological map constructed in the indoor simulation experiment verification of the present invention;

[0029] Figure 7 The time consumption of constructing the topological map in the indoor simulation experiment verification of the present invention;

[0030] Figure 8 It is a topological map constructed in the forest simulation experiment verification of the present invention;

[0031] Fig. 9 It is a topological map constructed in the actual cell experimental verification of the present invention; (Abstract Figure)

[0032] Fig.10 It is the time-consuming macro-planning in experimental verification of the present invention. DETAILED DESCRIPTION

[0033] In order to be able to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other. In the following description, many specific details are set forth in order to fully understand the present invention, and the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the art without making creative work are within the scope of protection of the present invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0034] In the various embodiments of the present invention, for the convenience of description and not to limit the present invention, the term "connection" used in the patent application specification and claims of the present invention is not limited to physical or mechanical connection, but may include electrical connection, whether direct or indirect. "Up", "down", "below", "left", "right", etc. are only used to indicate relative position relationship. When the absolute position of the described object changes, the relative position relationship also changes accordingly.

[0035] See Figures 1-5 for details. This embodiment provides an incremental topological map construction method based on pruned Voronoi diagrams. In urban semi-structured environments, off-road and rescue unstructured environments where obstacle shape fitting is difficult, topological relationships are complex, and the mapping range is large, as ground or low-altitude manned / unmanned mobile platforms travel, lightweight topological maps are incrementally and rapidly constructed autonomously to facilitate robot navigation and enhance human understanding of the environment. First, the boundary points are placed in the global grid map, and the maximum empty circle property of the Voronoi vertices is used to construct the minimum generation range of the Voronoi diagram to achieve incremental construction and update of the topological map; then, the Voronoi diagram is simplified by designing a pruning method to extract the skeleton information of the environment; and the breadth-first search algorithm is used to delete the edges without connection relationships to achieve full connectivity of the topological map and build a concise topological structure. The specific steps are as follows:

[0036] (1) Mapping the boundary points of the drivable area to the global grid map;

[0037] (2) Establish a rectangular envelope of the robot's circular perception range, expand the rectangular envelope outward until the vertices on the Voronoi edge that intersects the rectangular envelope meet the following properties: the number of base points in the largest empty circle of the vertex is not less than 3, and none of them is a newly added drivable area boundary point, and generate a traditional Voronoi diagram with all drivable boundary points in the rectangular envelope as base points;

[0038] (3) Expand the drivable boundary points in the generated area and delete the Voronoi edges in the occupied space;

[0039] (4) Construct a local undirected graph based on Voronoi edges and vertices, delete invalid edges with short length and only one connection, retain the skeleton information of the environment, and use the local undirected graph to replace the global undirected graph in the original rectangular frame;

[0040] (5) Starting from the robot's position, the undirected graph is traversed aimlessly based on the breadth-first search algorithm until the open list is empty, and the edges that have not been included in the open list and the closed list are deleted.

[0041] The preferred solution of this embodiment is that the boundary points of the drivable area are static obstacle points and effective perception boundary points in the environment, and when the boundary points of the drivable area are stored in the global grid map, only one point is retained in the same grid according to the grid map accuracy. In this embodiment, the boundary points of the drivable area are static obstacle points.

[0042] The preferred solution of this embodiment is that the maximum empty circle property of the vertex refers to at least 3 base points on the circumference of the maximum empty circle of the Voronoi vertex. The base point set on the circumference of the maximum empty circle is obtained by searching for the base point closest to the vertex.

[0043] The preferred solution of this embodiment is that the principle of deleting invalid edges is: the length of the invalid edge is less than the sum of the robot length and the length threshold, and the edge has only one connected relationship in the undirected graph. This embodiment performs two invalid edge deletions. After the invalid edges are deleted for the first time, the short edges with the same orientation are fitted as long edges according to the orientation angle, and the second invalid edge removal is performed according to this rule to achieve the purpose of simplifying invalid short edges.

[0044] A preferred solution of this embodiment is that the established undirected graph is a connected graph after being processed based on breadth-first search, and there is a path between any two vertices.

[0045] This example collects the positioning and perception results of the unmanned platform in an environment where a topological map needs to be built. It is often used together with the online SLAM module. The platform can be driven by autonomous exploration or manual driving. The perception results of the unmanned platform are preferably based on the results of lidar, millimeter-wave radar, and depth camera. The accurate location of obstacle points needs to be included, and the target types and grid points that do not need to appear in the map need to be identified. Topological map construction can be incrementally built as the platform runs and used for unmanned platform navigation in real time, or a map of the entire area can be built at once after the platform runs.

[0046] Example

[0047] Incremental topology map construction experiment using complex simulation environment and real community environment as examples

[0048] 1. Start the data receiving program of the 3D laser radar (including 1 32-line front laser radar and 1 16-line rear laser radar) on the platform, run the SLAM program and perception program, and the perception results include the classification information and speed information of pedestrians, cyclists and cars. The rest of the unclassified information is output in the form of obstacle points;

[0049] 2. Complete full coverage of the pre-established map area through manual or autonomous exploration;

[0050] 3. During the driving process of the platform, a topological map is constructed in real time with a construction frequency of 1 Hz. After receiving the perception results, pedestrians, cyclists and moving cars are removed and a topological map is constructed according to the steps of the present invention;

[0051] This embodiment aims at constructing a topological map in an environment represented by a metric grid map, and adopts a pruning method based on the Voronoi diagram; this embodiment aims at incremental topological map construction in a large environment. The existing incremental methods for constructing Voronoi diagrams include random incremental methods and methods based on dynamic Brushfire. This embodiment proposes an incremental method based on the maximum empty circle property. The specific method adopted is as follows:

[0052] (1) Perception result update

[0053] See Figure 1 for details, which is a schematic diagram of the update of environmental perception obstacle points. The input of the most recent frame of perception results is superimposed on the historical map. There are a large number of obstacle points in the large environment, and the obstacle points are stored in blocks through file storage. According to the grid index value, the input obstacle points are identified to determine whether they need to be added. Obstacle points in the same grid have the same index value. Figure 1A Contains the original obstacle points and the perception results of the most recent frame. The grid points that need to be added are determined according to the grid index value. Although the positions of the new obstacle points are slightly different from the positions of the historical obstacle points, they are still in the same grid, and the specific positions of the historical grid points are not changed. Figure 1B The superposition result is shown in the figure. After the superposition is completed, all points within the sensing range are taken as the base points of the Voronoi diagram. In addition, according to the annotation of dynamic obstacles in the sensing result, dynamic obstacles are not added to the base point set.

[0054] (2) Voronoi diagram generation region identification

[0055] The Voronoi Diagram (VD) is a partition of the nearest points in space. The traditional Voronoi diagram is a partition of the point set in space, also known as Thiessen polygons. Its definition is as follows: Let the point set P = {p1, p2, ..., p N} is a set of N points (base points) in space, d(p,q) is defined as the Euclidean distance between two points p and q in space, C(p i ) represents the Voronoi diagram corresponding to point p i The Voronoi cells are also called Voronoi domains. In each Voronoi cell, the properties in formula (1) exist. All Voronoi cells form the Voronoi diagram Vor(P). The elements in the set P are called base points. By definition, a Voronoi cell contains only one point p. i , and any point x to p in the Voronoi cell i The distance from x to any point p outside the unit is less than k The distance between the edge E of the Voronoi cell V The distance from any point on the grid to two points in adjacent cells is the same, so E V is the perpendicular bisector between any two adjacent points, so the triangle formed by connecting adjacent points that share an edge is a Delaunay triangle. Therefore, the Voronoi diagram is the dual graph of the Delaunay triangulation.

[0056] C(p i )={x|d(x,p i )<d(x,p k ),p i ,p k ∈P,i≠k,i,k∈{1,2,…,n}} (1)

[0057] The essence of the change of the Voronoi diagram is the change of the vertex, and the change of the position of the vertex is affected by the base point. Any Voronoi vertex q has the following properties: With q as the center, the circle constructed without any obstacle inside is called C(q), and the C with the largest radius max (q) is called the maximum empty circle. When C(q) is the maximum empty circle, there are at least three obstacle points on its circumference, such as Figure 2A As shown. After loading the obstacle points in the vicinity according to the current position and updating the perception results of the most recent frame, according to C max (q) It can be seen that the distances of the three points closest to point q found by the nearest neighbor should be equal. Figure 2B In the schematic diagram shown, if the newly added point is within the largest empty circle of some vertices, the vertex will change in the new Voronoi diagram and a new vertex may be generated;

[0058] See Figure 2 for details, which is a schematic diagram of the Voronoi diagram generation area recognition during incremental composition. The scanning line method is based on the geometric properties of the parabola (a parabola is composed of points with equal distances to a fixed point and a fixed straight line). By continuously moving a fixed straight line, the intersection of multiple parabolas is used to generate Voronoi edges. Inspired by this idea, the present invention designs a minimum area recognition method based on the scanning line method: Identify the newly added obstacle point set P N The minimum area R that will change in the Voronoi diagram N .like Figure 2B As shown, the set of all obstacle points P in the sensing range is obtained in the map = P N ∪P H , where P H is the historical obstacle point set. Get the rectangular envelope R of the midpoint P in the plane min , by the rectangular envelope R min The four sides of are translated separately, when the condition of formula (2) is satisfied or when E L When the set is empty, the translation ends, where x1 and x2 are the vertex coordinates of the rectangle, L(x1,x2) is the line segment formed by the two vertices, and E L is the set of historical Voronoi edges that intersect with L(x1,x2), V N For edge E L In the rectangular area R N The vertex inside, P C V N The largest empty circle C with vertex k max (k) The set of obstacle points on the circumference or inside the empty circle, consisting of K base points S k When any obstacle point in it belongs to a historical obstacle point, the vertex V Gk There will be no change.

[0059]

[0060] In the specific implementation, the kd tree is used to query the nearest neighbor to determine the maximum empty circle of the vertex: the base point set P is stored in the kd tree, and the nearest obstacle point S is found for the vertex k. k and its distance d k , and then search the kd tree for the node with a distance d from vertex k. k The set of all obstacle points P C , P C Then it is the set of obstacle points on the maximum empty circle. When the set P C The number of midpoints is greater than or equal to 3 and P C Each arbitrary point in the query is queried by index value and it belongs to the historical obstacle point P H When , the position of vertex k will not be affected by the newly added obstacle points;

[0061] (3) Voronoi diagram pruning method

[0062] The Voronoi diagram established with the obstacle point set P as the base point and the Euclidean distance as the distance representation method is represented as Vor(P). O ={E O1 ,E O2 ,…,E OM} and V O = {V O1 ,V O2 ,…,V ON} represents the set of edges and vertices in the graph, where M and N are the total number of edges and vertices, and E Ok =(V Oi ,V Oj ), i, j∈{1,2,…,M}. See Figure 3 for a schematic diagram of a pruning method based on barrier point expansion. The present invention prunes Vor(P) based on the expansion configuration space C, where Figure 3A is a complete graph generated based on the obstacle points, Figure 3B is the expanded configuration space, Figure 3C is the Vor(P) after pruning. According to the set threshold, the obstacle points in the space are expanded to obtain the free space C free and obstacle space C occupied , where C = C free ∪C occupied , then perform collision detection on each edge in Vor(P) in the configuration space C, and prune the invalid edges according to formula (3) to obtain the valid edge set E O ', with L(V Oi ,V Oj ) represents the line segment between two vertices, and the collision safety condition is

[0063]

[0064] (4) Undirected Graph Construction

[0065] The present invention uses an undirected graph G=(V G ,E G ) constructs a topological map, there is First, identify the connectivity of each edge in Vor(P), starting from the vertex V Gi 、V Gj and vertex V Gp 、V Gq The condition for the two sides to be connected is as follows: Gi or V Gj .

[0066]

[0067] According to step (3), there are many short edges with only one connected relationship in the pruning result Vor(P), which are invalid edges in the topological map. The present invention sets the rule shown in formula (5) to be no greater than the platform length d v , and the set E of all edges G The edges with only one connection relationship are considered invalid and deleted, that is, the leaf nodes with shorter length are deleted. p,q Vertex V Gp and V Gq The side formed, d buffer is the length threshold.

[0068]

[0069] (5) Global map update

[0070] R N The goal is to convert the global Voronoi diagram into R N The boundary is divided into two parts, the rectangular area R N The Voronoi edges outside the R do not change after the perception data is updated. To achieve this goal, it is necessary to N The base points in the largest empty circle corresponding to all vertices in the rectangle are contained in the rectangular area, so continue scanning and expanding to obtain area R max The present invention also sets the distance threshold d B , Figure 4 is a schematic diagram of the global Voronoi diagram update. max All obstacle points in the region are used as base points to generate the Voronoi diagram. N Delete all the edges outside R, and keep only all the vertices in R N The Voronoi edge set E in N . Delete all vertices in the global Voronoi diagram in R N The Voronoi edge set E in O , E N Add to the global Voronoi diagram and create it as follows Figure 4C The results are shown. After establishing the topological relationship, an incrementally constructed global topological map is obtained;

[0071] (6) Identification and removal of isolated edges

[0072] The local undirected graph G is divided into a limited number of block graphs according to the connection relationship. The block where the robot is located is selected as the main part, and the edges that cannot be connected to the main part are removed. The edges that have no connection relationship with the main part are called isolated edges in the present invention. The present invention uses a fast search algorithm to complete the identification and removal of isolated edges. The specific process of the algorithm is as follows: First, select the nearest edge E near the robot r , add the edge to the priority list Q P and open list Q O Secondly, when traversing the graph, there is no need to consider the length of each edge. The goal is to quickly traverse the entire graph. There is no need to find the shortest and optimal path. The breadth-first search algorithm is selected for target search until the priority list Q P When there is no data in the open list, the search is completed; finally, the edges that are not in the open list are removed, and multiple edges in the open list are searched according to the orientation angle θ≤θ max Merge into one edge, Figure 5 It is a topological map based on breadth-first search and geometric simplification. Figure 5 The left picture shows the initial pruning result after expansion pruning. Figure 5 The right picture shows the final pruning result after breadth-first search and geometric simplification.

[0073] After deleting the unreachable edges, the simplified Vor(P) is obtained, which is a connected graph with topological relationship, that is, there must be a connected path between any edges, and there are no parallel edges in the graph. There are two advantages of using the above method to approximate GVD: first, there is no need to cluster obstacles and fit polygons, which reduces fitting errors; second, it reduces the consumption of computing resources caused by obstacle fitting and reduces the complexity of map construction.

[0074] 4. The topological map constructed at any time during the driving process can be used for the macro planning of the current platform, and the shortest path is output based on the current existing map. The topological map constructed after the area exploration is completed can not only be used for macro guidance when other platforms re-enter the area, but also can be used to deepen the human operator's understanding of the environment.

[0075] 5. Experimental Verification and Data Analysis

[0076] We selected an indoor simulation environment (130m×100m), an outdoor forest simulation environment (150m×150m), and a real community environment (100m×100m) with complex topological relationships for topological map construction experiments. We used autonomous exploration in the simulation environment and manual driving in the real community to complete full coverage of the preset area.

[0077] The method of the present invention is compared with other methods in a simulation environment, including a method based on random sampling (referred to as predecessor method 1) and a method based on visibility graphs (referred to as predecessor method 2), and statistics are performed on the timeliness of map construction and map results. Figure 6 This is the topological map after the indoor environment exploration is completed, and two typical intersection areas are enlarged. The result shows that the topological map constructed by this method can obtain the skeleton information of the environment and is easier to understand the environment visually. Compared with the previous methods 1 and 2, there are no excessive redundant edges and nodes, and there are no unrepresented environmental areas. Figure 7 The single-frame mapping time in indicates that this method takes the shortest time in most frames. Figure 8 These are the experimental results in a forest environment. Compared with an indoor environment, a forest environment has more complex topological relationships and more irregular obstacles.

[0078] Table 1 shows the average time consumption of the three methods in each environment for building a single-frame topology map, as well as the average time consumption of each method in all environments. This method is affected by the number of obstacles, and the time consumption in the forest environment is higher than that in the indoor environment. However, the average time consumption in all environments is 26.8% lower than that of the predecessor method 1, which is much lower than the average time consumption of the predecessor method 2.

[0079] Table 1 Average time consumption for single frame mapping

[0080]

[0081] Table 2 Total number of nodes in the topology map

[0082]

[0083] Table 3 Total length of topological map

[0084]

[0085] Tables 2 and 3 show the total number of nodes and total length of the constructed topological maps. In terms of the number of nodes, the increase in the number of obstacles in the forest environment forms more complex topological relationships and more topological edges. Therefore, each method generates more topological nodes, but the number of nodes formed in this method is still lower than that of other methods. In terms of the total length of the map, the total length of the map generated by this method is several orders of magnitude lower than that of the other methods. The above experimental results prove the environmental adaptability and excellence of this method, and the generated map structure is more concise and lightweight.

[0086] The topological map of the real community is incrementally constructed based on the unmanned ground platform. The construction results are as follows: Fig. 9 As shown ( Fig. 9The right side is a partial enlarged view). The total number of nodes in the topological map is 147, and the total length of the map is 0.67 km. In the process of map construction, the present invention removes pedestrians, cyclists, and moving cars according to the classification information of the perception results, and does not put these targets into the base point when the Voronoi diagram is generated. Fig. 9 On the right are pedestrians, cyclists and moving cars encountered during driving. The corresponding map area is not affected by such targets. At the same time, a macro planning experiment is conducted based on the topological map constructed by the present invention. The starting point and the end point are randomly set within the map area. The orientation information of the starting point and the end point is not considered for the time being. The starting point and the end point are mapped to the nearest topological edge according to the Euclidean distance. Based on the A* algorithm, the shortest topological connection relationship is searched in the topological map. In about 470 random experiments, the average time consumption is 0.55ms. The time consumption curve is as follows: Fig.10 shown.

[0087] The above-mentioned detailed description of an incremental topological map construction method based on pruned Voronoi diagram with reference to the embodiment is illustrative rather than restrictive, and several embodiments can be listed according to the limited scope. Therefore, changes and modifications without departing from the overall concept of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for constructing an incremental topological map based on a pruned Voronoi diagram, characterized by: Map the boundary points of the drivable area to the global grid map; use the maximum empty circle property of Voronoi vertices to select the minimum generated area in the global grid map to achieve incremental topological map construction; prune the topological map by expanding the boundary points and filtering the invalid edges to retain the skeleton information of the environment; finally, build a fully connected topological map based on the breadth-first search algorithm. The specific steps are as follows: Step (1) Mapping the boundary points of the drivable area to the global grid map; Step (2) establish a rectangular envelope of the robot's circular sensing range, expand the rectangular envelope outward until the vertices on the Voronoi edge intersecting the rectangular envelope satisfy the following properties: the number of base points in the largest empty circle of the vertex is not less than 3, and none of them is a newly added drivable area boundary point, and generate a traditional Voronoi diagram with all drivable boundary points in the rectangular envelope as base points; Step (3) dilate the drivable boundary points in the generated area and delete the Voronoi edges in the occupied space; Step (4) construct a local undirected graph based on the Voronoi edges and vertices, delete invalid edges with short length and only one connection relationship, retain the skeleton information of the environment, and use the local undirected graph to replace the global undirected graph in the original rectangular frame; Step (5) starting from the robot position, traverse the undirected graph aimlessly based on the breadth-first search algorithm until the open list is empty, and delete the edges that have not been included in the open list and the closed list; The drivable area boundary points in step (1) are static obstacle points and effective perception boundary points in the environment, and when the drivable area boundary points are stored in the global grid map, only one point is retained in the same grid according to the grid map accuracy; The maximum empty circle property of the vertex in step (2) means that the circumference of the maximum empty circle of the Voronoi vertex contains at least 3 base points, and the base point set on the circumference of the maximum empty circle is obtained by finding the base point closest to the vertex; The principle of deleting the invalid edge in step (4) is: the length of the invalid edge is less than the sum of the robot length and the length threshold, and the invalid edge has only one connected relationship in the local undirected graph.

2. The incremental topological map construction method based on pruned Voronoi diagram according to claim 1 is characterized in that: The local undirected graph established in step (4) becomes a connected graph after being processed based on the breadth-first search algorithm, and there is a path between any two vertices.

Images