A local reference path generation method for autonomous driving in unknown environments
By generating local reference paths in unknown environments, the accuracy problem of path planning in unknown environments is solved, and higher trajectory planning robustness and accuracy are achieved.
Patent Information
- Application Number
- CN202211249772.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-12
- Publication Date
- 2025-08-12
- Estimated Expiration
- 2042-10-12
AI Technical Summary
The prior art is difficult to generate accurate local reference paths in unknown unstructured road environments, resulting in insufficient robustness and accuracy of path planning.
A local reference path generation method is adopted in an unknown environment, and an accurate local reference path is generated by establishing a global coordinate system, building a local occupancy grid map, obtaining road driving areas and boundary information, and calculating the horizontal and direction deviations.
It effectively reduces the position and curvature errors in the reference path, and improves the robustness and accuracy of local trajectory planning in unknown environments.
Smart Images
Figure CN115683141B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of autonomous driving trajectory planning, and in particular to a method for generating a local reference path for autonomous driving in an unknown environment. Background Art
[0002] Most current autonomous driving trajectory planning algorithms are based on the Frenet coordinate system. The Frenet coordinate system generally uses the centerline of the road as a reference path and uses the tangent vector and normal vector of the reference path to establish a coordinate system. Therefore, constructing an accurate and stable reference path is the basis for path planning for autonomous vehicles.
[0003] Most existing path planning methods target structured roads and are primarily based on known high-precision maps. However, there are few path planning methods for unknown, unstructured road environments. Global path planning in unknown environments is typically constructed offline based on satellite map information. Due to the limited accuracy of satellite maps, the road structure depicted based on satellite maps often deviates from the actual road structure. Furthermore, the perception information has significant uncertainty, making the acquisition of reference paths more difficult. Therefore, generating accurate local reference paths has become an important issue that needs to be addressed in path planning in unknown environments. Summary of the Invention
[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology. To achieve the above purpose, a local reference path generation method for autonomous driving in an unknown environment is adopted to solve the problems raised in the above background technology.
[0005] A method for generating a local reference path for autonomous driving in an unknown environment, comprising the following steps:
[0006] Step S1: Establish a global coordinate system, generate a global path according to the order of the task points, perform smooth fitting processing, obtain the positioning information of the autonomous driving vehicle, and intercept the local path segment as the initial global path segment;
[0007] Step S2: construct a local occupancy grid map of the area around the autonomous driving vehicle, and project obstacles and the initial global path segment around the vehicle's current position to obtain a local obstacle grid map;
[0008] Step S3: Obtain known drivable road areas and boundary information based on the local obstacle grid map to generate a reference path benchmark;
[0009] Step S4: Calculate the lateral deviation and directional deviation between the global path and the drivable area of the road, correct the deviation based on the global path, and generate an accurate local reference path.
[0010] As a further solution of the present invention: the specific steps of step S1 include:
[0011] A global coordinate system is established based on satellite map information. Each pixel in the global coordinate system is converted into a latitude and longitude coordinate point. The drivable road information in the map is depicted, topological relationships are constructed, task points are read, and a global path is generated in the topological map based on the order of task points using the A* algorithm.
[0012] The generated global path is smoothly fitted based on the cubic spline curve algorithm. The cubic spline curve consists of n polynomial curve segments: S i , i = 0, 1, ..., n-1, and satisfy the second-order continuous differentiability at each connection point;
[0013] Among them, each curve segment S i Defined as: S i (x) = a i +b i (xx i )+c i (xx i ) 2 +d i (xx i ) 3 ;
[0014] During the driving process of the autonomous driving vehicle, the vertical projection distance between the vehicle and each point on the global path is calculated, the vehicle is positioned according to the point with the minimum distance, and a path segment within a certain range in front and behind the vehicle is intercepted as the initial global path segment.
[0015] As a further solution of the present invention: the specific steps of step S2 include:
[0016] First, a local occupancy grid is constructed based on the real-time perception range around the autonomous vehicle, and a local coordinate system is established with the vehicle's current position as the center.
[0017] Obtain obstacle information around the vehicle's current location through the autonomous environmental perception system and project the obstacle points onto a grid map;
[0018] According to the relationship between the global coordinate system and the vehicle's local coordinate system, the initial global path segment is mapped to the local grid map through coordinate transformation.
[0019] As a further solution of the present invention: the specific steps of step S3 include:
[0020] The perception system obtains the drivable area information based on the road segmentation algorithm according to the laser point cloud data and expresses the result in the local grid map;
[0021] By randomly extracting discrete reference points within the drivable area of the local grid map and approximating the data based on the RBF network, the generated curve can fit any road shape and meet the constraints of vehicle motion characteristics.
[0022] Random discrete sampling is performed on the traversable areas in the local grid map, and the sampling results are used as sample input. The radial basis function of the Gaussian function and the gradient training method are used for network training. The final network output result is used as the benchmark for generating a reference path.
[0023] As a further solution of the present invention: the specific steps of step S4 include:
[0024] Calculate the lateral offset and angular deviation between the initial global path and the road center baseline. First, select the middle point M(x m ,y m ), calculate the point C(x c ,y c ), calculate the lateral offset: d c =x c -x m , and then according to M(x m ,y m ), C(x c ,y c ) The angle between the tangent lines at the two points is calculated to obtain the angle deviation θ;
[0025] Based on the initial global path segment, the reference path is offset corrected and curvature corrected to generate the final accurate local reference path.
[0026] Compared with the prior art, the present invention has the following technical effects:
[0027] By adopting the above technical solution, it is possible to construct an accurate reference path in an unknown unstructured environment. The initial global path is depicted and constructed in satellite aerial images. The path often deviates from the position and shape of the actual road. The reference path generation method of this solution can effectively reduce the position and curvature errors in the reference path, thereby improving the robustness and accuracy of local trajectory planning in unknown environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings:
[0029] Figure 1 A schematic diagram of the steps of a method for generating a local reference path according to an embodiment disclosed in this application;
[0030] Figure 2 This is a schematic diagram of the overall process of the embodiment disclosed in this application;
[0031] Figure 3 A schematic diagram of a positioning process of a vehicle in a global path according to an embodiment disclosed in this application;
[0032] Figure 4 A schematic diagram of a local grid map centered on a vehicle according to an embodiment disclosed in the present application;
[0033] Figure 5 This is a schematic diagram of the final generation result of the local reference path of the embodiment disclosed in this application. DETAILED DESCRIPTION
[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0035] Please refer to Figure 1 and Figure 2 In an embodiment of the present invention, a method for generating a local reference path for autonomous driving in an unknown environment includes the following steps:
[0036] Step S1: Establish a global coordinate system, generate a global path according to the order of the task points, perform smooth fitting processing, obtain the positioning information of the autonomous driving vehicle, and intercept the local path segment as the initial global path segment;
[0037] In this embodiment, a global coordinate system is established based on satellite map information, and each pixel in the coordinate system is converted into a latitude and longitude coordinate point. The calculation process is as follows:
[0038] Given the image coordinates, the GPS coordinates of the origin are (V lon , V lat ), heading angle θ, the image pixel coordinates of the point A to be converted are (A x , A y );
[0039] The first step is to calculate the horizontal and vertical distance differences Δx and Δy between the current point and the target point in the image coordinate system;
[0040] The second step is to find the actual distance L between the current point and the target point:
[0041]
[0042] The third step is to find the angle between the line connecting the two points and true north:
[0043]
[0044] Step 4: Calculate the horizontal and vertical distances in the GPS coordinate system based on the formulas in steps 2 and 3:
[0045]
[0046] The fifth step is to find the longitude and latitude coordinates of the target point, where R is the radius of the earth:
[0047]
[0048] The drivable road information on the map is mapped to construct a road network and stored as a map file. The road network represents accessible road information, consisting of a collection of waypoints and their connections. Task points are read, including their sequence number, longitude, latitude, altitude, and related attributes. Based on the order of the task points, an A* algorithm is used to search the road network and generate a global path.
[0049] The global path is smoothly fitted based on the cubic spline curve algorithm. The cubic spline curve consists of n polynomial curve segments: S i , i=0,1,…,n-1, and satisfy the second-order continuous differentiability at each connection point. i Defined as: S i (x) = a i +b i (xx i )+c i (xx i ) 2 +d i (xx i ) 3 .
[0050] During the driving process of the autonomous vehicle, based on the vertical projection distance between the vehicle and each point on the global path, the point P with the smallest distance between the vehicle and the global path is selected as the vehicle's positioning point, and the path segments within the range of 100 meters in front of the vehicle (point E) and 50 meters behind the vehicle (point S) are intercepted as the initial global reference path segment, as shown in the figure below. Figure 3 As shown in the figure, it is a schematic diagram of the positioning process of the vehicle position in the global path.
[0051] Step S2: construct a local occupancy grid map of the area around the autonomous driving vehicle, and project obstacles and the initial global path segment around the vehicle's current position to obtain a local obstacle grid map;
[0052] In this embodiment, the specific method is as follows:
[0053] Due to the lack of prior knowledge of the unknown environment, the global path is obtained from satellite aerial images, which often deviates from the position and shape of the actual road. When the deviation is large, the initial reference path needs to be corrected based on the vehicle's real-time environmental perception. First, a local grid map needs to be constructed to express the results of real-time perception. The local occupancy grid map is a grid map of real-time obstacle occupancy information within the current perception range, and the vehicle body position is used as the center to establish a coordinate system, such as Figure 4 The figure shows the final generated result of the local reference path. The grid map is 500×750 pixels in size, corresponding to an area of 100 meters by 150 meters in the real environment. Each cell corresponds to an area of 20 cm by 20 cm in the real environment. The center of the unmanned vehicle is fixed at (250, 500) in the spatiotemporal coordinate system. Secondly, the autonomous environmental perception system obtains obstacle information around the vehicle's current position and projects the obstacle points onto the grid map. Finally, based on the relationship between the latitude and longitude coordinate system and the vehicle's local coordinate system, the initial global path segment is mapped to the local grid map through coordinate transformation.
[0054] Step S3: Obtain known drivable road areas and boundary information based on the local obstacle grid map to generate a reference path benchmark;
[0055] In this embodiment, the drivable area information provided by the perception system refers to the road area in front of the autonomous driving vehicle that can pass without obstacles. It is generally obtained based on the road segmentation algorithm according to the laser point cloud data, and the result is finally expressed in a local grid map.
[0056] By randomly extracting discrete reference points within the drivable area of a grid map, the authors used a regularized network to approximate the reference point data based on an RBF neural network. The network parameters were then selected using a gradient descent method, a single-output RBF network learning method with a forgetting factor. The resulting trajectory can fit any road shape while satisfying the constraints of the vehicle's motion characteristics.
[0057] The activation function of the hidden node of the RBF network selects the radial basis function of the Gaussian function, and the distance function is used as the basis function of the hidden node. In the n-dimensional space, the radial basis function has spatial symmetry about its center point. Radial basis function Φ i (*);
[0058]
[0059] Among them, δ i is the spread constant (Spread), and the screening ability of the basis function decreases as the spread constant increases.
[0060] The output of the RBF network is:
[0061]
[0062] The learning algorithm of the RBF network adopts the gradient learning method, and its objective function is as follows:
[0063]
[0064] where β i is the forgetting factor, the error signal e j Defined as
[0065]
[0066] F(x) for data center c i and weight w i The gradients are:
[0067]
[0068]
[0069] Among them Ф i (x j ) is the i-th hidden node pair X j The output of , η is the learning rate.
[0070] In the specific implementation steps, the gradient training method is selected for network training, the learning rate is 0.001, the target error is 0.05, and the training results can ensure high accuracy and smoothness of the curve.
[0071] Step S4: Calculate the lateral deviation and directional deviation between the global path and the drivable area of the road, and generate an accurate local reference path based on the global path.
[0072] Calculate the lateral offset and angular deviation between the initial global path and the road center baseline. The offset calculation method is as follows: select the middle point M(x m ,y m ), calculate the point C(x c ,y c ), calculate the lateral offset: d c =x c -x m , the angle deviation θ is determined by (x m ,y m ), (x c ,y c ) is obtained by calculating the angle between the tangent directions at the two points.
[0073] The reference path will be generated based on the initial global path by two correction methods, namely offset correction and curvature correction. First, offset correction is performed. According to the calculated lateral offset dc, all points (x i ,y i ) is corrected, and the calculation formula is as follows:
[0074]
[0075] In order to avoid the instability of vehicle control performance caused by frequent corrections, Δs is defined as:
[0076]
[0077] That is, when the offset is small, it is considered that the initial path does not require lateral offset correction.
[0078] Similarly, when the angle deviation is small, the initial path is considered to need no correction; when the angle deviation θ is greater than 10°, correction is performed by adjusting the curvature of the initial path curve. A new cubic polynomial curve is fitted to reconstruct the path by adding control points: the control points include the starting point S(x s ,y s )、End point E(x e ,y e ), the midpoint of the road center baseline M(x m ,y m ) and the vehicle's position V(x v ,y v ), and generate a cubic polynomial curve by interpolating these control points. The curve result is as follows Figure 5 As shown in the figure, it is a schematic diagram of the final generated result of the local reference path, and the curve is output as the final accurate local reference path.
[0079] Although the embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents, and all should be included within the scope of protection of the present invention.
Claims
1. A method for generating a local reference path for autonomous driving in an unknown environment, characterized in that: The specific steps include: Step S1: Establish a global coordinate system, generate a global path according to the order of the task points, perform smooth fitting processing, obtain the positioning information of the autonomous driving vehicle, and intercept the local path segment as the initial global path segment; Step S2: construct a local occupancy grid map of the area around the autonomous driving vehicle, and project obstacles and the initial global path segment around the vehicle's current position to obtain a local obstacle grid map; Step S3: Obtain known drivable road areas and boundary information based on the local obstacle grid map to generate a reference path benchmark. The specific steps include: The perception system obtains the drivable area information based on the road segmentation algorithm according to the laser point cloud data and expresses the result in the local grid map; By randomly extracting discrete reference points within the drivable area of the local grid map and approximating the data based on the RBF network, the generated curve can fit any road shape and meet the constraints of vehicle motion characteristics; Random discrete sampling is performed on the traversable areas in the local grid map. The sampling results are used as sample input. The radial basis function of the Gaussian function and the gradient training method are used for network training. The final network output is used as the benchmark for generating the reference path. Step S4: Calculate the lateral deviation and directional deviation between the global path and the drivable area of the road, correct the deviation based on the global path, and generate an accurate local reference path. The specific steps include: Calculate the lateral offset and angular deviation between the initial global path and the road center baseline. First, select the middle point M(x m , y m ), calculate the point C(x c , y c ), calculate the lateral offset: d c = x c -x m , and then according to M(x m , y m ), C(x c , y c ) The angle between the tangent lines at the two points is calculated to obtain the angle deviation θ; Based on the initial global path segment, the reference path is corrected, including offset correction and curvature correction, to generate the final accurate local reference path.
2. The method for generating a local reference path for autonomous driving in an unknown environment according to claim 1, characterized in that: The specific steps of step S1 include: A global coordinate system is established based on satellite map information. Each pixel in the global coordinate system is converted into a latitude and longitude coordinate point. The drivable road information in the map is depicted, topological relationships are constructed, task points are read, and a global path is generated in the topological map based on the order of task points using the A* algorithm. The generated global path is smoothly fitted based on the cubic spline curve algorithm. The cubic spline curve consists of n polynomial curve segments: , and satisfy the second-order continuous differentiability at each connection point; Each curve segment Defined as: ; During the driving process of the autonomous driving vehicle, the vertical projection distance between the vehicle and each point on the global path is calculated, the vehicle is positioned according to the point with the minimum distance, and a path segment within a certain range in front and behind the vehicle is intercepted as the initial global path segment.
3. The method for generating a local reference path for autonomous driving in an unknown environment according to claim 1, characterized in that: The specific steps of step S2 include: First, a local occupancy grid is constructed based on the real-time perception range around the autonomous vehicle, and a local coordinate system is established with the vehicle's current position as the center. Obtain obstacle information around the vehicle's current location through the autonomous environmental perception system and project the obstacle points onto a grid map; According to the relationship between the global coordinate system and the vehicle's local coordinate system, the initial global path segment is mapped to the local grid map through coordinate transformation.