A method for dividing a passing boundary and generating a reference line under an unstructured scene

Abstract

The application discloses a kind of unstructured scene under the method for dividing and generating reference line of passing boundary, it is related to the technical field of car intelligence planning, and its technical solution main points are: obtaining discrete path data, pre-processing, obtaining pre-processed data, and calculating the heading and curvature information of reference line;Receive road network information and input A* algorithm, obtain global rough path, convert global rough path from Cartesian coordinate system to Frenet coordinate system, then establish driving space node graph, input driving space node graph into dynamic programming algorithm, obtain left and right rough boundary, realize smoothing involving equality constraint initial solution in combination with quadratic programming model;Establish the kinematic model of vehicle motion of arc length parameter and discretization AL-iLQR solver combined with multi-constraint optimization, generate smooth reference line.The application significantly improves the path modeling efficiency and noise resistance, optimizes the real-time and safety of passing boundary smoothing, and efficiently generates kinematically feasible trajectory.

Description

Technical Field

[0001] This invention relates to the field of intelligent planning technology for automobiles, and more specifically, to a method for defining traffic boundaries and generating reference lines in unstructured scenarios. Background Technology

[0002] In the field of spatiotemporal joint motion planning, the motion state of a vehicle at different times can be described by spatiotemporal motion trajectories. These trajectories can be represented using either a Cartesian coordinate system or a Frenet coordinate system. The Cartesian coordinate system is favored due to its wide applicability; however, in motion trajectory planning, it is also necessary to consider the vehicle's position relative to the road or obstacles, which requires more complex mathematical calculations than simply calculating lateral and longitudinal distances. Therefore, the Frenet coordinate system was introduced, in which the vehicle's position relative to the road or obstacles is used... This is used to represent the longitudinal displacement of the vehicle along the reference line. This represents the lateral displacement of the vehicle relative to the reference line, and the vehicle's trajectory can be considered as... and The Fleury coordinate system is a function that varies with time T. This representation not only simplifies the mathematical description of the trajectory and provides a more concise coordinate representation, but also allows for the accurate depiction of trajectory points at each moment by pre-setting a reference line and ignoring the influence of road curvature changes on the trajectory representation. This facilitates path planning and decision-making in autonomous driving systems. Therefore, the Fleury coordinate system has been widely used in the navigation systems of autonomous vehicles.

[0003] Spatiotemporal joint planning problems require simultaneous consideration of temporal and spatial influences, resulting in high complexity. To reduce the computational cost of joint planning, sufficient constraint information and differentiable representations are needed for the trajectory optimization process.

[0004] In view of this, the present invention proposes a method for dividing the passage boundary and generating reference lines in unstructured scenes. Summary of the Invention

[0005] The above-mentioned technical objective of the present invention is achieved through the following technical solution: The first aspect of this invention provides a method for defining traffic boundaries and generating reference lines in unstructured scenes, comprising the following steps: Acquire discrete path data, perform preprocessing, obtain preprocessed data, and calculate the heading and curvature information of the reference line; The system receives road network information and inputs it into the A* algorithm to obtain a global coarse path. The global coarse path is then converted from the Cartesian coordinate system to the Frenet coordinate system. A driving space node graph is then established and input into a dynamic programming algorithm to obtain left and right coarse boundaries. The left and right coarse boundaries are then smoothed using a quadratic programming model to obtain smooth boundaries. The smoothing process accelerates the solution by setting an initial solution containing equality constraints for the quadratic programming problem. Obtain the front wheel angle, current vehicle speed, and wheelbase, establish a vehicle kinematic model with arc length parameters, and input the front wheel angle, current vehicle speed, wheelbase, and curvature information. Use the discretized AL-iLQR solver combined with multi-constraint optimization to generate a smooth reference line. The multi-constraint optimization includes: obstacle avoidance corridor constraints, which use smooth boundaries to avoid collisions between vehicles and obstacles.

[0006] In conjunction with the first aspect, the present invention is further configured such that the preprocessing is spline curve interpolation and thinning and denoising.

[0007] In conjunction with the first aspect, the present invention is further configured such that: the conversion from Cartesian coordinates to Frenet coordinates is achieved by constructing a reference line model and solving it using Newton's method to determine the projection point.

[0008] In conjunction with the first aspect, the present invention is further configured such that: establishing the driving space node map includes: selecting equally spaced nodes along the reference path. The sampling points are used as the vertical discrete layer, resulting in a series of blue nodes as shown in the figure. The total number of nodes is denoted as For each vertical sampling point, at equal intervals along the horizontal direction... Symmetric sampling One node; Record the state of each node. and connect and Layer nodes. From node ( , ) to node ( , The connection of ) represents the possible motion paths of the vehicle, and the heading angle of the vehicle can be estimated by the difference method: ; In the formula Indicates longitudinal displacement The heading angle at the reference line.

[0009] In conjunction with the first aspect, the present invention is further configured such that the vehicle kinematic model is: ; in, This is the cumulative arc length of the vehicle relative to the starting point of the reference line. This is the lateral offset. For heading angle deviation, The rotation angle of the front wheels. The vehicle's current speed. This refers to the vehicle's wheelbase. Let be the curvature of the reference line projection point.

[0010] In conjunction with the first aspect, the present invention is further configured such that: the method for calculating the heading and curvature information of the reference line is: ; In conjunction with the first aspect, the present invention is further configured such that the multi-constraint optimization also includes: vehicle motion capability constraints.

[0011] A second aspect of the present invention also provides an apparatus / device / system for a method of defining traffic boundaries and generating reference lines in unstructured scenes, 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 above method.

[0012] A third aspect of the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0013] A fourth aspect of the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0014] In summary, the present invention has the following beneficial effects: 1. Significantly Improved Path Modeling Efficiency and Noise Robustness: In Step 1 (Establishing the Frenet Coordinate System), the Douglas-Peucker algorithm for thinning and natural cubic spline interpolation adaptively thins the original dense path points. This significantly reduces the computational cost of spline interpolation while preserving key geometric features (such as curvature abrupt changes) and avoids oscillations caused by noise points. Natural cubic spline interpolation is then used to reconstruct discrete path points, ensuring the reference line is second-order differentiable (continuous curvature), providing a smooth geometric foundation for subsequent motion planning.

[0015] 2. Real-time and Safety Optimization of Passage Boundary Smoothing: In step two (quadratic programming passage boundary smoothing), the design of dynamic programming coarse boundary generation + constrained quadratic programming modeling + hot-start initial solution transforms the discrete boundary generated by dynamic programming into a convex quadratic programming problem. An analytical initial solution based on the fixed width assumption (hot-start) is designed, significantly reducing the number of iterations and meeting real-time requirements. By introducing lateral displacement constraints (ensuring the boundary width is not less than the narrowest point) and directional constraints (avoiding excessive compression of the passage space), obstacle avoidance capability is maintained while smoothing the boundary. The balanced design of the smoothing term (three-point vector curvature penalty) and the deviation term in the objective function ensures that the boundary closely approximates the original feasible region and that the curvature is continuous.

[0016] 3. Efficient generation of kinematically feasible trajectories: The Frenet kinematic model transformation and AL-iLQR solver design in step three (reference line smoothing) convert the time-domain model into a model with the reference line arc length as a parameter, and introduces a curvature change rate control quantity to directly constrain the vehicle's steering rate. The generated trajectory conforms to the actual vehicle kinematic characteristics (such as maximum steering angle limit). By using the augmented Lagrangian method (AL-iLQR), hard constraints such as obstacle avoidance corridor constraints and curvature constraints are transformed into unconstrained optimization problems with penalty terms. Combined with iterative updates of the Lagrange multipliers in the inner and outer loops, the solution efficiency is improved while ensuring safety, which is superior to the numerical instability of the traditional penalty function method. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for dividing access boundaries and generating reference lines in an unstructured scene according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the driving space node diagram in an embodiment of the present invention; Figure 3 This is a flowchart of a method for dividing traffic boundaries and generating reference lines in an unstructured scene according to Embodiment 2 of the present invention. Detailed Implementation

[0018] 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.

[0019] Example 1: A method for boundary delineation and reference line generation in unstructured scenes, such as... Figure 1 As shown, it includes the following steps: Step 1: Establishing the Frenet coordinate system (1) Frenet reference line model The Frenet coordinate system is an intuitive coordinate representation method for planning tasks, and it is better suited to the characteristics of curved roads compared to the Cartesian coordinate system. The Frenet coordinate system uses longitudinal displacement. With lateral displacement This describes the location of any point on the vehicle or reference path. It usually represents the cumulative arc length along the reference line. This indicates the lateral offset relative to the reference line.

[0020] The reference line is represented by discrete path points in the Cartesian coordinate system. A smooth and continuous curve is reconstructed by interpolation of natural cubic spline curves, preserving the geometric features of discrete points and ensuring second-order differentiability.

[0021] When path points are dense or noisy, the Douglas-Peucker algorithm is used to thin out and denoise the path: taking the first and last line segments as the reference, if the maximum deviation of the line segment is less than the threshold, it is simplified to the endpoint; otherwise, it is divided by the point with the maximum deviation and processed recursively, reducing redundant points while retaining key geometric features.

[0022] Using the arc length of the reference line Curve model with parameters It is possible to backtrack the first derivative of the curve with respect to the arc length. Second derivative And thus calculate the heading of the reference line. With curvature Information is used to assist in subsequent coordinate transformation and motion planning.

[0023] (1.1) (1.2) (2) Conversion from Frenet coordinate system to Cartesian coordinate system Given a position vector in Cartesian coordinates To obtain the reference line for this point Representation of Frenet coordinate system projection points This can be viewed as a constrained optimization problem of finding the nearest point on a parametric curve, i.e. (1.3) The solution is obtained iteratively using Newton's method: the objective function is approximated by Taylor expansion, the gradient descent direction is calculated, and the solution is updated iteratively. The longitudinal arc length of the projected point is finally obtained until convergence (the residual is less than the threshold or the maximum number of iterations is reached). and lateral offset .

[0024] Step 2: Smoothing of traffic boundaries based on quadratic programming (1) Generating rough traffic boundaries First, the road network information is received and searched for a global coarse path using the A* algorithm. Based on this, a Frenet coordinate system is created, and then a driving space node graph is established. This mainly involves two steps: first, selecting equally spaced nodes along the reference path... The sampling points are used as the longitudinal discrete layer to obtain, as shown below. Figure 2 A series of blue nodes are shown. The total number of nodes is denoted as Then, for each vertical sampling point, at equal intervals along the horizontal direction... Symmetric sampling Each node, such as Figure 2 The orange nodes are shown.

[0025] Record the state of each node. and connect and Layer nodes. From node ( , ) to node ( , The connection of ) represents the possible motion paths of the vehicle, and the heading angle of the vehicle can be estimated by the difference method: (1.4) In the formula Indicates longitudinal displacement The heading angle at the reference line.

[0026] The driving space node graph is input into a dynamic programming algorithm. The dynamic programming algorithm traverses the node connection relationships, calculates the optimal cost of each node (considering the degree of deviation from the reference line and the distance to obstacles), backtracks to obtain the optimal path sequence, and then determines the left and right rough boundaries (critical nodes before collision).

[0027] (2) Modeling of boundary smoothing problem The rough boundary obtained earlier cannot be used directly and requires further smoothing. To achieve boundary smoothness, both smoothness and fit to the original boundary must be considered. Smoothness is evaluated using a three-point vector relationship: for continuous points... The degree of proximity of the line segment connecting these three consecutive points to the straight line can be described as a vector. , The magnitude of the vector sum is given. A smaller magnitude indicates better smoothness. A smoothness cost is defined to penalize abrupt curvature changes, making the boundary closer to a straight line feature. The smoothness cost formula is as follows: (1.5) Meanwhile, to prevent the smoothed boundary from deviating excessively from the original feasible region, a deviation cost is introduced to ensure that the smoothed boundary is close to the rough boundary. The formula is as follows: (1.6) To ensure traffic safety, two types of constraints need to be set: one is the boundary width constraint. First, ensure the boundary width is not less than the initial narrowest point to avoid compressing the passage space; second, directional constraints. and To prevent the boundary from shifting towards the obstacle. Considering the above factors, the optimization objective is... (where the weights are the coefficients), and transform it into a standard quadratic programming problem to achieve smooth boundary processing and obtain a smooth boundary.

[0028] (3) Initial solution design for boundary smoothing problem Ignoring directional constraints and assuming a fixed boundary width, the problem is transformed into an equality-constrained QP problem. An analytical solution is obtained by solving the block linear system, which serves as the initial solution for the "hot start" of the complete QP problem, reducing the number of iterations.

[0029] Step 3: Smoothing the reference line considering kinematic constraints (1) Vehicle kinematic model in Frenet coordinate system Establish the vehicle kinematic differential equations in the Frenet coordinate system, with the state variables as follows: These represent the cumulative arc length, lateral offset, and heading angle deviation of the vehicle relative to the reference line starting point, respectively; control quantities include the front wheel rotation angle. Current speed of the vehicle In addition, it involves the vehicle wheelbase Curvature of reference line projection points The derivation of the kinematic equations is based on the rate of change of the heading angle deviation. It is equal to the rate of change of the vehicle's own heading angle. Rate of change of heading angle at the projection point relative to the reference path The difference, that is .in, ,and Therefore, the complete kinematic equations are: (1.7) Among these, the front wheel angle and the vehicle's current speed need to be acquired in real time, the vehicle wheelbase is set according to the specific parameters of the vehicle, and the curvature of the projection point is obtained using the method in step one.

[0030] Since the reference line smoothing problem does not involve interaction with dynamic obstacles, in order to eliminate the model's dependence on velocity... The dependency can be utilized The countdown will be in time. The kinematic model with parameters is converted to one with reference line arc length. This is a parametric model. Turning curvature can also be introduced. Eliminating the tangent function improves the model's nonlinear characteristics; defining the actual driving distance instead of the cumulative arc length of the reference line, and replacing the control variable with the rate of change of the path curvature relative to the actual driving distance. Eliminate the influence of the speed term.

[0031] (2) Reference line smoothing performance indicators and constraint design The cost function consists of three parts: reference line deviation cost (expressed as a quadratic form of the state variables, i.e.) This makes the path closer to the original reference line, and the smoothness cost (penalty for path curvature and rate of change of curvature, i.e.) Terminal cost (the degree of deviation between the endpoint and the reference line state, avoiding excessive tail-end offset). Constraints include: obstacle avoidance corridor constraints. ( The path is confined within a safe boundary. The obstacle avoidance corridor constraint uses the previously obtained smooth boundary to prevent collisions between the vehicle and obstacles. Vehicle motion capability constraints (curvature and rate of change of curvature must be within a threshold range, i.e.) , All constraints are rearranged into standard linear form. .

[0032] (3) AL-iLQR solver design After designing the performance indicators and constraints, the AL-iLQR solver was used for solving the problem. This solver uses the traditional iLQR method based on dynamic programming and LQR theory, approximating the initial trajectory using Newton's method, solving the LQR problem through backpropagation, and verifying the optimization direction through forward simulation. However, it cannot handle constraints outside the dynamic equations. The penalty function method transforms constraints into penalty terms in the objective function, but due to limitations in computer precision, it is difficult to strictly satisfy the constraints, easily leading to oscillations of the trajectory near the boundary. An improved approach is adopted using the augmented Lagrange method (AL-iLQR): the core of which is to construct an augmented objective function. in Let Lagrange multiplier vectors be used. The penalty coefficient is adjusted via a dynamic activation matrix, applying penalty only to unmet constraints. A two-level iterative mechanism is employed: the inner loop solves the unconstrained optimization problem and updates the control input. With state ;Outer loop dynamic adjustment and This continues until the termination condition is met. Each stage constructs an augmentation term. Terminal stage construction Ultimately, the original problem is transformed into an optimization problem with state equation constraints, and a smooth reference line that satisfies kinematic constraints is generated efficiently.

[0033] Example 2: A method for defining traffic boundaries and generating reference lines in unstructured scenes includes the following steps: S100. Acquire discrete path data, perform preprocessing, obtain preprocessed data, and calculate the heading and curvature information of the reference line; S200. Receives road network information and inputs it into the A* algorithm to obtain a global coarse path. It then converts the global coarse path from the Cartesian coordinate system to the Frenet coordinate system, establishes a driving space node graph, and inputs the driving space node graph into a dynamic programming algorithm to obtain the left and right coarse boundaries. S300. The left and right rough boundaries are smoothed by combining a quadratic programming model to obtain smooth boundaries; the smoothing process accelerates the solution by setting an initial solution containing equality constraints for the quadratic programming problem. S400. Obtain the front wheel angle, current vehicle speed, and vehicle wheelbase, establish a vehicle kinematic model with arc length parameters, and input the front wheel angle, current vehicle speed, vehicle wheelbase, and curvature information. Use the discretized AL-iLQR solver combined with multi-constraint optimization to generate a smooth reference line. Multi-constraint optimization includes: obstacle avoidance corridor constraints, using smooth boundaries to avoid collisions between vehicles and obstacles.

[0034] In step S100 of this embodiment: preprocessing consists of spline curve interpolation and thinning / denoising.

[0035] In step S100 of this embodiment: the method for calculating the heading and curvature information of the reference line is as follows: (2.1) (2.2) in, The heading is for reference lines. This is curvature information.

[0036] In step S200 of this embodiment: the conversion from Cartesian coordinates to Frenet coordinates is achieved by constructing a reference line model and solving it using Newton's method to determine the projection points.

[0037] In step S200 of this embodiment: establishing the driving space node map includes: selecting nodes with equal spacing along the reference path. The sampling points are used as the vertical discrete layer, resulting in a series of blue nodes as shown in the figure. The total number of nodes is denoted as For each vertical sampling point, at equal intervals along the horizontal direction... Symmetric sampling One node; Record the state of each node. and connect and Layer nodes. From node ( , ) to node ( , The connection of ) represents the possible motion paths of the vehicle, and the heading angle of the vehicle can be estimated by the difference method: (2.3) In the formula Indicates longitudinal displacement The heading angle at the reference line.

[0038] In step S400 of this embodiment: the vehicle kinematic model is: (2.4) in, This is the cumulative arc length of the vehicle relative to the starting point of the reference line. This is the lateral offset. For heading angle deviation, The rotation angle of the front wheels. The vehicle's current speed. This refers to the vehicle's wheelbase. Let be the curvature of the reference line projection point.

[0039] In step S400 of this embodiment: the multi-constraint optimization further includes: vehicle motion capability constraints.

[0040] The present invention also provides an apparatus / device / system for a method of defining traffic boundaries and generating reference lines in unstructured scenarios, 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 above method.

[0041] The present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0042] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0043] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for boundary delineation and reference line generation in unstructured scenes, characterized by: Includes the following steps: Acquire discrete path data, perform preprocessing, obtain preprocessed data, and calculate the heading and curvature information of the reference line; The system receives road network information and inputs it into the A* algorithm to obtain a global coarse path. The global coarse path is then converted from the Cartesian coordinate system to the Frenet coordinate system. A driving space node graph is then established and input into a dynamic programming algorithm to obtain left and right coarse boundaries. The left and right coarse boundaries are then smoothed using a quadratic programming model to obtain smooth boundaries. The smoothing process accelerates the solution by setting an initial solution containing equality constraints for the quadratic programming problem. Obtain the front wheel steering angle, current vehicle speed, and vehicle wheelbase, establish a vehicle kinematic model with arc length parameter, and use the discretized AL-iLQR solver combined with multi-constraint optimization to generate a smooth reference line; The multi-constraint optimization includes: obstacle avoidance corridor constraints, which use smooth boundaries to avoid collisions between vehicles and obstacles.

2. The method for boundary delineation and reference line generation in unstructured scenes according to claim 1, characterized in that: The preprocessing consists of spline curve interpolation and thinning / denoising.

3. The method for boundary delineation and reference line generation in unstructured scenes according to claim 1, characterized in that: The conversion from Cartesian coordinates to Frenet coordinates is achieved by constructing a reference line model and solving it using Newton's method to determine the projection points.

4. The method for boundary delineation and reference line generation in unstructured scenes according to claim 1, characterized in that: The process of establishing the driving space node map includes: selecting nodes at equal intervals along the reference path. The sampling points are used as the vertical discrete layer, resulting in a series of blue nodes as shown in the figure. The total number of nodes is denoted as For each vertical sampling point, at equal intervals along the horizontal direction... Symmetric sampling One node; Record the state of each node. and connect and Layer nodes. From node ( , ) to node ( , The connection of ) represents the possible motion paths of the vehicle, and the heading angle of the vehicle can be estimated by the difference method: ； In the formula Indicates longitudinal displacement The heading angle at the reference line.

5. The method for boundary delineation and reference line generation in unstructured scenes according to claim 1, characterized in that: The vehicle kinematic model is as follows: ； in, This is the cumulative arc length of the vehicle relative to the starting point of the reference line. This is the lateral offset. For heading angle deviation, The rotation angle of the front wheels, The vehicle's current speed, This refers to the vehicle's wheelbase. Curvature of the reference line projection point.

6. The method for boundary delineation and reference line generation in unstructured scenes according to claim 1, characterized in that: The method for calculating the heading and curvature information of the reference line is as follows: ； ； in, The heading is for reference lines. This is curvature information.

7. The method for boundary delineation and reference line generation in unstructured scenes according to claim 1, characterized in that: The multi-constraint optimization also includes: vehicle motion capability constraints.

8. An apparatus / device / system for a method of defining traffic boundaries and generating reference lines in unstructured scenes, 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 1-7.

9. 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 1-7.

10. 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 1-7.

Images