An autonomous valet parking planning method and system based on a scale adaptive index model

Abstract

This invention discloses an autonomous valet parking planning method and system based on a scale-adaptive index model, belonging to the field of autonomous driving technology. The method includes: acquiring radar scan data of a parking lot and converting the parking scene map into a multi-scale grid map; constructing a scale-adaptive index model based on the multi-scale grid map; extracting the main channel skeleton path based on the scale-adaptive index model and optimizing and smoothing the skeleton; calculating the number of intersections between each edge of the parking space and the multi-scale grid map based on the optimized main channel skeleton path; calculating the parking position offset and yaw angle and determining the geometric parking position attitude; and performing autonomous valet parking planning based on the geometric parking position attitude and the scale-adaptive index model. This invention solves the problems of low efficiency and low success rate of existing path planning methods in complex and narrow parking lot environments.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to an autonomous valet parking planning method and system based on a scale-adaptive indexing model. Background Technology

[0002] In autonomous valet parking (AVP) scenarios, the complex environment of narrow parking lots poses a significant challenge to path planning algorithms. Traditional path planning methods have significant limitations: some algorithms perform well in static environments but have poor adaptability to dynamic environments and high computational complexity; while other algorithms rely on the initial solution and the accuracy of environment modeling, making them prone to getting trapped in local optima; existing algorithms generally adopt a fixed-scale planning framework, which is difficult to adapt to the multi-dimensional constraints in parking lots, leading to planning failures or suboptimal paths in complex scenarios.

[0003] In autonomous valet parking (AVP) scenarios, the complex environment of narrow parking lots poses a significant challenge to path planning algorithms. Existing technologies have significant limitations:

[0004] First, they have poor adaptability to dynamic environments. Some traditional algorithms perform well in static environments, but when faced with dynamic obstacles, their insufficient real-time update capability prevents them from responding quickly, leading to invalid or infeasible planning results. For example, if a pedestrian or other vehicle suddenly appears in a parking lot, the algorithm may not be able to complete the replanning in a short time, making it difficult to meet real-time requirements.

[0005] Secondly, they are prone to getting trapped in local optima. Some algorithms are highly dependent on the accuracy of the initial solution or environmental modeling, making it difficult to perform path search from a global perspective. In a narrow parking lot, if the algorithm fails to fully consider global characteristics during initial planning and relies only on local information, it is prone to getting trapped in local optima. For example, gradient descent-based algorithms may be "pulled" into local optima by initial conditions due to the non-convexity of the objective function, making it difficult to escape and find the global optimum.

[0006] Finally, the planning framework is rigid. Existing algorithms generally use a fixed-scale planning framework, which is difficult to adapt to changes in multi-dimensional constraints within parking lots. For example, when faced with railings of different heights or passageways of different widths, the fixed-scale framework cannot accurately model them, leading to failed planning paths or failure to meet actual needs. In addition, the fixed-scale framework struggles to capture multi-scale information about environmental features; for example, local fine-grained information (lane edge changes) may be ignored, and overall large-scale information (parking lot layout) is not fully utilized, resulting in suboptimal planning results in complex scenarios. Summary of the Invention

[0007] This invention addresses the problems of poor adaptability to path planning in the complex environment of narrow parking lots, susceptibility to local optima, and rigid planning frameworks. It provides an autonomous valet parking (AVP) planning method based on a scale-adaptive indexing model. The proposed model establishes multi-scale feature associations through dynamic environmental perception, intelligently adapting to different scale planning requirements and optimizing path search efficiency. This invention breaks through the dependence of traditional algorithms on fixed scales, providing a novel solution to the path planning challenge in narrow spaces, and is of great significance for improving the operational efficiency and success rate of AVP systems in complex scenarios.

[0008] According to one aspect of this specification, an autonomous valet parking planning method based on a scale-adaptive indexing model is provided, comprising:

[0009] S1. Convert the acquired parking scene map into a multi-scale grid map;

[0010] S2. Construct a scale-adaptive index model based on a multi-scale grid map, and extract the main channel skeleton path from the multi-scale grid map using the scale-adaptive index model.

[0011] S3. Based on the main channel skeleton path, calculate the number of intersections between each side of the parking space and the multi-scale grid map to obtain the parking edge;

[0012] S4. Calculate the parking position offset and yaw angle based on the parking edge to obtain the geometric parking position attitude;

[0013] S5. Based on the geometric parking position posture and scale adaptive index model, autonomous valet parking planning is carried out.

[0014] Furthermore, a scale-adaptive indexing model is constructed, including:

[0015] Each grid map at each scale is divided into a set of tiles of a specified size;

[0016] Spatial indexes are built based on multi-scale grid maps to retrieve multi-scale spatial data;

[0017] Based on tile sets and multi-scale spatial data, a recursive index model is constructed using recursive downsampling techniques;

[0018] By performing max pooling on the recursive index model, a scale-adaptive index model is obtained.

[0019] Further, the following steps are performed using the scale-adaptive indexing model, including:

[0020] The total cost at different scales is calculated using the following expression:

[0021]

[0022] in, Representing scale The total cost function under the following conditions Representing scale The complexity cost function is as follows. Representing scale The cumulative path length cost function is used to calculate the cumulative path length cost. Representing scale The security cost function is as follows. This indicates the weights of the relative importance of each factor in the corresponding cost function.

[0023] Selecting the optimal scale based on total cost at different scales includes:

[0024]

[0025]

[0026] in, The optimal scale is the one with the lowest cost. Represents the set of all scales. It is the updated scale. This represents the total cost at the current scale.

[0027] Furthermore, the main channel skeleton path is extracted from the multi-scale grid map using a scale-adaptive indexing model, including:

[0028] Calculate the minimum distance from each free voxel in the multi-scale grid map to the nearest obstacle, and combine it with the set feasible region threshold to identify feasible regions in the multi-scale grid map;

[0029] Intersection operations are performed on feasible regions at different scales to obtain the main channel skeleton path;

[0030] For free voxels that are not within the feasible area, eliminate noise and connect the broken corridor sections;

[0031] Redundant and noisy points in feasible areas of multi-scale grid maps are removed, and the extracted main channel skeleton path is optimized and smoothed.

[0032] Further, S3 includes:

[0033] The index of the current parking space is extracted based on a multi-scale grid map, the four corner points of the parking space are obtained, and the four corner points of the parking space are sorted in a clockwise direction.

[0034] Based on the sorted four corner points, calculate the number of intersections between each edge of the parking space and the multi-scale grid map, and find the edge with the fewest intersections as the parking edge.

[0035] Further, S4 includes:

[0036] Calculate the vector perpendicular to the parking edge to determine the offset direction of the parking position;

[0037] Calculate the offset vector of the parking location based on the parking type;

[0038] Based on the offset direction and offset vector of the parking position, and combined with the actual dimensions of the vehicle, the geometric parking position posture is calculated.

[0039] Further, S5 includes:

[0040] Based on the main channel skeleton path, a search is performed on the attitude from the starting point to the geometric parking position;

[0041] Based on the geometric parking position pose and the four corner points of the parking space, a search is performed from the geometric parking position pose to the parking target point;

[0042] Based on the search from the starting point to the geometric parking position and from the geometric parking position to the parking target point, the autonomous valet parking plan is completed.

[0043] Furthermore, the scale-adaptive indexing model is expressed as follows:

[0044]

[0045] in, This represents a scale-adaptive indexing model. It is the set of tiles of the bottom-level grid map. Indicates the first A collection of tiles on a scale.

[0046] According to one aspect of this specification, an autonomous valet parking planning system based on a scale-adaptive indexing model is provided, comprising:

[0047] The multi-scale grid map acquisition module is used to convert the acquired parking scene map into a multi-scale grid map;

[0048] The main channel skeleton path extraction module is used to construct a scale-adaptive index model based on a multi-scale grid map, and extract the main channel skeleton path from the multi-scale grid map through the scale-adaptive index model.

[0049] The parking edge acquisition module is used to calculate the number of intersections between each edge of the parking space and the multi-scale grid map based on the main channel skeleton path, so as to obtain the parking edge.

[0050] The geometric parking position attitude acquisition module is used to calculate the parking position offset and yaw angle based on the parking edge to obtain the geometric parking position attitude.

[0051] The autonomous valet parking planning module is used to plan autonomous valet parking based on a geometric parking position posture and scale adaptive index model.

[0052] According to one aspect of the present invention, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps of the autonomous valet parking planning method based on a scale-adaptive indexing model.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] 1. This invention constructs a scale-adaptive index model through multi-scale feature maps and dynamic cost functions, which significantly improves the efficiency and robustness of the search algorithm in complex environments, thereby achieving faster path planning in large-scale parking lots.

[0055] 2. This invention proposes a multi-scale grid map, which can more accurately model the parking lot environment, reduce planning errors, enhance adaptability to large-scale parking lot scenarios, and provide more reasonable topological information for path generation.

[0056] 3. This invention introduces channel skeleton path search into the field of parking lot path planning. By constructing the topological skeleton of the parking lot, the complex two-dimensional path search problem is simplified into an efficient search of the one-dimensional skeleton graph, which significantly reduces the computational complexity and improves the success rate of path planning, especially in parking lot environments with complex obstacle layouts. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 The flowchart illustrates an autonomous valet parking planning method based on a scale-adaptive indexing model, as provided in this embodiment of the invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Specifically, such as Figure 1 As shown, this embodiment of the invention provides an autonomous valet parking planning method based on a scale-adaptive index model, including: S101, acquiring radar scan data of a parking lot (the radar is installed on the roof of the autonomous vehicle), and converting the parking scene map obtained from the radar scan data into a multi-scale grid map; S102, constructing a scale-adaptive index model based on the multi-scale grid map; S103, extracting the main channel skeleton path based on the scale-adaptive index model, and optimizing and smoothing the skeleton; S104, calculating the number of intersections between each edge of the parking space and the multi-scale grid map based on the optimized main channel skeleton path to obtain the parking edge; S105, calculating the parking position offset and yaw angle based on the parking edge to obtain the geometric parking position attitude; S106, performing autonomous valet parking planning based on the geometric parking position attitude and the scale-adaptive index model.

[0061] Specifically, the embodiments of the present invention provide the specific process for each step:

[0062] S101: Convert the known parking scene map into a grid format, where each grid cell (tile) encodes obstacles using binary values.

[0063] S102: Define the tile size parameter, divide the grid map into a set of tiles of size s×s, each tile representing a sub-region of the map; construct a spatial index to efficiently retrieve multi-dimensional spatial data; use max pooling operation to construct a multi-scale tile set, preserving obstacle information while reducing data dimensionality; select the optimal scale by comprehensively considering computational complexity, path length, and safety factors.

[0064] S103: Calculate the minimum distance from each free voxel to the nearest obstacle, set a feasible region threshold, and identify feasible regions; perform intersection operations on feasible regions at different scales, merge feasible regions, and retain key structural information; for free voxels not in feasible regions, eliminate noise and connect broken corridor parts; remove redundant and noise points, and optimize and smooth the extracted skeleton points.

[0065] S104: Extract the index of the current parking space from the multi-scale grid map, obtain the coordinate set of the four corner points of the parking space, and sort the four corner points of the parking space in a clockwise direction. Calculate the number of intersections between each edge of the parking space and the multi-scale grid map, find the midpoint and vector of the edge with the fewest intersections, and the edge with the fewest intersections is the parking edge.

[0066] S105: Calculate the orthogonal unit vector of the parking edge based on the parking edge, i.e., the vector perpendicular to the parking edge, to determine the offset direction of the parking position; calculate the parking position offset vector according to the parking type (perpendicular parking or parallel parking); calculate the geometric parking position posture based on the offset direction and offset vector of the parking position, combined with the actual size of the vehicle.

[0067] S106: Based on the main channel skeleton path, perform a search for the attitude from the starting point to the geometric parking position; based on the attitude of the geometric parking position and the four corner points of the parking space, perform a search for the parking target point from the attitude of the geometric parking position; based on the search from the starting point to the attitude of the geometric parking position and the search from the attitude of the geometric parking position to the parking target point, complete the autonomous valet parking planning.

[0068] Specifically, in S101, each grid cell is encoded with a binary value to represent an obstacle, as follows:

[0069] (1)

[0070] in, This indicates the state of the cell in the i-th row and j-th column of the map grid, where i and j are the row index and column index of the tile, respectively. A value of 1 indicates that there is an obstacle at the cell location, and a value of 0 indicates that the cell location is an empty area. This indicates that there is an obstacle in the cell at row i and column j in the map grid; This indicates that the cell in the i-th row and j-th column of the map grid is unobstructed.

[0071] Specifically, the embodiments of the present invention provide specific steps in S102, including:

[0072] S1021. Define the tile size parameter to divide the grid map into a set of tiles of size s×s; specifically, define the tile size parameter to divide the grid map into a set of tiles of size s×s.

[0073] (2)

[0074] in, It is the set of tiles of the bottom-level grid map. This represents a sub-region of the bottom-level grid map, where s represents the number of grid cells in each tile, i is the row index of the tile, j is the column index of the tile, and M and N are the original dimensions of the bottom-level grid map, respectively. The union operator U represents merging the states of all grid cells within the grid map, while... This represents the state of a grid cell in a grid diagram, where x and y represent tiles. The coordinates of the grid within the range. , This represents the range of (x, y) coordinates for the bottom-level grid map. This indicates the scale of the lowest-level grid map.

[0075] S1022. Construct a spatial index to efficiently retrieve multidimensional spatial data, including:

[0076] (3)

[0077] Here, Node represents all nodes in the spatial index, containing sets of child nodes at different levels. This represents a node in layer k (i.e., the k-th layer grid map), which contains several child nodes or tiles. This indicates the tile size for the current level. )express The bottom left coordinate of the rectangular region (i,j) in the grid map, and ( This indicates the upper right coordinate of the region.

[0078] S1023. Subsequently, a recursive indexing model is constructed using recursive downsampling technology. This model is built layer by layer starting from the base tile map, with each layer's tile map obtained by performing a max-pooling operation on its lower-level grid data. Max-pooling is used to construct a multi-scale tile set, preserving obstacle information while reducing data dimensionality, including:

[0079] (1) Generate high-level tiles through max pooling operation:

[0080] (4)

[0081] in, This represents the value at position (i,j) in the k-th layer of tiles. and This indicates that the k-th layer tile is in a lower level (the k-th layer tile). Adjacent position index in layer), Indicates the first The value of position (x, y) in the tile layer.

[0082] (2) Construct a multi-scale tile set to retain obstacle information while reducing data dimensionality:

[0083] (5)

[0084] in, Indicates the first A collection of tiles on a scale This represents the tile located at position (i,j) on this scale. and These represent the number of rows and columns in the tile set at this scale, respectively, by dividing the original size by... get.

[0085] (3) Construct a scale-adaptive indexing model:

[0086] (6)

[0087] Here, Pyramid represents the entire scale adaptive indexing model.

[0088] S1024. Considering computational complexity, path length, and security factors, select the optimal metric, including:

[0089] (1) Considering computational complexity, path length, and security factors:

[0090] (7)

[0091] (8)

[0092] (9)

[0093] (10)

[0094] in, Representing scale The total cost function under the following conditions Representing scale The complexity cost function is as follows. Representing scale The cumulative path length cost function is used to calculate the cumulative path length cost. Representing scale The security cost function is as follows. To adjust the weights of the relative importance of each factor in the cost function. For scale The amount of data to be processed This represents the amount of data at the finest scale. Let be the Euclidean distance between consecutive path points. The minimum distance between the path and the nearest obstacle. For predefined security thresholds.

[0095] (2) During the path search process, the system evaluates the cost of each potential scale and selects the scale with the lowest cost as the current search scale. Selecting the optimal scale:

[0096] (11)

[0097] in, The optimal scale, where cost is lowest, is s, which represents the scale of a certain map layer. This represents the set of all scales. The scaling ratio is dynamically adjusted based on real-time performance feedback and environmental changes. If the path quality or safety at the current scaling ratio is substandard, the system will automatically recalibrate to a new optimal scaling ratio.

[0098] (12)

[0099] in, It is the updated scale. This represents the total cost at the current scale. Indicates the current scale.

[0100] Specifically, the embodiments of the present invention provide specific steps in S103, including:

[0101] S1031. Obtain the minimum distance from each free voxel to the nearest obstacle, set a feasible region threshold, and identify feasible regions, including:

[0102] (1) Obtain the minimum distance from each free voxel to the nearest obstacle, and set a feasible region threshold:

[0103] (13)

[0104] in, Let represent the total cost of the target point (i,j), and m and n represent the neighborhood grid indices. Use the weight function to represent This indicates a local adjustment item.

[0105] (2) Identify regions that do not require morphological operations, such as rectangular columns:

[0106] (14)

[0107] in, Indicates the first Feasible regions at various scales This indicates the threshold for the feasible region.

[0108] S1032. Perform intersection operations on feasible regions at different scales, merge feasible regions, and retain key structural information, including:

[0109] (15)

[0110] Where n represents the total number of scales, Representing different scales The feasible area above, This represents the intersection of features across all scales of the grid map.

[0111] S1033. For free voxels not within the feasible area, eliminate noise and connect broken corridor sections, including:

[0112] (16)

[0113] in, This represents the new free voxel value or eigenvalue at position (i,j) after processing. This represents the original value of the input image or feature map at position (i,j). This indicates that the original image is processed first. The corrosion process is performed, followed by an expansion process on the corrosion results.

[0114] S1034. Remove redundant and noise points, and optimize and smooth the extracted skeleton points, including:

[0115] (1) Remove redundant and noisy points:

[0116] (17)

[0117] in, This represents the state value at the current time t. This represents the state value at the next time step t+1. This indicates that a sparsification operation will be performed on the current state value.

[0118] (2) Optimize and smooth the extracted skeleton points:

[0119] (18)

[0120] in, Represented as a skeleton point set, This represents the result calculated using the Euclidean distance formula. Represented as the number of points in the neighborhood. This represents the minimum number of points required within the neighborhood. This represents the distance threshold; if the condition is met... Then point It belongs to the ε neighborhood of gi.

[0121] Specifically, the embodiments of the present invention provide specific steps in S104, including:

[0122] S1041. Extract the index of the current parking space from the gridded multi-scale grid map, obtain the coordinate set of the four corner points of the parking space, and sort the four corner points of the parking space in a clockwise direction.

[0123] S1042. Based on the sorted four corner points, calculate the number of intersections between each edge of the parking space and the multi-scale grid map, and obtain the midpoint and vector of the edge with the fewest intersections, including:

[0124] (19)

[0125] in, , Indicates the corner points of adjacent parking spaces. 、( The coordinates of the two adjacent corner points of the parking space are x and y, respectively. It is a vector.

[0126] S1043, the edge with the fewest intersections is the parking edge of each parking space.

[0127] Specifically, the embodiments of the present invention provide specific steps in S105, including:

[0128] S1051. Based on the parking edge, calculate the orthogonal unit vector of the parking edge, i.e., the vector perpendicular to the parking edge, to determine the offset direction of the parking position, including:

[0129] (20)

[0130] in, For vectors The components on the x-axis, It is a vector Components on the y-axis This represents an orthogonal unit vector.

[0131] S1052. Calculate the parking position offset vector based on the parking type (perpendicular parking or parallel parking), including:

[0132] (twenty one)

[0133] (twenty two)

[0134] in, To adjust the factor, and This represents the perpendicular parking offset vector and the parallel parking offset vector.

[0135] S1053. Based on the offset direction and offset vector of the parking position, and combined with the actual dimensions of the vehicle, calculate the geometric parking position attitude, including:

[0136] (twenty three)

[0137] (twenty four)

[0138] in, This represents the desired phase angle. This represents the phase angle obtained from actual calculation. 、( ) are the x and y coordinates of two adjacent corner points of the parking space.

[0139] S1054. Merge the information to obtain the geometric parking position pose, including:

[0140] (25)

[0141] in, Indicates the geometric parking position. The midpoint of the edge with the fewest intersection points. This is the offset vector.

[0142] Specifically, the embodiments of the present invention provide specific steps in S106, including:

[0143] S1061. Determine the intermediate position pose based on the geometric parking location, serving as a transition point between multi-scale skeleton path search and local multi-scale mesh map search, specifically:

[0144] (26)

[0145] in, This is the intermediate position pose, corresponding to the geometric parking position pose. Nearest neighbor locations on a multi-scale skeleton, where These are the horizontal coordinates, vertical coordinates, yaw angle, and direction of travel of the parking point, respectively. dir=1 indicates forward direction, and dir=-1 indicates reverse direction.

[0146] S1062. Path search from the starting point to the intermediate position is completed based on the multi-scale skeleton;

[0147] S1063. Complete the search from the intermediate position attitude to the parking target point in the local multi-scale grid map.

[0148] The implementation of the various embodiments of the present invention is based on programmed processing through a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide an autonomous valet parking planning system based on a scale-adaptive index model. This system is used to execute an autonomous valet parking planning method based on a scale-adaptive index model from the above method embodiments.

[0149] The system includes: a multi-scale grid map acquisition module, which converts the acquired parking scene map into a multi-scale grid map; a main channel skeleton path extraction module, which constructs a scale-adaptive index model based on the multi-scale grid map and extracts the main channel skeleton path from the multi-scale grid map using the scale-adaptive index model; a parking edge acquisition module, which calculates the number of intersections between each edge of the parking space and the multi-scale grid map based on the main channel skeleton path to obtain the parking edge; a geometric parking position attitude acquisition module, which calculates the parking position offset and yaw angle based on the parking edge to obtain the geometric parking position attitude; and an autonomous valet parking planning module, which performs autonomous valet parking planning based on the geometric parking position attitude and the scale-adaptive index model.

[0150] This invention provides an autonomous valet parking planning system based on a scale-adaptive index model. Addressing the problems of poor adaptability to path planning in the complex environment of narrow parking lots, susceptibility to local optima, and rigid planning frameworks, this system employs several modules and proposes a scale-adaptive index model. This model establishes multi-scale feature associations through dynamic environmental perception, intelligently adapting to different scale planning requirements and optimizing path search efficiency. This invention breaks through the dependence of traditional algorithms on fixed scales, providing a novel solution to the challenge of path planning in narrow spaces, and is of great significance for improving the operational efficiency and success rate of AVP systems in complex scenarios.

[0151] This invention also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, this program breaks through the dependence of traditional algorithms on fixed scales, providing a novel solution to the problem of path planning in narrow spaces, and is of great significance for improving the operational efficiency and success rate of AVP systems in complex scenarios.

[0152] In summary, the present invention acquires radar scan data of a parking lot, converts the parking scene map into a multi-scale grid map, and thus constructs a scale-adaptive index model. The main channel skeleton path is extracted using the constructed scale-adaptive index model, and the number of intersections between each edge of the parking space and the multi-scale grid map is calculated. The parking position offset and yaw angle are calculated respectively to determine the geometric parking position attitude, thereby enabling autonomous valet parking planning. This invention solves the problems of low efficiency and low success rate of existing path planning methods in complex and narrow parking lot environments.

[0153] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and many variations are possible. Any simple modifications, equivalent changes, and alterations made to the above specific embodiments based on the technical essence of the present invention should be considered within the protection scope of the present invention.

Claims

1. An autonomous valet parking planning method based on a scale-adaptive indexing model, characterized in that, include: S1. Convert the acquired parking scene map into a multi-scale grid map; S2. Construct a scale-adaptive index model based on a multi-scale grid map, and extract the main channel skeleton path from the multi-scale grid map using the scale-adaptive index model. Constructing a scale-adaptive indexing model includes: Each grid map at each scale is divided into a set of tiles of a specified size; Spatial indexes are built based on multi-scale grid maps to retrieve multi-scale spatial data; Based on tile sets and multi-scale spatial data, a recursive index model is constructed using recursive downsampling techniques; By performing max pooling on the recursive index model, a scale-adaptive index model is obtained. The following steps are performed using the scale-adaptive indexing model: The total cost at different scales is calculated using the following expression: ， in, Representing scale The total cost function under the following conditions Representing scale The complexity cost function is as follows. Representing scale The cumulative path length cost function is used to calculate the cumulative path length cost. Representing scale The security cost function is as follows. This indicates the weights of the relative importance of each factor in the corresponding cost function. Selecting the optimal scale based on total cost at different scales includes: ， ， in, The optimal scale is the one with the lowest cost. Represents the set of all scales. It is the updated scale. This represents the total cost at the current scale; S3. Based on the main channel skeleton path, calculate the number of intersections between each side of the parking space and the multi-scale grid map to obtain the parking edge; S4. Calculate the parking position offset and yaw angle based on the parking edge to obtain the geometric parking position attitude; S5. Based on the geometric parking position posture and scale adaptive index model, autonomous valet parking planning is carried out.

2. The autonomous valet parking planning method based on a scale-adaptive indexing model according to claim 1, characterized in that, The main channel skeleton path is extracted from the multi-scale grid map using a scale-adaptive indexing model, including: Calculate the minimum distance from each free voxel in the multi-scale grid map to the nearest obstacle, and combine it with the set feasible region threshold to identify feasible regions in the multi-scale grid map; Intersection operations are performed on feasible regions at different scales to obtain the main channel skeleton path; For free voxels that are not within the feasible area, eliminate noise and connect the broken corridor sections; Redundant and noisy points in feasible areas of multi-scale grid maps are removed, and the extracted main channel skeleton path is optimized and smoothed.

3. The autonomous valet parking planning method based on a scale-adaptive indexing model according to claim 1, characterized in that, Step S3 includes: The index of the current parking space is extracted based on a multi-scale grid map, the four corner points of the parking space are obtained, and the four corner points of the parking space are sorted in a clockwise direction. Based on the sorted four corner points, calculate the number of intersections between each edge of the parking space and the multi-scale grid map, and find the edge with the fewest intersections as the parking edge.

4. The autonomous valet parking planning method based on a scale-adaptive indexing model according to claim 1, characterized in that, Step S4 includes: Calculate the vector perpendicular to the parking edge to determine the offset direction of the parking position; Calculate the offset vector of the parking location based on the parking type; Based on the offset direction and offset vector of the parking position, and combined with the actual dimensions of the vehicle, the geometric parking position posture is calculated.

5. The autonomous valet parking planning method based on a scale-adaptive indexing model according to claim 3, characterized in that, Step S5 includes: Based on the main channel skeleton path, a search is performed on the attitude from the starting point to the geometric parking position; Based on the geometric parking position pose and the four corner points of the parking space, a search is performed from the geometric parking position pose to the parking target point; Based on the search from the starting point to the geometric parking position and from the geometric parking position to the parking target point, the autonomous valet parking plan is completed.

6. The autonomous valet parking planning method based on a scale-adaptive indexing model according to claim 1, characterized in that, The scale-adaptive indexing model is expressed as follows: ， in, This represents a scale-adaptive indexing model. It is the set of tiles of the bottom-level grid map. Indicates the first A collection of tiles on a scale.

7. An autonomous valet parking planning system based on a scale-adaptive indexing model, characterized in that, An autonomous valet parking planning method based on a scale-adaptive indexing model as described in any one of claims 1-6 includes: The multi-scale grid map acquisition module is used to convert the acquired parking scene map into a multi-scale grid map; The main channel skeleton path extraction module is used to construct a scale-adaptive index model based on a multi-scale grid map, and extract the main channel skeleton path from the multi-scale grid map through the scale-adaptive index model. The parking edge acquisition module is used to calculate the number of intersections between each edge of the parking space and the multi-scale grid map based on the main channel skeleton path, so as to obtain the parking edge. The geometric parking position attitude acquisition module is used to calculate the parking position offset and yaw angle based on the parking edge to obtain the geometric parking position attitude. The autonomous valet parking planning module is used to plan autonomous valet parking based on a geometric parking position posture and scale adaptive index model.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the autonomous valet parking planning method based on the scale-adaptive index model as described in any one of claims 1 to 6.

Images