Grid model map construction method for intelligent robot path planning

Abstract

The present invention provides a method for constructing a regular hexagonal grid model map for intelligent robot path planning, comprising: dividing a plane area within a visual range into a plurality of seamlessly connected, non-overlapping regular hexagonal blank basic grids according to a preset grid size combined with map data; obtaining point cloud data of each blank basic grid, assigning terrain features, cell attributes and grid edge attributes to each blank basic grid in combination with map data to obtain a basic grid, forming a grid model covering the plane area, determining the coordinates of the robot and the laser point in the grid model, calculating the probability estimate of the basic grid where the laser point is located, and determining whether the grid is in an occupied state or an idle state; updating the state of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located; and incrementally constructing a grid model of the environment according to the state of each basic grid. The map obtained by this method can be applied to map construction and path planning of intelligent robots.

Description

Technical Field

[0001] The present invention relates to a robot-oriented map construction method, in particular to a regular hexagonal grid model map construction method for intelligent robot path planning. Background Art

[0002] In the field of computer science, Ian Foster (1998) proposed the concepts of grid and grid computing. It integrates high-speed Internet, high-tech computers, large databases, sensors, and remote devices to provide users with more resources, functions, and interactivity. In essence, this type of grid refers to the establishment of a grid system to make information resources as ready-to-use as electricity resources. However, grid computing is too idealistic, and the composition of information resources is very complex, far from being comparable to electricity resources; sharing resources and solving problems in complex heterogeneous environments across platforms, organizations, and trust domains also presents great technical difficulties. Therefore, grid computing in the ubiquitous field has gradually been replaced by cloud computing, and the concept of grid has gradually weakened (Wang et al., 2016).

[0003] However, after the concept of grid was proposed, the concept of "spatial information grid" was proposed in the field of geographic information science, which is the fusion and integration of grid technology and spatial information technology. The grid model described in this article draws on the concept of grid in the field of geographic information science. The grid model for intelligent robots takes the robot as the main body to recognize the spatial environment it is in, which is essentially the same as taking people as the main body to recognize the spatial environment in which people are in. With the development of artificial intelligence, robots actively recognize the spatial environment, which is an inevitable product of technological development.

[0004] In the traditional sense, the grid in the field of geographic information science is another data organization form corresponding to the vector data structure, called the raster data model. It divides the spatial environment into regular grids, and assigns corresponding attribute values ​​to each grid to represent spatial entity data, which is ultimately used to describe the spatial environment. Chen Shupeng et al. (2002) believed that although the grid map based on the raster data model is a relatively simple map type, it divides the mapping area into grids according to plane coordinates or the longitude and latitude of the earth, and uses the grid as a unit to describe or express the attribute classification, statistical classification and change parameters therein, which is equivalent to expressing the law of dynamic spatiotemporal changes in two-dimensional space. Therefore, it has strong adaptability and diversity.

[0005] On the other hand, Li Deren et al. (2003) proposed the concept of spatial information multi-level grid from the perspective of urban management applications. It divides the global and national scope into grids of different coarse and fine levels according to different grid sizes. Each grid determines the geographical location by the longitude and latitude coordinates of its center point, and records basic data items closely related to the grid. Yuan Xiuxiao et al. (2005), Li Deren et al. (2006), and Zhang Yongzhi (2014) have studied the basic principles and application modes of spatial information multi-level grid from the perspectives of precision analysis, grid division, coding design, data retrieval, and scheduling strategies. Since the concept of spatial information multi-level grid originates from the extension of the connotation of urban management applications, most of the grids used are irregular polygons corresponding to administrative regions, which have great limitations.

[0006] At the same time, the gradual deepening of theoretical research and application practice of global-level geographic information requires a more effective multi-scale data management mechanism to achieve multi-level management and call of massive global data (Zhao Xuesheng, 2004). Therefore, GoodChild et al. (2000) proposed a spherical discrete grid (Discrete Global Grid) that "divides the large areas of the earth into gradually refined and mutually nested small areas based on the concept of cyclic segmentation." The spherical discrete grid abstracts the surface of the earth as a sphere, divides the sphere into a series of grid units according to certain rules, uses grid units to describe spatial position, form, and distribution, organizes and manages spatial data, and realizes spatial object modeling, analysis, and expression (Dutton, 1998; Sahr et al., 2003; Ben Jin, 2006; Yuan Wen, 2004; Zhao Xuesheng et al., 2007).

[0007] Spherical discrete grids not only meet the requirements of computers for data discretization processing, but also get rid of the constraints of map projections. They are expected to fundamentally solve the problems of data discontinuity, geometric deformation and topological inconsistency in global spatial data management and multi-scale operations of traditional plane models. However, from a technical perspective, spherical discrete grids are more of a spatial positioning scheme and spatial retrieval mechanism (Li Deren et al., 2003).

[0008] The research results of spatial information grids provide more segmentation methods for the research content of this article. Relatively speaking, the field of robotics focuses on probabilistic positioning, path planning, automatic control and other aspects, and less on the map model used by robots. Quijano and Garrido (2007) used hexagonal grids instead of quadrilateral grids to simulate the robot's exploration environment, and used a variety of different path search algorithms to analyze the impact of the two grids on the search efficiency (Quijano HJ, Garrido L. Improving Cooperative Robot Exploration Using an Hexagonal World Representation [C] / / Electronics, Robotics & Automotive Mechanics Conference. IEEE Computer Society, 2007.). Li et al. (2017) used hexagonal partitioning of the sampling area during the environmental monitoring of unmanned surface vehicles, and planned the path based on the spanning tree planning method (Spanning Tree-based Planning), and analyzed the path planning efficiency of unmanned surface vehicles (LI, Teng, et al. A hexagonal grid-based sampling planner for aquatic environmental monitoring using unmanned surface vehicles. In: 2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC). IEEE, 2017. p. 3683-3688.). Tao Zhe et al. (2020) fully compared the construction process of traditional grids, diamond grids, hexagonal grids and triangular grids in view of the shortcomings of traditional quadrilateral grids in environment description and path planning, studied the arrangement coding of hexagonal grids, analyzed the motion indicators corresponding to different grids based on the octree strategy, and verified the advantages of hexagonal grids in environment description, path planning and number of iterations through A* algorithm, Dijkstra algorithm and BFS algorithm (Tao Zhe, Gao Yuefei, Zheng Tianjiang, et al. Research on path planning based on A* algorithm in honeycomb grid map. Journal of North University of China: Natural Science Edition, 2020(4):310-317.). The above research contents have certain reference significance for this paper. However, they focus more on path planning itself rather than grid model, and also have certain limitations. Summary of the invention

[0009] In order to solve the above technical problems, the purpose of the present invention is to provide a method for constructing a regular hexagonal grid model map for intelligent robot path planning. The method improves the grid model from three aspects: terrain features, grid element attributes and grid edge attributes by expanding the attribute information of the grid model, so that the grid model is transformed from a subdivision algorithm into a new data structure that describes the spatial environment. The data structure can be well applied to the map construction and path planning process of the intelligent robot.

[0010] Based on the above purpose, the present invention provides a method for constructing a regular hexagonal grid model map for intelligent robot path planning, the method comprising:

[0011] Divide the plane area within the visual range into multiple seamlessly connected, non-overlapping regular hexagonal blank basic grids according to the preset grid size and map data;

[0012] The point cloud data of each blank basic grid is obtained, and the terrain features, cell attributes and grid edge attributes are given to each blank basic grid in combination with the map data to obtain a basic grid, thereby forming a grid model covering the plane area, wherein the terrain features are used to describe the ups and downs of the spatial environment; the cell attributes are used to describe the dominant geographical elements in the spatial environment where the basic grid is located; and the grid edge attributes are used to describe the geographical elements at the boundary of the spatial environment where the basic grid is located;

[0013] Determine the coordinates of the robot and the laser point in the grid model, calculate the probability estimate of the base grid where the laser point is located, and thus determine whether the base grid is in an occupied state or an idle state;

[0014] Update the status of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located;

[0015] The grid model of the environment is incrementally mapped according to the status of each basic grid.

[0016] Preferably, the terrain features use terrain feature factors to describe the terrain undulation within the range occupied by each basic grid, and the terrain feature factors include micro factors and macro factors. The micro factors include slope, aspect, plane curvature and profile curvature, and the micro factors include average slope, surface cutting depth, terrain undulation and elevation variation coefficient.

[0017] Preferably, the cell attributes include point elements, surface elements and some linear elements in the geographical environment, and the grid edge attributes include linear elements located at the boundaries of the basic grid in the geographical environment.

[0018] Preferably, the cell attributes of the basic grid include a superposition of place-occupying cells and indicative cells, wherein the place-occupying cells are spatial elements that occupy the largest area of ​​the current basic grid, or occupy a dominant position, or occupy an important position; and the indicative cells are spatial elements that occupy an extremely important position within the current basic grid area.

[0019] Preferably, the grid edge attribute of the basic grid is a superposition of an obstacle type grid edge and a placeholder type grid edge, the obstacle type is used to describe the "on" and "off" relationship between two adjacent basic grids, and the placeholder type grid edge exists in dependence on the obstacle type grid edge and is used to represent additional information on the grid edge.

[0020] As a preferred method, the specific method of dividing the plane area within the visual range into a plurality of seamlessly connected, non-overlapping regular hexagonal blank basic grids according to the preset grid size combined with the map data is:

[0021] Determine the grid starting point and grid orientation;

[0022] Determine the coordinates of each point and each edge of the grid, where the edges are labeled A, B, C, D, E, F, and the points are labeled 1, 2, 3, 4, 5, 6;

[0023] According to the mesh generation algorithm, the mesh size and the starting point coordinates are used to generate all the point coordinates of the entire blank mesh model, thereby realizing the mesh generation of the blank mesh model of the entire plane area.

[0024] Preferably, the specific method for determining the coordinates of the robot and the laser point in the grid model is:

[0025] Determine the positions of the robot and the laser point in the plane rectangular coordinate system, and use the grid direct coding method to determine the coordinates of the two in the grid model based on the mutual conversion relationship between the grid code and the plane rectangular coordinates in the blank basic grid model.

[0026] Preferably, a hidden Markov model is used to calculate the probability estimate of the basic grid where the laser point is located.

[0027] Preferably, the Bresenham algorithm is used to update the states of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located.

[0028] Preferably, the incrementally mapping the grid model of the environment according to the status of each basic grid is specifically: incrementally mapping the grid model of the environment using a map matching algorithm.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] First, the grid structure is used to expand the multiple expressions of spatial environment attributes, and the grid model is improved from three aspects: terrain characteristics, grid element attributes, and grid edge attributes. The grid model is transformed from a subdivision algorithm to a new data structure that describes the spatial environment. An innovative map model for robot cognition is proposed, which can be well applied to the map construction and path planning process of intelligent robots.

[0031] Second, the more advantageous regular hexagonal grid model is used to replace the occupancy grid map model, making the description of the spatial environment more reasonable. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings in the specification, which constitute a part of the present application, are used to provide a further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute a limitation on the present application.

[0033] Figure 1 is a schematic diagram of grid center distance analysis in an embodiment of the present invention;

[0034] FIG. 2(a)-(b) are frequency spectrum diagrams of a continuous band-limited signal after passing through a rectangular dot matrix sampling grid in an embodiment of the present invention;

[0035] FIG3 (a)-(b) are schematic diagrams of the spectrum and spatial sampling direction vector of a continuous signal after passing through a regular hexagonal sampling grid in an embodiment of the present invention.

[0036] Figure 4 is a schematic diagram of hexagonal grid properties in an embodiment of the present invention;

[0037] Figure 5 is a schematic diagram of a "forest" cell in an embodiment of the present invention;

[0038] Figure 6 is a schematic diagram of a "city" cell in an embodiment of the present invention;

[0039] Figure 7 is a schematic diagram of a “railway” cell in an embodiment of the present invention;

[0040] Figure 8 is a schematic diagram of a “river” cell in an embodiment of the present invention;

[0041] Fig. 9 is a flowchart of constructing a regular hexagonal grid model based on laser radar in an embodiment of the present invention;

[0042] Fig.10 is a schematic diagram of the grid size, starting point and orientation in an embodiment of the present invention;

[0043] Fig.11 is a schematic diagram of a direct encoding mechanism of a regular hexagonal grid model in an embodiment of the present invention;

[0044] Fig.12 Schematic diagram of equal-interval division of a rectangular coordinate system in an embodiment of the present invention;

[0045] Fig.13 is a hidden Markov model in an embodiment of the present invention;

[0046] Fig.14 It is a flow chart of a method for constructing a regular hexagonal grid model map for intelligent robot path planning in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present embodiment have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0050] This embodiment provides a method for constructing a regular hexagonal grid model map for intelligent robot path planning. Fig.14 As shown, the method includes:

[0051] Divide the plane area within the visual range into multiple seamlessly connected, non-overlapping regular hexagonal blank basic grids according to the preset grid size and map data;

[0052] The point cloud data of each blank basic grid is obtained, and the terrain features, cell attributes and grid edge attributes are given to each blank basic grid in combination with the map data to obtain a basic grid, thereby forming a grid model covering the plane area, wherein the terrain features are used to describe the ups and downs of the spatial environment; the cell attributes are used to describe the dominant geographical elements in the spatial environment where the basic grid is located; and the grid edge attributes are used to describe the geographical elements at the boundary of the spatial environment where the basic grid is located;

[0053] Determine the coordinates of the robot and the laser point in the grid model, calculate the probability estimate of the base grid where the laser point is located, and thus determine whether the base grid is in an occupied state or an idle state;

[0054] Update the status of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located;

[0055] The grid model of the environment is incrementally mapped according to the status of each basic grid.

[0056] As a preferred implementation method, the terrain features use terrain feature factors to describe the terrain undulation within the range occupied by each basic grid. The terrain feature factors include micro factors and macro factors. The micro factors include slope, aspect, plane curvature and profile curvature, etc. The micro factors include average slope, surface cutting depth, terrain undulation and elevation variation coefficient, etc.

[0057] As a preferred implementation, the cell attributes include point elements, surface elements and some linear elements in the geographical environment, and the grid edge attributes include linear elements located at the boundaries of the basic grid in the geographical environment.

[0058] As a preferred implementation, the cell attributes of the basic grid include the superposition of placeholder cells and indicator cells, wherein the placeholder cells are spatial elements that occupy the largest area of ​​the current cell or occupy a major position, or occupy an important position, such as water systems, residential areas, vegetation, etc. ( Figure 4 The indicative cell is a spatial element that occupies an extremely important position in the current cell area, such as a railway, a highway, etc.

[0059] Specifically, a placeholder cell refers to the main element that occupies the grid area. That is, if the basic grid occupies a certain spatial range, then within this range, the element that occupies the largest area, or occupies the main position, or occupies an important position is used as the cell attribute of the grid. Placeholder cells must highlight the important elements and ignore the minor elements, thereby roughly expressing the spatial environment. For example, the "forest" cell ( Figure 5 ) and the "city" cell ( Figure 6 ).

[0060] Indicative cells are special cells. They are not enough to constitute place-occupying cells. However, due to their extreme importance, they need to be transformed into place-occupying cells in specific applications. In order to distinguish them from place-occupying cells, this paper names them as indicative cells. For example, the traffic element, which has an important influence on robot motion and simulation model deduction, is still given cell attributes even though it is a linear element. For example, "railway" ( Figure 7 ); even if a grid has been assigned a placeholder attribute, in order to highlight the traffic information, the traffic information can be superimposed on the original attribute and defined as an indicative cell with obvious traffic characteristics, for example, "the main road in the city".

[0061] As a preferred implementation, the grid edge attribute of the basic grid is a superposition of an obstacle-type grid edge and a placeholder-type grid edge. The obstacle type is used to describe the "on" and "off" relationship between two adjacent grids. For example, the grid edge attribute is an uncrossable river, indicating that the current grid cannot cross the grid edge to reach the adjacent grid; the placeholder-type grid edge exists in dependence on the obstacle-type grid edge and is used to represent additional information on the grid edge, for example, the river ( Figure 8 ) is the main factor that hinders passage and becomes the main source of obstacle-type grid edges. However, for rivers, there are bridges and other appendages on the river. They depend on the river to exist, and the river is no longer an obstacle-type grid edge because of the bridge. Figure 4 A, B, C, D, E and F are shown. The attribute information of the grid model mainly comes from the map data. The terrain features, grid element attributes and grid edge attributes can be used as the data basis for real-time modeling of intelligent robots to better serve path planning applications.

[0062] For the description of spatial environment, terrain features describe the ups and downs of spatial environment, cell attributes describe point elements and surface elements, and some linear elements in spatial environment, and cell edge attributes describe linear elements in spatial environment. Together, they determine the basic framework of the spatial environment conceptual model. A large number of seamless, continuous and non-overlapping basic grids, as well as the accompanying attribute information, together constitute the basic appearance of spatial environment. In the path planning of intelligent robots or simulation models, terrain features, cell attributes and cell edge attributes play different roles. Intelligent robots or simulation models are affected by the slope of terrain. As the complexity of terrain increases, they need to pay more when passing through the grid, which in turn affects the farthest point they can reach. Cell attributes determine the different costs they need to consume when passing through cells with specific attributes. Similar to grids of different soils, hard soil cells can always consume less cost and run farther than soft soil cells. The grid edge properties determine whether they can cross the grid to reach the adjacent grid. Grid edge properties such as rivers, steep slopes, and gullies have a certain hindering effect on robot movement and simulation model deduction. Only when the performance of robots and simulation models exceeds the limits of rivers, steep slopes or gullies will they not have an hindering effect on them. Otherwise, their movement may be hindered.

[0063] The terrain characteristics, cell attributes and edge attributes of the grid model are all reflected in the path planning application through cost values.

[0064] As a preferred implementation, the specific method of dividing the plane area within the visual range into a plurality of seamlessly connected, non-overlapping regular hexagonal blank basic grids according to the preset grid size combined with map data is as follows:

[0065] Determine the grid starting point and grid orientation;

[0066] Determine the coordinates of each point and each edge of the grid, where the edges are labeled A, B, C, D, E, F, and the points are labeled 1, 2, 3, 4, 5, 6;

[0067] According to the mesh generation algorithm, the mesh size and the starting point coordinates are used to generate all the point coordinates of the entire blank mesh model, thereby realizing the mesh generation of the blank mesh model of the entire plane area.

[0068] As a preferred implementation, the specific method for determining the coordinates of the robot and the laser point in the grid model is:

[0069] Determine the positions of the robot and the laser point in the plane rectangular coordinate system, and use the grid direct coding method to determine the coordinates of the two in the grid model based on the mutual conversion relationship between the grid code and the plane rectangular coordinates in the blank basic grid model.

[0070] As a preferred implementation, a hidden Markov model is used to calculate the probability estimate of the basic grid where the laser point is located.

[0071] As a preferred implementation, the Bresenham algorithm is used to update the states of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located.

[0072] As a preferred implementation, the incremental mapping of the grid model of the environment according to the status of each basic grid is specifically: incremental mapping of the grid model of the environment using a map matching algorithm.

[0073] In actual application, map data and laser sensors are the two main sources for modeling regular hexagonal grid models for intelligent robots. As the main way to describe the spatial environment, map data provides a basic grid for the regular hexagonal grid model. Laser sensors can provide very current point cloud data for the regular hexagonal grid model, which is suitable for real-time modeling and rapid updating of the grid model. In practical terms, when using laser sensors to build a regular hexagonal grid model, the basic grid can be built based on the map data, and then the relative point cloud data obtained by the laser sensor can be integrated to achieve real-time modeling and rapid updating of the grid model.

[0074] The grid model refers to dividing the area into a series of regular polygonal grid units according to certain rules. The grid units are used to describe the spatial position, element attributes and spatial distribution, and to realize the organization, management, modeling, analysis and expression of spatial data. The following differences between the regular quadrilateral model and the regular hexagonal model in terms of environment description and path planning demonstrate the rationality and feasibility of applying the regular hexagonal grid model to robot maps:

[0075] The geometric structure refers to a series of regular polygonal grid units covering a plane or sphere. They continuously, seamlessly and non-overlappingly divide the spatial environment, determine the spatial position and describe the spatial attributes. Considering the particularity of the grid covering the study area, these regular polygonal grids should meet at least two conditions: First, the distance from the center point of the grid to each edge is the same in all directions. Ideally, the distance from the center point of the grid to the edge in the 360° direction is equal, that is, a circle. However, the circle cannot be continuously spliced ​​in the region, so it is necessary to select an inscribed regular polygon of the circle instead of the circle for approximation. Second, the grid must be a graphic that can be spliced ​​continuously and without overlap in the region.

[0076] According to the above two conditions, the geometric structure of the appropriate mesh model can be determined.

[0077] Assume that there is a regular n-gon, then each interior angle ΔΑ is:

[0078]

[0079] If the figure can form a continuous, seamless and non-overlapping splicing on the plane, it must be necessary that m regular n-gons each have an internal angle spliced ​​at one point, just completely covering the plane. Therefore,

[0080]

[0081] From equation (1) and equation (2), we can get:

[0082]

[0083] Where m and n are integers, and n can only be 3, 4, and 6. Therefore, there are only three types of regular polygons that conform to the geometric structure of the mesh model, namely regular triangles, regular quadrilaterals, and regular hexagons. In theory, these three types of regular polygons can be used as the geometric structure of the mesh model.

[0084] (1) Grid center distance analysis

[0085] The center distance of a mesh refers to the distance from the center point of the mesh to the edge of the mesh. The geometric structure of the mesh model is usually approximated by a regular polygon inscribed in a circle. Obviously, the distance from the center point of the mesh to the edge of the mesh will change unequally. Among them, the vertical distance from the center point of the mesh to one side of the regular polygon is the closest center distance (D min ), which is also called the equidistant direction; the distance from the center point of the grid to the vertex of the regular polygon is the farthest center distance (D max The ratio of the closest center distance to the farthest center distance is called the grid center distance error ( Figure 1 ). Generally, a nearly ideal grid should have more equidistant directions and smaller grid center distance errors.

[0086] The grid center distance analysis is performed on regular triangles, regular quadrilaterals and regular hexagons respectively, and the analysis results are shown in Table 1.

[0087] Table 1 Grid center distance error table

[0088] category Equidistant direction ( / ) error Regular triangle 3 2.000 Regular quadrilateral 4 1.414 Regular hexagon 5 1.150

[0089] It can be found that: ① From the perspective of equidistant directions, the regular hexagon has the most equidistant directions. The more equidistant directions there are, the more directions can be selected from the center point of the grid, thus obtaining more reasonable experimental results. ② From the perspective of grid center distance error, the errors of regular quadrilaterals and regular hexagons are smaller than those of regular triangles. The smaller the error, the more accurate the approximation results can be obtained. Therefore, both regular quadrilaterals and regular hexagons can better meet the experimental requirements, among which the advantages of regular hexagons are more prominent.

[0090] (2) Grid sampling density analysis

[0091] Covering the study area with a suitable grid and then determining the attribute information of each grid is a process similar to building a "digital image" and is called "digital image" sampling. The grid density within a certain range is called grid sampling density.

[0092] Assume that the Fourier transform result of the two-dimensional continuous image information is shown in the shaded part of Figure 2(a), that is, the frequency band of the signal is in a circular area with a radius of W in the frequency domain. The spectrum of the signal after the continuous band-limited image passes through a regular quadrilateral sampling grid is shown in Figure 2(b). The sampling interval in the horizontal direction is T1, and the sampling interval in the vertical direction is T2. The sampling interval satisfies the Nyquist sampling theory, that is,

[0093] The arrangement of the spectrum of the sampled signal in the frequency domain is determined by the sampling matrix. The sampling matrix of the regular quadrilateral grid is a diagonal matrix V, that is:

[0094]

[0095] As can be seen from Figure 2(a), the space occupied by the regular hexagon circumscribed by the circle is smaller than that of the regular quadrilateral. After the continuous image signal passes through a regular hexagonal sampling grid that satisfies the Nyquist sampling theorem, the spectrum of the signal formed is shown in Figure 3(a). It can be seen that the signal spectrum after sampling is much tighter in the frequency domain space arrangement than the signal spectrum after sampling through the regular quadrilateral sampling grid. The spatial sampling interval is determined by two direction vectors v1 and v2, as shown in Figure 3(b). The sampling matrix V is composed of two direction vectors v1 and v2, that is:

[0096]

[0097] The sampling grid in Figure 3 is a regular hexagon, so there is Since the spectrum of the continuous signal is in a circular area with a radius of W, in order to satisfy the Nyquist sampling theorem, that is, The sampling matrix of a regular hexagon is:

[0098]

[0099] The sampling density is the inverse of the determinant of the sampling matrix. Therefore, by comparing equation (4) and equation (6), it can be proved that the minimum sampling density of the regular hexagonal structure is 13.4% lower than that of the regular quadrilateral structure. It can be seen that in the theoretical case where the grids have the same sampling density, the approximation quality of the regular hexagonal grid is always better than that of the regular triangle or regular quadrilateral grid.

[0100] Combining the basic grid constructed based on the map data, the relative position coordinates of the robot, and the plane rectangular coordinates of the laser point, the basic process of constructing a regular hexagonal grid model based on the laser sensor is as follows: Fig. 9 As shown. First, a basic grid model is constructed according to the map data and the set grid size, including the establishment of the geometric structure of the grid model, and the assignment of terrain features, cell attributes and grid edge attributes to the grid model according to the map elements. This process is called grid modeling based on map data; however, it is not intended to fully describe how to use map data to fill the grid model. Therefore, only a simple description of the geometric modeling of the grid is given, that is, the basic grid model generated in this step is a blank grid model. Secondly, according to the coordinates of the robot and the laser point, their coordinates in the grid model are determined, and the hidden Markov model is used to calculate the probability estimate of the grid where the laser point is located, and the occupied state or idle state of the grid is determined; thirdly, the Bresenham algorithm is used to update the state of other grids between the two grids; finally, the map matching algorithm is used to realize the incremental mapping of the environmental grid model.

[0101] Specifically, the geometric modeling of the mesh model requires clear information such as mesh size, mesh starting point, mesh orientation, etc. Fig.10 As shown, the grid size refers to the distance H between opposite sides of the regular hexagonal grid, the grid starting point is the lower left corner, and the grid orientation is that the grid side faces north;

[0102] Each regular hexagonal grid of the grid model can be regarded as an independent unit, storing uniformly encoded attribute information and interacting with the robot or simulation model independently. Each grid consists of two parts: grid element and grid edge. Therefore, the geometric modeling of the grid model is essentially to determine the coordinate values ​​of each point and each edge of the grid, where the edges are marked as A, B, C, D, E, F, and the points are marked as 1, 2, 3, 4, 5, 6, such as Figure 4 shown.

[0103] Assuming that the coordinates of the center point of the lower left grid of the grid model are O(X0, Y0) and the grid size is H, then the coordinates of the center point of the grid in the row and column can be calculated by formula (7), and the coordinates of each point in the grid can be calculated by formula (8).

[0104]

[0105] According to the mesh generation algorithm, the coordinates of all points of the entire mesh model are generated using information such as the mesh size and the coordinates of the starting point, thereby achieving mesh generation of the mesh model of the entire plane area.

[0106]

[0107] In the process of interaction between a robot or simulation model and a grid model, it is usually necessary to convert grid codes to plane rectangular coordinates, such as Fig.11 As shown in Figure 11, the robot or simulation model can be quickly and effectively operated in space. Therefore, network coding needs to meet the requirements of coordinate transformation and spatial operation. Typical grid coding mechanisms include generalized balanced ternary (GBT) (Gibson et al., 1982; Sahr et al., 2003), "subordinate graph" structure (Ben Jin, 2005; Zhang Yongsheng et al., 2007; Tong Xiaochong, 2010), PYXIS structure (Perry, 2006; Vince et al., 2009; YONG et al., 2003), etc. The grid direct coding method effectively meets the above two requirements (Figure 11). It uses a two-dimensional array to manage the attribute information of the grid. The direct coding method can save nearly half of the computer storage space, which is of great significance for grid models in large areas.

[0108] The conversion formula from plane rectangular coordinates to grid coding is shown in equation (9).

[0109]

[0110] The conversion formula from grid coding to plane rectangular coordinates is shown in equation (10).

[0111]

[0112] Among them, I hex , J hex are the horizontal and vertical codes of the grid, respectively, and X and Y are the horizontal and vertical coordinates of the rectangular coordinate system. min ,Y min) is the lower left corner of the study area in the rectangular coordinate system, (X max ,Y max ) is the upper right corner of the study area in the rectangular coordinate system, O is the origin of the rectangular coordinate system, X dis is the distance between them in the X direction, dis is the spacing distance in the Y direction (such as Fig.12 ).

[0113] The regular hexagonal grid model mapping for intelligent robots is essentially a process of simultaneous positioning and mapping of robots. Generally speaking, it uses a probabilistic method to represent the possibility of the robot's state at a certain moment. Assume that the robot's state information at a certain moment is x t , the control information is u t , the sensor measurement information is z t , then the robot system mainly solves two probabilities: state transition probability p(x t |x t-1 ,u 1：t ) and the measurement probability p(z t |x t ).

[0114] Hidden Markov models describe the basic process of robot movement and measurement ( Fig.13 ), which shows that the robot's state at this moment is only related to the state x at the previous moment t-1 and the current control information u t Regarding the robot’s measurement information z at time t t Only the current state information x t related.

[0115] The odometer detects the robot's linear velocity, angular velocity and other control variables in real time, and determines the robot's current position information based on the state information at the previous moment; the laser sensor scans the surrounding environment at a specific frequency to obtain measurement information of the environment for building a grid model.

[0116] Assuming that the robot's position is known, the construction of the grid model solves the problem of how to generate a consistent map based on noisy and uncertain measurement data. The grid model uses a series of binary random variables to represent the map. The value of the binary random variable indicates whether the current position is free or occupied by an obstacle. Therefore, the map construction process is to calculate the posterior probability p(m|x 1：t , z 1：t ) process.

[0117] The grid model discretizes the two-dimensional space into independent units of the same size, each of which represents the area contained in the grid. Since each unit in the grid model is either occupied or idle, we can use p(m i ) represents the probability that a unit is occupied by an obstacle, p(m i )=1 indicates that the unit is occupied, p(m i )=0 indicates that it is in an idle state. Assuming that each unit in the grid map is independent of each other, the posterior probability p(m|x 1：t , z 1：t )for:

[0118]

[0119] Among them, x 1：t is the state (pose) sequence of the robot, z 1：t is the measurement sequence of the robot.

[0120] For any independent unit, the posterior probability can be obtained by the Bayesian criterion:

[0121]

[0122] According to the hidden Markov model, when x t When zt is known, 与 x 1：t-1 and z 1：t-1 It doesn't matter, so

[0123] p(z t |m i , x 1：t , z 1：t-1 )=p(z t |m i , x t ), p(z t |x 1：t , z1: t-1 )=p(z t |x t ). p(z t |m i , x t ) is simplified by the Bayesian criterion as:

[0124]

[0125] If the measurement information z t If it is unknown, then the robot state x at time t t Does not include independent regions m in the raster map i information, that is, x t For m i Cannot provide any information, there is p(mi |x t )=p(m i ),

[0126] p(m i |x 1：t , z 1：t-1 )=p(m i |x 1：t-1 , z 1：t-1 ). Rearrange p(m i |x 1：t , z 1：t ) can be obtained:

[0127]

[0128] Each unit has only two states: occupied and idle, so the probability of the corresponding independent unit being idle is It is expressed as:

[0129]

[0130] in, Represents the probability that the unit is idle.

[0131] and, so:

[0132]

[0133] Eliminate the irrelevant terms in equation (15) and equation (16) by reducing them, then:

[0134]

[0135] Taking the logarithm of formula (17), we can obtain:

[0136]

[0137] If the log-odds ratio is used to represent the posterior probability, then equation (16) is expressed as:

[0138]

[0139] Among them, l t-1，i is the logarithmic geometry of the robot’s posterior probability at time t-1, l 0，i is the initial value before the sensor measurement, and Equation (19) satisfies the structure of recursive calculation of unit occupancy probability. That is to say, through the robot's posture state and observation information at time t, the occupancy probability of any unit can be updated based on the existing grid model probability at the previous moment.

[0140] In this way, the construction algorithm of the grid model is shown in Algorithm 1.

[0141]

[0142] The process of updating the grid model is to determine the grid through which the connection line passes based on the connection line between the grid where the robot is located and any grid with a certain state, and estimate the state of the corresponding grid. For the regular quadrilateral grid model, this process uses the Bresenham algorithm to update, and for the regular hexagonal grid model, a specific processing algorithm is required.

[0143] Assume the robot's position is (x R ,y R ), the coordinates returned by the sensor scan are (x S ,y S ), according to the regular hexagonal grid model, the grid number of the coordinates returned by the sensor scan is (i S , j S ), then the probability value of it being occupied by an obstacle is determined by Algorithm 1. S , j S ) is determined, the occupied (or idle) state of other grids passed by the scan line can be determined based on the robot's position, the coordinates and grid numbers returned by the sensor scan. This process is shown in Algorithm 2.

[0144]

[0145] Probability estimation and map updating solve the problem of building a robot map when it is in a fixed position. As the robot continues to move, the probability values ​​of all grids are constantly updated. Therefore, different grid models of the same resolution need to be continuously spliced ​​through incremental mapping until a complete grid model is formed.

[0146] The computer stores each grid of the grid model in a matrix array, and each matrix element corresponds to a grid. In this way, the complete environment grid model corresponds to a pixel in the matrix array. Assume that the robot constructs the grid model Map at different times G and Grid Model Map W If there are overlapping areas, the problem is how to stitch them together to achieve incremental mapping of the environment mesh model. First, use the edge extraction algorithm to extract the edges from the mesh model Map. G and Grid Model Map W Extract edge pixels to form an edge pixel point set, that is, and where g i and w i is a vector, N g and N w Respectively represent the number of elements in the edge pixel set G and W. Grid model MapG and Grid Model Map W The edge pixel point set of the overlapping area is G ε It is a subset of G and W, and the overlap percentage is ε. Then, the incremental mapping problem of the grid model at different times is transformed into an image registration problem. The rigid body transformation T = {R, t} of the robot is calculated so that the point set T (G) after the transformation can be well matched with the point set W. Then, the incremental mapping problem of the grid model is expressed as a minimization problem, that is:

[0147]

[0148] where is the control parameter, |·| represents the minimum overlap percentage of the number of elements in the set, and ε min Through the continuous updating and matching of the mesh model, the mesh model of the robot is gradually and incrementally constructed, and finally a complete mesh model is formed.

[0149] The present invention improves the grid model from three aspects, namely, terrain features, cell attributes and grid edge attributes, by expanding the attribute information of the grid model, so that the grid model is transformed from a subdivision algorithm into a new data structure for describing the spatial environment. This data structure can be well applied to map construction and path planning applications of intelligent robots.

[0150] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. Ordinary technicians in the field can change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and purpose of the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for constructing a regular hexagonal grid model map for intelligent robot path planning, characterized in that: include: Divide the plane area within the visual range into multiple seamlessly connected, non-overlapping regular hexagonal blank basic grids according to the preset grid size and map data; The point cloud data of each blank basic grid is obtained, and the terrain features, cell attributes and grid edge attributes are given to each blank basic grid in combination with the map data to obtain a basic grid, thereby forming a grid model covering the plane area, wherein the terrain features are used to describe the ups and downs of the spatial environment; the cell attributes are used to describe the dominant geographical elements in the spatial environment where the basic grid is located; and the grid edge attributes are used to describe the geographical elements at the boundary of the spatial environment where the basic grid is located; Determine the coordinates of the robot and the laser point in the grid model, calculate the probability estimate of the base grid where the laser point is located, and thus determine whether the base grid is in an occupied state or an idle state; Update the status of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located; Incrementally map the grid model of the environment based on the status of each basic grid; The terrain features use terrain feature factors to describe the terrain undulation within the scope of each basic grid. The terrain feature factors include micro factors and macro factors. The micro factors include slope, slope aspect, plane curvature and profile curvature. The micro factors include average slope, surface cutting depth, terrain undulation and elevation variation coefficient. The cell attributes include point elements, surface elements and some linear elements in the geographical environment, and the grid edge attributes include linear elements located at the boundaries of the basic grid in the geographical environment; The cell attributes of the basic grid include the superposition of place-occupying cells and indicating cells. The place-occupying cells are the spatial elements that occupy the largest area of ​​the current basic grid or occupy a major position or an important position; the indicating cells are the spatial elements that occupy an extremely important position within the area of ​​the current basic grid. The grid edge attribute of the basic grid is the superposition of the obstacle type grid edge and the placeholder type grid edge. The obstacle type is used to describe the "connected" and "disconnected" relationship between two adjacent basic grids. The placeholder type grid edge exists in dependence on the obstacle type grid edge and is used to represent additional information on the grid edge.

2. A method for constructing a regular hexagonal grid model map for intelligent robot path planning according to claim 1, characterized in that: The specific method of dividing the plane area within the visual range into multiple seamlessly connected, non-overlapping regular hexagonal blank basic grids according to the preset grid size and map data is as follows: Determine the grid starting point and grid orientation; Determine the coordinates of each point and each edge of the grid, where the edges are labeled A, B, C, D, E, F, and the points are labeled 1, 2, 3, 4, 5, 6; According to the mesh generation algorithm, the mesh size and the starting point coordinates are used to generate all the point coordinates of the entire blank mesh model, thereby realizing the mesh generation of the blank mesh model of the entire plane area.

3. The method for constructing a regular hexagonal grid model map for intelligent robot path planning according to claim 1, characterized in that: The specific method for determining the coordinates of the robot and the laser point in the grid model is: Determine the positions of the robot and the laser point in the plane rectangular coordinate system, and use the grid direct coding method to determine the coordinates of the two in the grid model based on the mutual conversion relationship between the grid code and the plane rectangular coordinates in the blank basic grid model.

4. The method for constructing a regular hexagonal grid model map for intelligent robot path planning according to claim 1, characterized in that: A hidden Markov model is used to calculate the probability estimate of the underlying grid where the laser point is located.

5. The method for constructing a regular hexagonal grid model map for intelligent robot path planning according to claim 1, characterized in that: The Bresenham algorithm is used to update the status of other basic grids between the basic grid where the robot is located and the basic grid where the laser point is located.

6. The method for constructing a regular hexagonal grid model map for intelligent robot path planning according to claim 1, characterized in that: The incremental mapping of the grid model of the environment according to the status of each basic grid is specifically: using a map matching algorithm to incrementally map the grid model of the environment.

Images