Automatic driving method, computer device, storage medium, program product, and mobile platform

Abstract

The invention discloses an automatic driving method, computer equipment, a storage medium, a program product and a mobile platform. The method comprises the following steps: determining a topological path and a road edge outer contour line in a sensing range according to environment sensing data; generating a road center line according to the topological path and the road edge outer contour line; and generating a driving track for automatic driving at least according to the road center line. The topological path and the along outer contour line are determined according to the environmental perception data of the vehicle, and the road center line is further generated according to the topological path and the along outer contour line for automatic driving. The method does not need to depend on the recognition of the lane line, and achieves the automatic driving on an unstructured road.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, and in particular to an automatic driving method, a computer device, a storage medium, a program product and a mobile platform. BACKGROUND

[0002] Current mainstream automatic driving solutions mainly focus on structured roads, and driving safety and stability mainly rely on lane lines constructed by humans. For unstructured roads (for example, rural roads, road sections with chaotic lane lines), current perception can only provide road edges and non-road areas, and cannot generate rich road semantics. Most self-driving systems need user intervention to take over when encountering scenes with low road quality. Therefore, there is an urgent need for a solution that can achieve automatic driving on unstructured roads. SUMMARY

[0003] Embodiments of the present application provide an automatic driving method, a computer device, a storage medium, a program product and a mobile platform, which are used to at least solve one of the above technical problems.

[0004] In a first aspect, embodiments of the present application provide an automatic driving method, comprising: determining a topological path and a road edge contour line within a perception range according to environmental perception data; generating a road center line according to the topological path and the road edge contour line; generating a driving trajectory for automatic driving according to at least the road center line.

[0005] In some embodiments, the determining a topological path within a perception range according to environmental perception data comprises: generating a Euclidean distance potential field map according to environmental perception data; generating a Vornoi diagram based on the Euclidean distance potential field map; and determining a topological path according to the Vornoi diagram.

[0006] In some embodiments, the generating a Euclidean distance potential field map according to environmental perception data comprises: generating a grid map of a current scene; determining an impassable area according to environmental perception data, and filling corresponding grids on the grid map with a preset value; determining a passable area according to environmental perception data, and filling the distance to the nearest obstacle in the corresponding grid on the grid map to obtain a Euclidean distance potential field map.

[0007] In some embodiments, the determining a road edge contour line within a perception range according to environmental perception data comprises: obtaining road edge scatter point data according to environmental perception data, and performing convex hull processing on the road edge scatter point data to obtain a single polygon contour; determine a road edge contour line according to the topological path and the polygonal contour.

[0008] In some embodiments, the single polygonal contour is obtained by performing convex hull processing on the road edge scatter data, including: performing convex hull clustering on the road edge scatter data; performing convex hull merging processing on the convex hull clustering result to obtain an initial polygonal contour; performing singular point filtering on the initial polygonal contour to obtain the single polygonal contour.

[0009] In some embodiments, the road centerline is generated according to the topological path and the road edge contour line, including: dividing the topological path into a constrained valid segment and an unconstrained fuzzy segment according to attributes of obstacles on both sides of the topological path; wherein the attributes of the obstacles include road element obstacles and non-road region obstacles, the road element obstacles correspond to the constrained valid segment, and the non-road region obstacles correspond to the unconstrained fuzzy segment; moving the road edge contour line corresponding to the constrained valid segment to the vehicle direction by a preset distance to generate the road centerline.

[0010] In some embodiments, before determining the topological path and the road edge contour line within the perception range according to the environmental perception data, the autonomous driving method further includes: generating the driving mode switching instruction in response to a driving mode switching operation of a user; or, performing scene recognition according to the environmental perception data, and generating the driving mode switching instruction when the scene recognition result meets a preset condition; when detecting the driving mode switching instruction, controlling the vehicle to switch from a first driving mode to a second driving mode; wherein the first driving mode is the current driving mode of the vehicle, and the second driving mode is an unstructured road autonomous driving mode.

[0011] In some embodiments, the autonomous driving method further includes: selecting the centerline according to at least one of safety, comfort and consistency; performing trajectory planning according to the selected centerline and obstacles on the centerline, for completing autonomous driving on unstructured roads.

[0012] In a second aspect, the present application provides a computer device, including a memory, a processor and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the method according to any one of the embodiments of the present application.

[0013] In a third aspect, the present application provides a computer readable storage medium, having stored thereon computer programs / instructions, wherein the computer programs / instructions, when executed by a processor, implement the steps of the method according to any one of the embodiments of the present application.

[0014] In a fourth aspect, the present application provides a computer program product, comprising computer programs / instructions, wherein the computer programs / instructions, when executed by a processor, implement the steps of the method according to any one of the embodiments of the present application.

[0015] In a fifth aspect, the present application provides a mobile platform, wherein the mobile platform is installed with the computer device according to any one of the embodiments of the present application.

[0016] The present application determines a topology path and an along-outline line based on the environment perception data of the vehicle, and further generates a road center line based on the topology path and the along-outline line for automatic driving. The automatic driving on an unstructured road is achieved without relying on the recognition of lane lines. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Figure 1 Flowchart of an embodiment of the automatic driving method of the present application; Figure 2 Flowchart of another embodiment of the automatic driving method of the present application; Figure 3 Effect diagram of the Euclidean distance potential field calculated by an embodiment of the present application; Figure 4 Schematic diagram of generating a Voronoi diagram by an embodiment of the present application; Figure 5 Schematic diagram of sampling a goal point candidate set by an embodiment of the present application; Figure 6 Flowchart of another embodiment of the automatic driving method of the present application; Figure 7 Schematic diagram of convex hull merging and single-side road along-outline extraction by an embodiment of the present application; Figure 8 Flowchart of another embodiment of the automatic driving method of the present application; Figure 9 Schematic diagram of local map cutting in the present application; Figure 10Flowchart of another embodiment of the automatic driving method of the present application; Figure 11 Schematic diagram of a rural mode state machine used in the automatic driving method of the present application; Figure 12 Flowchart of another embodiment of the automatic driving method of the present application; Figure 13 Structural schematic diagram of an embodiment of the computer device of the present application. DETAILED DESCRIPTION

[0019] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0020] It should also be noted that, in this document, the terms “comprise”, “contain”, not only include those elements, but also include other elements not explicitly listed, or include elements inherent to the process, method, article or device. Without more limitations, the elements defined by the statement “comprise” do not exclude the presence of other identical elements in the process, method, article or device comprising the described elements.

[0021] Current mainstream automatic driving schemes are mostly limited to structured roads, and are still in an exploratory pre-research stage for unstructured roads. Most automatic driving systems need user intervention to take over when encountering scenes with low road quality. Unstructured roads have complex working conditions, high diversity, difficult data collection, poor annotation consistency, low reliability of lane lines in perception and reasoning, and cannot guarantee driving safety and consistency.

[0022] The present application proposes an automatic driving method which can be applied to unstructured road scenes. The unstructured road is relative to the structured road.

[0023] The structured road generally refers to highways, urban trunk roads and other well-structured roads. Such roads have clear road marking lines, the background environment of the road is relatively simple, and the geometric characteristics of the road are also relatively obvious. Therefore, the road detection problem for it can be simplified as the detection of lane lines or road boundaries.

[0024] Unstructured road generally refers to urban non-main roads, rural roads and other roads with low degree of structuring. Such roads have no lane lines and clear road boundaries, and are difficult to distinguish between road areas and non-road areas due to the influence of shadows, water marks and the like. The variable road types, complex environmental backgrounds, shadows, water marks and changing weather are the difficulties faced by unstructured road detection and the main research direction of current road recognition technology.

[0025] As Figure 1 Fig. 1 shows a flowchart of an embodiment of the automatic driving method of the present application. The execution subject of the present application is a computer device, a server or a processor with computing function. The present application is illustrated by taking the application of the method in an automatic driving vehicle as an example. When the method is applied in an automatic driving vehicle, the execution subject of the method is an automatic driving vehicle control unit, and the present application includes the following steps: S10, determining a topological path and a road edge contour line in a perception range according to environmental perception data.

[0026] Exemplarily, the environmental perception data is obtained by sensors carried by the vehicle. The sensors include but are not limited to: a binocular camera, which can be installed at the front of the vehicle, used for shooting front road images, and outputting traffic elements (such as road edges, lane lines, center lines, guide strips, triangular islands and traffic lights, etc.) and static / dynamic obstacles (wherein the static obstacles are mainly temporary roadblocks such as cone barrels and stones; the dynamic obstacles include vehicles parked by the roadside or vehicles in motion, which are processed by computer vision as dynamic obstacles); a millimeter wave radar, which is installed at the front of the vehicle, used for detecting the distance of static / dynamic obstacles; a fisheye camera, which is installed at the side of the vehicle, with similar functions to the binocular camera, but with lower precision.

[0027] The environmental perception data includes but is not limited to at least one of the image data collected by the binocular camera, the detection data of the millimeter wave radar and the image data collected by the fisheye camera. According to the environmental perception data collected by one or more sensors carried by the vehicle, the topological path and the road edge contour line in the perception range are calculated.

[0028] S20, generating a road center line according to the topological path and the road edge contour line.

[0029] Exemplarily, the road center line can be generated according to the road edge contour lines on both sides of the topological path.

[0030] S30, generating a driving trajectory for automatic driving according to at least the road center line.

[0031] For example, the road centerline is used as the driving trajectory for autonomous driving. Furthermore, when an obstacle is detected on the road centerline, a driving trajectory to avoid the obstacle is generated based on the obstacle's location information and the road centerline for autonomous driving.

[0032] In this embodiment, the topological path and roadside outline are determined using the vehicle's environmental perception data. Furthermore, a road centerline is generated based on the topological path and roadside outline for autonomous driving. This eliminates the need for lane line recognition, enabling autonomous driving on unstructured roads.

[0033] like Figure 2 The diagram shown is a flowchart of another embodiment of the autonomous driving method in this application. In this embodiment, determining the topological path within the perception range based on environmental perception data includes: S11. Generate a Euclidean distance potential energy field diagram based on environmental perception data.

[0034] For example, a grid map of the current scene is generated; impassable areas are determined based on environmental perception data, and the corresponding grid cells on the grid map are filled with preset values; passable areas are determined based on environmental perception data, and the corresponding grid cells on the grid map are filled with the distance to the nearest obstacle to obtain a Euclidean distance potential energy field map. Here, the grid map is a grid-based map representation method, where the map area is divided into uniform grid cells, and each grid cell is assigned specific attribute information (e.g., whether it is occupied by an obstacle). The grid map can be drawn using tools such as MATLAB, and this application does not limit its creation.

[0035] The preset value can be 0, which means that the corresponding grid is a non-passable area. Non-passable areas include the roadside area (border area) and the non-road area (Nonroad area). Areas outside the non-passable areas are passable areas.

[0036] In this embodiment, various road elements (moving / stationary obstacles, etc.) are acquired through environmental perception data. Combined with border and nonroad regions, the values ​​of corresponding grid cells in the raster map are filled (border / nonroad grid cells represent a distance of 0 to the nearest obstacle, so the distance to the border / nonroad itself is also 0). Then, according to EDF (Euclidean Distance Field), a column scan is performed first, followed by a row scan, to fill each grid cell with the Euclidean distance to the nearest border and nonroad obstacles. All grid cells inside obstacles are set to 0 (i.e., all grid cells corresponding to the obstacle positions on the raster map are set to 0). Formula 1 below describes the update of the row direction after the column scan is completed. i The algorithm for calculating the Euclidean distance between points.j For row scanning and i Points on the same straight line in the grid.

[0037] f edf 2 = min j [ f row ( C j ) + || C j - C i || 2 ] Formula 1 Formula 1 shows how EDF updates the nearest obstacle distance to a given grid cell using a row-by-row scan (column scans are simply performed based on Euclidean distance). f row ( C j ) represent j The grid distance is the Euclidean distance to the nearest obstacle in the same column (this distance is squared to avoid the computational problems associated with square root extraction). Here, the grid coordinates in a row are represented as... C 1， C 2 … C M ,so C i This represents the geometric position of the target grid (obstacle grid) being solved. C j This represents other rasters that are in the same row as the target raster. f row ( C j ) represent j Distance to the nearest obstacle on the grid, || C j - C i || 2 represent i Grid to j The Euclidean distance of the grid. For example... Figure 3 The image shown is a rendering of the Euclidean distance potential energy field diagram calculated according to an embodiment of this application. It includes vehicle markings, purple areas representing impassable areas, green areas representing curb areas, and areas between green areas representing passable areas.

[0038] S12. Generate a Vinio diagram based on the Euclidean distance potential energy field diagram. The Vinio diagram in this embodiment is as follows: Figure 4As shown in the diagram. A Veno diagram is a set of continuous polygons (e.g., composed of nodes and edges) consisting of perpendicular bisectors of lines connecting two adjacent points, and has the following characteristics: (1) The Veno diagram divides the plane into n polygonal domains, each polygon containing only one generator; (2) The distance from a point within each polygon to that generator is shorter than its distance to other generators; (3) Points on the boundary of a polygon are equidistant from the two generators that generated the boundary; (4) The Voronoi polygon boundaries of adjacent graphs are subsets of the original adjacent boundaries; (5) A Voronoi graph has at most 2 * n - 5 vertices and 3 * n - 6 edges; (6) The generators inside the polygon are the centers of the outer circles of the triangle formed by the three points that make up the three sides, and all of these outer circles do not contain any vertices other than these three points (hollow circle property).

[0039] This application does not impose any restrictions on the specific algorithm for generating the Venn diagram.

[0040] For example, a fan-shaped region is constructed centered on the vehicle (the radius of the fan-shaped region is determined according to the sensing range, for example, taking the longest distance of 200m, and the central angle of 315°), and then samples are taken at equal intervals on the edge of the fan as candidate points for the target region; then, for the candidate points, representative points are calculated. n A representative point is selected as the final goal point. The sampling method is as follows: Figure 5 As shown, a candidate set of goal points is sampled, and combined with the grid probability value of the grid where the sampled point is located, sampling points that meet the set road threshold are selected (the road threshold is the threshold for whether a point is a non-road point; this road threshold can be set empirically, and the higher the road threshold, the higher the probability that a point is a non-road point; it is generally 0~10, so meeting the set road threshold means that the grid probability value of the grid where a point is located is lower than the road threshold). The sampling points that meet the set road threshold are then clustered using the k-means algorithm. The grid probability value represents the probability that a vehicle encounters an obstacle in the corresponding grid, and can be calculated by perceiving and analyzing the environment. The specific implementation algorithm is not limited in this application. The sampling point is the point sampled on the sector based on the central angle (e.g., 5°). The candidate point is the point where the Nonroad probability of the sampled point is less than 10. The goal point is selected from the candidate points by first clustering them according to geometric distance, and then selecting the point farthest from the obstacle in each cluster as the final target point to be reached by this road.

[0041] In some embodiments, a path search algorithm is used to select the point farthest from the obstacle in each cluster as the final destination point of the road. The path search algorithm typically uses a cost function plus a heuristic function to search for a path from the vehicle's position to the goal point. For each point on this optimal path, the cost function is simply the cost required to reach the current point from the starting point (including Euclidean distance and a cost coefficient for each grid cell depending on whether it contains obstacles). The heuristic function is the remaining cost from the current point to the goal point (also using Euclidean distance, grid cost, and angle cost). Finally, a path is found that minimizes the total cost of all points on the path plus the heuristic function. The heuristic function of the k-means algorithm uses the Euclidean distance between candidate points, as shown in Formula 2. After clustering, the road exit area for the current scene is obtained. Here, a road exit is defined as a path that can be taken from the vehicle's position to the boundary within the current perception range. Therefore, it is necessary to first find points below a certain threshold within the perception edge (in this case, a fan-shaped area). At the exit of each road, a set of points will be found (because the road width is generally not less than the perception accuracy, there will not be only one point). A set of points obviously cannot be used as exit points to assist the search in the following text. Moreover, the role of these points is actually to assist in searching for possible exits to this road. Therefore, a clustering operation is performed, selecting only one representative point for each road exit to simplify the problem and facilitate the search in the following text.

[0042] The above is Formula 2, where, x This refers to the two-dimensional coordinates of a road candidate point in the world coordinate system. x i This refers to the first point in the set. i Two-dimensional coordinates of a point u i yes x Calculate the average two-dimensional coordinates of all points. x i - u j Indicates the first i The Euclidean distance from each point to the average coordinates. C i It refers to the first i The set of heap points D ij Representative points concentration i Point to number j The distance between the centers of the point set.

[0043] Furthermore, the node corresponding to the vehicle's location is determined as the starting node of the Venn diagram, and the node corresponding to the exit point obtained in the above embodiment is determined as the ending node of the Venn diagram, thereby obtaining a directed Venn diagram. The edges in the Venn diagram can be constructed based on the passable paths determined by the Euclidean distance potential energy field diagram (for example, the path corresponding to the grid in the Euclidean distance potential energy field whose Euclidean distance is greater than a set threshold, where the set threshold may be, for example, the minimum width through which the vehicle can pass, and the specific value is not limited in this application).

[0044] S13. Determine the topological path based on the Vinio diagram.

[0045] For example, in the Vinonic path (i.e. Figure 4 The red topological path shown in the Veno diagram is used to search for all possible topological paths that conform to the vehicle's direction of travel using a depth-first search. The edges of the topological path are formed by points whose Euclidean distance meets a threshold (as mentioned in the previous embodiment, the Euclidean distance is greater than a set threshold, which will not be elaborated here). Simultaneously, the geometric positions of the nearest obstacles on both sides of the topological path are recorded (e.g., distances to the nearest two obstacles). Therefore, when a topological path is obtained, a set of information recording the geometric positions of the nearest obstacles on both sides of the topological path is acquired, which can be used to construct the edges of the obstacles on both sides of the topological path.

[0046] For example, during the process of determining the topology path, appropriate pruning is performed using rules such as path length and filtering duplicate paths to remove redundant paths. Finally, the backbone of the potential route and the edges formed by its associated obstacles are obtained. The backbone of the potential route serves as the topology path, and the edges formed by the associated obstacles define the boundaries of the backbone, which are then displayed. Figure 4 The Vinno diagram shown.

[0047] In some embodiments, path length refers to the Euclidean length of the edge in the Venn diagram. Pruning using path length includes directly deleting shorter paths. Pruning using duplicate path filtering rules includes filtering duplicate paths by, during depth-first traversal of the topological path search, if a duplicate edge identifier is encountered (each edge is pre-assigned a unique edge identifier), it is considered a loop topological path and the topological path is deleted.

[0048] like Figure 6 The diagram shown is a flowchart of another embodiment of the autonomous driving method in this application. In this embodiment, determining the outer contour line of the roadside within the perception range based on environmental perception data includes: S15. Obtain roadside scatter data based on environmental perception data, and perform convex hull processing on the roadside scatter data to obtain the outer contour of a single polygon.

[0049] Optionally, the following steps are included: performing convex hull clustering on the roadside scatter data; performing convex hull merging on the convex hull clustering results to obtain an initial polygonal outer contour; and performing singular point filtering on the roadside scatter points of the initial polygonal outer contour to obtain a single polygonal outer contour. For example, based on the border scatter data (roadside scatter data) obtained from environmental perception data, a small-scale convex hull clustering is first performed to obtain multiple convex hulls, and then the multiple convex hulls are merged into an initial polygonal outer contour.

[0050] Furthermore, during the development of this application, the inventors discovered that singularities exist in the initial polygonal outline due to perceptual errors. Therefore, they further filtered the obtained initial polygonal outline for singularities to obtain a usable polygonal outline. These singularities are falsely detected border points that intrude into the lane due to perceptual errors, leading to problems such as false braking. Therefore, it is necessary to eliminate the interference of these falsely detected points when acquiring the outline. Specifically, based on the characteristic that singularities cause raised bumps on the polygonal outline, these border points causing the raised outline can be identified and removed during the generation of the polygonal outline.

[0051] The above method for obtaining the outer contour of a polygon saves nearly 70% of the computational power compared to directly using alpha shape (an algorithm for extracting boundary points) to find the outer contour of a point set. The outer contour can meet the optimization requirements of this scenario.

[0052] S16. Determine the outer contour line of the curb based on the topological path and the outer contour of the polygon.

[0053] like Figure 7 The diagram shown illustrates the convex hull merging and single-sided path outer contour extraction process in one embodiment of this application. It includes the following steps: a. Convex hull cutting: This involves cutting off roadside points obtained from environmental perception data (e.g., ...). Figure 7 The red dots in the image are used for convex hull cutting to obtain multiple convex hulls. For example... Figure 7 As shown in the figure, the scattered points along the roadside are clustered and divided into three rectangular regions. b. Calculate the midpoint: Calculate the center point of the scattered points along the roadside contained in each region (i.e., the centroid of the convex hull corresponding to the scattered points in each region, such as...). Figure 7 (green dots in the middle) c. Merge polygons (i.e., convex hulls) and connect their centroids: e.g. Figure 7 As shown, the centroids of adjacent convex hulls are connected, and the line connecting the two centroids intersects the two outer contour edges of the two convex hulls to form two intersection points. d. Merge polygons and refine edges: Refine the outer contours of the two convex hulls. Then, when the distance between the two intersection points is less than a set threshold (e.g., 0.5m), merge the two convex hulls (e.g., merge the four vertices of the two outer contour edges to form a new polygon). Figure 7 The two new green edges in the middle are deleted simultaneously. Figure 7 The two gray outer contours of the convex hulls merge into one. e. Find the bottom left point and connect the entire outline: (e.g.) Figure 7 As shown in the image, find the bottom left blue dot and connect all the scattered points in the three rectangular areas to obtain the complete outline; f. Delete the line connecting the outline of the rectangular region and the centroid to obtain the outer outline of the whole polygon. g. Use the line selection results to obtain the single-sided outer contour: Use the line selection results of the previous frame to determine multiple road edge points on the side of the polygon outer contour that are close to the line selection results, and connect the multiple road edge points to form the single-sided outer contour.

[0054] In this embodiment, convex hull clustering is mainly used to first cluster small borders to obtain convex polygons (convex hulls). Then, adjacent convex hulls are gradually merged (adjacent is defined as centroid proximity; since the line connecting the centroids of a convex hull outwards must pass through the outer contour of its convex polygon, the line connecting the outer contours of two centroids must pass through one edge of their respective outer contours, leaving an intersection point). If the Euclidean distance between the two intersection points is less than a certain threshold (e.g., 0.5m), it is considered that the two convex polygons can be merged through these two outer contour edges. The four vertices of the two line segments are merged to form two new edges, and the two polygons are merged into one. Figure 7 As shown in the image, the gray lines have been separated and two green lines have been added. Following the steps described above, the convex polygons formed by the convex hull are gradually merged into a single convex polygon. Then, the lines selected in the previous frame (such as...) are used. Figure 7 The arrow to the right of the yellow line (the selected line can be understood as a reference line for planning and selection in autonomous driving, used to guide the vehicle along this line) finds the outer contour of the curb on the side closer to the selected line on the convex polygon (e.g., Figure 8 (The yellow line in the diagram). This method, compared to alphashape, does not require triangulation algorithms; it only uses convex hull clustering to quickly reduce the number of candidate point clusters, saving computational power.

[0055] like Figure 9 The diagram shown is a flowchart of another embodiment of the autonomous driving method of this application. In this embodiment, generating a road centerline based on the topological path and the outer contour line of the road edge includes: S21. Based on the attributes of obstacles on both sides of the topological path, the topological path is divided into constrained valid segments and unconstrained fuzzy segments; wherein, the attributes of the obstacles include road element obstacles and non-road area obstacles, the road element obstacles correspond to the constrained valid segments, and the non-road area obstacles correspond to the unconstrained fuzzy segments.

[0056] For example, based on the searched topological path, the nearest obstacle in the EDF potential energy field is sampled and matched with a certain precision (e.g., precision of 2m, i.e., searching for the information of the nearest left and right obstacles every 2m along the topological path). According to the obstacle attributes, the local map is divided into effective segments constrained by road edges and unconstrained fuzzy segments. If a certain area is constrained by road edges, lane lines, triangular islands, or other road elements on the left and right, it can be considered a constrained effective segment. The constrained effective segment can use the outer contour extraction scheme mentioned above to extract the single-sided outer contour, providing a half-plane constraint for the generation of the center line (preventing a point on the center line from crossing the half-plane of the single-sided outer contour). If there are no road edges, lane lines, triangular islands, or other road elements constraining, and only non-road constraints, it is an unconstrained fuzzy segment. In the optimization of the center line, only curvature smoothing is performed to clarify the region of interest for each topological path (e.g., the effective segment constrained by road edges, i.e., the area where the nearest left and right obstacles mentioned above can be matched with road elements).

[0057] like norm The diagram illustrates the local map segmentation process in this application. The yellow area represents constrained valid segments, the gray area represents unconstrained fuzzy segments, and the red line segments represent the filtered topological paths. The region of interest (i.e., the constrained valid segments) is associated with the roadside contour to determine the roadside's influence on the centerline, its left and right attributes, etc., for subsequent optimization calculations to generate the centerline. Here, the centerline serves as a reference line representing lane semantics in decision planning. Each reference line represents a lane within the decision planning process, and the trajectory of the decision planning follows the reference line. Therefore, the reference line itself ensures smoothness, no collisions, and provides lane semantics. Data association refers to the fact that the EDF potential field mentioned in the previous embodiment can acquire the nearest obstacle to a given grid (e.g., road elements encountered in rural areas). Therefore, when determining the left and right boundaries of a point in the region of interest, the two-dimensional coordinates and types of the nearest obstacle to the current point and its second nearest opposite obstacle can be directly obtained from the EDF potential field; this is called data association.

[0058] S22. Move the outer contour line of the roadside corresponding to the constrained effective segment a preset distance in the direction of the vehicle to generate the road centerline.

[0059] For example, the outer contour of the right road edge associated with the data is translated laterally towards the vehicle's direction. That is, the outer contour of the right road edge is moved laterally (e.g., along the y-axis) towards the vehicle's position. The curve after the movement is half the width of the lane (e.g., 3.75 / 2m) is used as the conjectured initial centerline. Points on the initial centerline have an initial state (including the point's two-dimensional coordinates, direction, curvature, etc.).

[0060] Furthermore, the hypothesized initial centerline is optimized. For example, points can be taken on the initial centerline at preset step sizes (e.g., intervals of 2m), and the initial state of these points can be optimized. The calculation of the angles and positions of each initial state on the initial centerline is shown in Formula 3. For the... i +1 point i +1 state, using the first i The first point i The position of the state is along the first... i The direction propagation of each state uses LQR (Linear Quadratic Regulator) to optimize the two-dimensional coordinates, direction, curvature, etc. of the conjectured initial centerline using gradient descent, and finally outputs the centerline on the right.

[0061] X i+1 = X i + X iq * norm Formula 3, Initial Solution State Propagation Formula in, X To optimize the state of a point on the line (including the point's two-dimensional coordinates, direction, curvature, etc.). X i+1 For the first i +1 state, X i For the first i One state, X iq * Figure 10 Representative at i At the position, advance a distance of precision along the direction of the angle (sin(θ), cos(θ)) (e.g., the precision of 2m mentioned in the previous embodiment).

[0062] like Figure 11 The diagram shown is a flowchart of another embodiment of the autonomous driving method of this application. In this embodiment, before determining the topological path and the outer contour line of the roadside within the perception range based on environmental perception data, the autonomous driving method further includes: S40. Generate the driving mode switching command in response to the user's driving mode switching operation; or, perform scene recognition based on environmental perception data, and generate the driving mode switching command when the scene recognition result meets preset conditions.

[0063] S50. When a driving mode switching command is detected, the vehicle is controlled to switch from a first driving mode to a second driving mode; wherein the first driving mode is the current driving mode of the vehicle, and the second driving mode is an unstructured road autonomous driving mode. The autonomous driving method described in the preceding embodiments is executed in the unstructured road autonomous driving mode; in the current driving mode of the vehicle (e.g., a structured road driving mode), an autonomous driving method suitable for structured roads can be executed, and the specific implementation can refer to relevant prior art, which is not limited in this application.

[0064] For example, in practical use, this embodiment allows users to manually trigger the generation of driving mode switching commands as needed, thereby completing the driving mode switch. For instance, when the vehicle is about to reach unstructured road scenarios such as rural areas, the user can trigger the driving mode switching command through voice commands, gesture operations, or touch operations on the vehicle's interactive screen, thereby switching the driving mode to the unstructured road autonomous driving mode.

[0065] Furthermore, scene recognition can be performed using perception data from the sensors mounted on the vehicle. When the scene recognition result indicates that the current scene is an unstructured road scene, the driving mode switching command is automatically generated, thereby switching the driving mode to the unstructured road autonomous driving mode. In this embodiment, no manual switching is required from the user, completely freeing the user's hands.

[0066] For example, scene recognition includes: identifying the proportion of structured roads in a road network based on perception data; when the identification result shows that the proportion of structured roads is lower than a set threshold (e.g., 70%), the current identified scene is determined to be an unstructured road scene. Further, scene recognition is performed using multiple consecutive frames of data; when the identification results of multiple consecutive frames all indicate that the proportion of structured roads is lower than the set threshold (e.g., 70%), the current identified scene is determined to be an unstructured road scene. Using multiple consecutive frames of data for scene recognition and judgment reduces misjudgments caused by single-frame errors and improves the reliability of determining unstructured road scenes.

[0067] In some embodiments, scene recognition further includes: determining the current scene as an unstructured road scene only when the navigation positioning determines that the current road is not a highway and the perception data identifies that there are missing or broken lane lines in the road, so as to prevent accidental touches in similar scenes such as highway intersections.

[0068] Furthermore, in order to improve the accuracy and safety of scene recognition based on environmental perception data and automatic triggering of driving mode switching, this application proposes to introduce a rural mode state machine.

[0069] like Figure 12 The diagram shows the rural mode state machine used in the autonomous driving method of this application. It includes an invalid state, an enabled rural mode state, an activated rural mode state, and a disabled rural mode state. These three states form a closed loop, with the invalid state being the state where the rural mode state machine is not triggered at all. The conditions for entering the enabled rural mode state in the disabled or invalid state are: the navigation positioning indicates the vehicle is not on a highway, and the perception data identifies low road quality (e.g., the proportion of structured roads in the current scene is below a set threshold). Only after the state machine enters the enabled rural mode state, and a non-intersection is detected with no normal route selection (normal route selection means an unusable route cannot be identified through structured roads), does it switch to the activated rural mode state. In the activated rural mode state, if the perception identifies high road quality or an intersection area, the state machine switches to the disabled rural mode state.

[0070] In this embodiment, the state transition of the rural mode state machine is triggered based on the scene recognition result, switching to rural mode to save computing power consumption in normal mode. At the same time, the rural mode state machine adds system mode recognition (that is, the decision planning internally identifies the scene based on navigation information and lane line quality assessment, and this mode is only enabled when the road quality is identified as low based on perception data and the current navigation positioning is not a highway), to prevent accidental touches in similar scenes such as highway intersections.

[0071] In some embodiments, after controlling the vehicle to switch from a first driving mode to a second driving mode (unstructured road autonomous driving mode) according to a detected driving mode switching command, the autonomous driving method of this application further includes: Choose a centerline based on at least one of safety, comfort, and consistency. Based on the selected centerline and the obstacles on the centerline, trajectory planning is performed to complete autonomous driving on unstructured roads.

[0072] For example, this application also includes a lane selection state machine for centerline selection. The lane selection state machine selects the centerline upon receiving an enable flag (e.g., contained in a driving mode switching command, which can be obtained by parsing the command). The criteria for selecting the centerline are: prioritizing safety and comfort, thereby selecting the centerline with the best consistency for lane keeping. Safety is defined as no collision ahead of the centerline (excluding various obstacles); comfort is defined as the lateral and longitudinal acceleration of the centerline being within a preset range (e.g., ±3.5 m / s^2); consistency requires that the points corresponding to the final centerline determined in the aforementioned embodiments and the initial centerline be sampled point-by-point, maintaining a certain threshold distance (0.5 m) in the lateral direction to evaluate the similarity of the two lines. When more than 80% of the sampled points meet the above threshold, the two curves are deemed to have the best consistency. Finally, a matching centerline is selected based on at least one of the above three factors; furthermore, path planning, based on the selected centerline, plans lateral deviation and bypass trajectories for static obstacles and stationary vehicles on the centerline to complete road driving on rural sections. This embodiment enables exploratory movement in unstructured road scenarios. The planned centerline trajectory, selected based on three factors—safety, comfort, and consistency—also aligns with the natural human habit of driving close to the side of the road. Furthermore, by planning trajectories that avoid stationary obstacles and vehicles, the usability of the trajectory is ensured.

[0073] like Figure 13 The diagram shown is a flowchart of another embodiment of the autonomous driving method of this application. In this embodiment, the autonomous driving method mainly consists of three parts: centerline estimation, state control, and path planning. The centerline prediction process includes: firstly, generating an EDF potential energy field based on the perceived input (e.g., environmental perception data obtained from vehicle sensors), obtaining a Venn diagram based on the EDF potential energy field, and then performing a topological path search to obtain the required topological path; secondly, performing convex hull merging and clustering polygon processing on the scattered points along the roadside based on the perceived input to obtain the single-sided outer contour of the roadside; furthermore, performing data association between the topological path and the single-sided outer contour of the roadside, as well as generating and optimizing the centerline, to finally obtain a candidate centerline that can be used for autonomous driving in unstructured road scenarios.

[0074] The state control section includes: road quality assessment and rural mode state machine. The state switching of the rural mode state machine is based on the road quality assessment results. When switching to the active rural mode state, an enable flag is generated and sent to the route selection state machine.

[0075] In some embodiments, the switching of autonomous driving modes can be implemented through a state control section. For example, switching between a structured road driving mode and an unstructured road driving mode can be implemented. Exemplarily, the state control section controls the execution of the centerline estimation section to switch to the unstructured road driving mode. For example, the state control section only controls the centerline estimation section to perform the centerline estimation step when it performs a road quality assessment and switches to the activated rural mode state, generating an enable flag.

[0076] The path planning section includes a route selection state machine and obstacle avoidance. When the route selection state machine receives an enable flag from the state control section, it selects a centerline (the specific selection method is described in the previous embodiment and will not be repeated here). Then, obstacle avoidance planning is performed based on the selected centerline, and the final planned trajectory is output for autonomous driving. This embodiment enables exploratory advancement in unstructured road scenarios. The centerline trajectory selected based on safety, comfort, and consistency aligns with the natural human tendency to drive close to the side of the road. Furthermore, by planning trajectories to avoid stationary obstacles and vehicles, the usability of the trajectory is ensured.

[0077] In some embodiments, this application provides a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in any embodiment of this application.

[0078] In some embodiments, this application provides a computer-readable storage medium storing a computer program / instructions thereon, characterized in that the computer program / instructions, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0079] In some embodiments, this application provides a computer program product including a computer program / instructions, characterized in that the computer program / instructions, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0080] In some embodiments, this application provides a mobile platform, characterized in that it is equipped with the computer device described in any embodiment of this application. The mobile platform may be a mobile device such as a vehicle, robot, or unmanned aerial vehicle.

[0081] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0082] Figure 13 This is a schematic diagram of the hardware structure of a computer device for executing an autonomous driving method according to another embodiment of this application, as shown below. Figure 13 As shown, the device includes: One or more processors 1310 and memory 1320, Figure 13 Take the 1310 processor as an example.

[0083] The device for performing the autonomous driving method may also include an input device 1330 and an output device 1340.

[0084] The processor 1310, memory 1320, input device 1330, and output device 1340 can be connected via a bus or other means. ​ Taking the example of a connection between China and Israel via a bus.

[0085] The memory 1320, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the autonomous driving method in the embodiments of this application. The processor 1310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 1320, thereby realizing the autonomous driving method of the above-described method embodiments.

[0086] Memory 1320 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the autonomous driving device, etc. Furthermore, memory 1320 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 1320 may optionally include memory remotely located relative to processor 1310, and this remote memory may be connected to the autonomous driving device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0087] Input device 1330 can receive input digital or character information and generate signals related to user settings and function control of the autonomous driving device. Output device 1340 may include display devices such as a display screen.

[0088] The one or more modules are stored in the memory 1320, and when executed by the one or more processors 1310, they execute the autonomous driving method in any of the above method embodiments.

[0089] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0090] The computer device in this application embodiment exists in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include: smartphones (e.g., iPhones), multimedia phones, feature phones, and low-end phones, etc.

[0091] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include PDAs, MIDs, and UMPCs, such as the iPad.

[0092] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes audio and video players (such as iPods), handheld game consoles, e-book readers, as well as smart toys and portable car navigation devices.

[0093] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0094] (5) Other electronic devices with data interaction functions.

[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An autonomous driving method, comprising: The topological path and outer contour of the roadside are determined based on environmental perception data; Generate the road centerline based on the topological path and the outer contour line of the roadside; At least the driving trajectory is generated based on the road centerline for autonomous driving.

2. The method according to claim 1, characterized in that, The step of determining the topological path within the sensing range based on environmental sensing data includes: Generate a Euclidean distance potential energy field map based on environmental perception data; A Vinno diagram is generated based on the Euclidean distance potential energy field diagram. The topological path is determined based on the Vinio diagram.

3. The method according to claim 2, characterized in that, The process of generating the Euclidean distance potential energy field map based on environmental perception data includes: Generate a raster map of the current scene; Based on environmental perception data, impassable areas are identified, and the corresponding grids on the grid map are filled with preset values. Based on environmental perception data, passable areas are determined, and the distances to the nearest obstacles are filled into the corresponding grids on the grid map to obtain a Euclidean distance potential energy field map.

4. The method according to claim 1, characterized in that, The step of determining the outer contour line of the curb within the sensing range based on environmental sensing data includes: Based on environmental perception data, roadside scattered data is obtained, and the convex hull of the roadside scattered data is processed to obtain the outer contour of a single polygon. The outer contour line of the curb is determined based on the topological path and the outer contour of the polygon.

5. The method according to claim 4, characterized in that, The convex hull processing of the roadside scatter data yields a single polygonal outer contour, including: Perform convex hull clustering on the roadside scatter data; The initial polygonal outer contour is obtained by performing convex hull merging on the convex hull clustering results. The initial polygonal outer contour is filtered for singular points to obtain a single polygonal outer contour.

6. The method according to any one of claims 1-5, characterized in that, Generating a road centerline based on the topological path and the outer contour of the road edge includes: Based on the attributes of obstacles on both sides of the topological path, the topological path is divided into constrained valid segments and unconstrained fuzzy segments; wherein, the attributes of the obstacles include road element obstacles and non-road area obstacles, the road element obstacles correspond to the constrained valid segments, and the non-road area obstacles correspond to the unconstrained fuzzy segments; The outer contour line of the roadside corresponding to the constrained effective segment is moved a preset distance in the direction of the vehicle to generate the road centerline.

7. The method according to any one of claims 1-5, characterized in that, Before determining the topological path and curb outline within the sensing range based on environmental sensing data, the process also includes: The driving mode switching command is generated in response to the user's driving mode switching operation; or, the driving mode switching command is generated when the scene recognition result meets the preset conditions based on environmental perception data. When a driving mode switching command is detected, the vehicle is controlled to switch from the first driving mode to the second driving mode; wherein the first driving mode is the current driving mode of the vehicle, and the second driving mode is the unstructured road automatic driving mode.

8. The method according to claim 7, characterized in that, After controlling the vehicle to switch from the first driving mode to the second driving mode based on the detected driving mode switching command, the following is also included: Choose a centerline based on at least one of safety, comfort, and consistency. Based on the selected centerline and the obstacles on the centerline, trajectory planning is performed to complete autonomous driving on unstructured roads.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-8.

11. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-8.

12. A mobile platform, characterized in that, The computer device of claim 9 is installed.

Images