A toll station diverging area vehicle trajectory planning method and system considering vehicle interaction and a storage medium

Abstract

The present application relates to the technical field of automatic driving, in particular to a toll station shunt area vehicle trajectory planning method and system considering vehicle interaction, and a storage medium, comprising: determining an initial target toll lane based on vehicle information and environmental information of the toll station; determining a reference driving trajectory of an autonomous vehicle in a Cartesian coordinate system and converting to a Frenet coordinate system; generating a lateral trajectory set and a longitudinal trajectory set under a cruising, overtaking and following strategy according to whether there is an obstacle vehicle entering the trajectory planning area, and fitting to obtain all planning trajectories of the autonomous vehicle; using a three-dimensional space-time occupancy graph to eliminate planning trajectories with collision risks, and then selecting an optimal driving trajectory based on a comprehensive evaluation index of the trajectory and vehicle dynamics constraints. The present application can effectively deal with vehicle interaction in complex environments and improve driving safety and efficiency.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and specifically to a method, system, and storage medium for vehicle trajectory planning in toll station diversion areas that takes into account vehicle interaction. Background Technology

[0002] Highway toll stations, due to their complex structure and the lack of clear lane markings in the diversion areas, allow for highly flexible vehicle trajectories, making them a long-standing challenge for traffic management. Particularly in the toll station diversion areas, vehicles need to complete operations such as deceleration, lane selection, queuing, and payment within a short distance, easily leading to short-term peaks during morning and evening rush hours or holidays, resulting in congestion and queuing delays. Different types of vehicles exhibit significant differences in acceleration capabilities, lane-changing intentions, and traffic strategies. Some vehicles may frequently change lanes to choose faster lanes, while others tend to adjust their driving trajectory in advance and maintain a stable queue, thereby reducing weaving and lane-changing operations. This behavioral difference exacerbates the risk of conflicts and accidents in this area. Furthermore, the relatively fixed number and layout of toll lanes, lack of flexibility, and unclear lane markings further increase the complexity of trajectory planning and traffic organization.

[0003] Most current trajectory planning methods focus on conventional road scenarios (such as highways and urban roads), where vehicles typically travel along lane centerlines and exhibit well-defined spatial characteristics. However, in special scenarios such as tollbooth merging areas, vehicle trajectories are not constrained by fixed lane lines, and their movement patterns are influenced by complex interactions with surrounding vehicles and environmental factors. How to perform real-time dynamic trajectory planning in these scenarios, avoiding conflicts with surrounding vehicles and improving driving comfort and safety, remains a pressing issue.

[0004] In summary, there is an urgent need for a vehicle trajectory planning method, system, and storage medium for toll station diversion areas that considers vehicle interaction in order to solve the problems existing in the prior art. Summary of the Invention

[0005] The purpose of this invention is to provide a vehicle trajectory planning method for toll station diversion areas that considers vehicle interaction. It aims to solve the problem of how to perform real-time dynamic trajectory planning for autonomous vehicles within toll station diversion areas, avoiding conflicts with surrounding vehicles and improving driving comfort and safety. The specific technical solution is as follows:

[0006] A vehicle trajectory planning method for toll station diversion areas considering vehicle interaction includes:

[0007] S1. Determine the initial target toll lane based on vehicle and environmental information at the toll station;

[0008] S2. Determine the reference driving trajectory of the autonomous vehicle in the Cartesian coordinate system and transform the Cartesian coordinate system to the Frenet coordinate system;

[0009] S3. Sample along the lateral and longitudinal directions within the trajectory planning area in front of the autonomous vehicle;

[0010] S4, Predicting the Future Whether any obstructing vehicles enter the trajectory planning area at any given time;

[0011] If the future When an unobstructed vehicle enters the trajectory planning area, the autonomous vehicle is set to cruise mode, generating a set of lateral trajectories based on lateral sampling points and a set of longitudinal trajectories based on longitudinal sampling points.

[0012] If the future If an obstacle vehicle enters the trajectory planning area, longitudinal sampling is performed again based on the ST diagram according to overtaking and following strategies. A lateral trajectory set is generated based on the lateral sampling points, and a longitudinal trajectory set is generated based on the resampled longitudinal sampling points. Here, S represents the longitudinal displacement along the reference driving trajectory, and T represents the travel time. This represents any point in time between the current sample and the next sample.

[0013] S5. Based on the fitting of the lateral trajectory set and the longitudinal trajectory set, all planned trajectories of the autonomous vehicle during the period from the current sampling to the next sampling are obtained. Based on the predicted trajectory of the obstacle vehicle, the planned trajectories with collision risk are eliminated using the three-dimensional spatiotemporal occupancy map, and the trajectory selection set composed of optional trajectories is obtained.

[0014] S6. Find the optimal driving trajectory from the current sampling to the next sampling from the trajectory selection set. If the optimal driving trajectory is found, output it directly. If the optimal driving trajectory is not found, return to step S3.

[0015] Preferably, an ST graph is constructed based on the spatiotemporal interaction relationship between the autonomous vehicle and the obstacle vehicle entering the trajectory planning area, and an overtaking sampling area and a following sampling area are divided based on the ST graph. Longitudinal sampling is then performed again in the overtaking sampling area and the following sampling area to obtain the sampling points for the overtaking and following strategies.

[0016] The predicted trajectory based on obstacle vehicles uses a 3D spatiotemporal occupancy map to eliminate planned trajectories with collision risks. Specifically:

[0017] Future The predicted and planned trajectories of vehicles that will enter the trajectory planning area at any time are projected onto the spatiotemporal occupancy map to analyze the occupancy status at different times. The planned trajectories with conflict points are removed, and the planned trajectories without conflict points are added to the trajectory selection set as optional trajectories.

[0018] Preferably, the specific method for finding the optimal driving trajectory from the trajectory selection set is as follows:

[0019] Calculate the comprehensive evaluation index of each selectable trajectory in the optimal trajectory set. Each selectable trajectory is evaluated according to a comprehensive evaluation index. Sort the trajectories from low to high, and check whether each optional trajectory satisfies the vehicle dynamics constraints in turn. The first optional trajectory that satisfies the vehicle dynamics constraints is output as the optimal driving trajectory. If all optional trajectories in the trajectory selection set do not satisfy the vehicle dynamics constraints, return to step S3 to resample and plan the trajectory.

[0020] Preferred, selectable trajectory comprehensive evaluation index Represented as:

[0021]

[0022] in, , , , and All are weighting factors. This represents the ratio of the length of the selectable trajectory to the length of the reference trajectory. This indicates the degree of curvature change in the selectable trajectory. This indicates the degree of deviation of the selected trajectory from the reference trajectory. This represents the ratio of the time corresponding to the output of the optional trajectory to the output time of the reference driving trajectory. This represents the collision cost of the optional trajectory.

[0023] Preferred, optional trajectory collision cost Represented as:

[0024]

[0025] in, Indicates the future Trajectory points of the selectable trajectory at time t. The distance between the trajectory points and the predicted trajectory of the obstacle vehicle. This represents the number of trajectory points into which the optional trajectory is discretized. This indicates the preset maximum horizontal distance;

[0026] The degree of curvature change of the selectable trajectory Represented as:

[0027]

[0028] in, The first of the optional trajectories The curvature at the location of each trajectory point This represents the maximum curvature of the preset trajectory points;

[0029] The degree of deviation of the selectable trajectory from the reference driving trajectory Represented as:

[0030]

[0031] in, The first of the optional trajectories The lateral offset distance between each trajectory point and the reference driving trajectory. This indicates the preset maximum lateral offset distance;

[0032] The ratio of the length of the selectable trajectory to the length of the reference trajectory. Represented as:

[0033]

[0034] in, Indicates the length of the optional trajectory. Indicates the length of the reference driving trajectory;

[0035] The ratio of the time corresponding to the selectable trajectory to the output time of the reference trajectory. Represented as:

[0036]

[0037] in, This indicates the time corresponding to the output of the optional trajectory. This indicates the output time of the reference driving trajectory.

[0038] Preferably, the vehicle dynamics constraint is expressed as:

[0039]

[0040] in, This indicates the minimum permissible acceleration. Indicates the maximum permissible acceleration. This represents the acceleration of any point on the selectable trajectory. Indicates the minimum permitted driving speed. Indicates the maximum permissible driving speed. This represents the speed at any point on the selectable trajectory. This represents the minimum permissible curvature. Indicates the maximum permissible curvature. This represents the curvature of any point on the selectable trajectory.

[0041] Preferably, the predicted trajectory of the obstacle vehicle is obtained as follows:

[0042] A1. Obtain the vehicle Multi-source time-series variables, including vehicles Before the start of trajectory planning Frame trajectory data, vehicles Attribute information and vehicle Environment variables;

[0043] A2. Regarding vehicles forward The trajectory data of the frame is smoothed and filtered to obtain preprocessed trajectory data;

[0044] A3. Using short-time Fourier transform, the vehicle's trajectory in the preprocessed trajectory data is analyzed. shaft speed and the vehicle's direction of travel Angle between axes Perform frequency domain analysis to determine the vehicle Has a lateral offset occurred?

[0045] A4. Vehicle The lateral offset recognition results, preprocessed trajectory data, attribute information, and environmental variables are input together into the iTransformer model for prediction. Then, the prediction results of the iTransformer model are input into the LSTM decoder module to generate the vehicle. The trajectory prediction results.

[0046] Preferably, determine the vehicle Whether lateral offset occurs specifically refers to:

[0047] If both frequency domain feature one and frequency domain feature two show peak values ​​at a certain moment, then the vehicle is considered to be at that moment... Lateral offset occurred; among which, frequency domain feature one is generated by the vehicle along... shaft speed After short-time Fourier transform, the second frequency domain feature is obtained from the vehicle's driving direction and... Angle between axes It is obtained through short-time Fourier transform.

[0048] The present invention also provides a vehicle trajectory planning system for a toll station diversion area that takes into account vehicle interaction, including a memory and a processor. The memory stores a computer program, and the processor executes the trajectory planning method when running the computer program.

[0049] The present invention also provides a storage medium storing a computer program, which, when run, executes the trajectory planning method.

[0050] The application of the technical solution of the present invention has the following beneficial effects:

[0051] This invention addresses the frequent vehicle weaving behavior in toll station diversion areas without lane markings by proposing a vehicle trajectory planning method that considers vehicle interaction. During the autonomous vehicle trajectory planning process, the predicted trajectories of surrounding vehicles are incorporated to fully consider their interaction relationships: when generating lateral and longitudinal trajectory sets, longitudinal sampling is performed above and below the projection area of ​​the predicted trajectory of the obstacle vehicle based on the ST diagram, generating longitudinal sampling points for following and overtaking strategies respectively to avoid longitudinal collisions between the autonomous vehicle and the obstacle vehicle. Simultaneously, a 3D spatiotemporal occupancy map is used to describe the spatial relative positions of the obstacle vehicle and the autonomous vehicle, eliminating planned trajectories with lateral conflict points with the autonomous vehicle, thus selecting optional trajectories with no collision risk with the obstacle vehicle. Finally, the optimal driving trajectory is selected from among the optional trajectories based on comprehensive evaluation indicators and vehicle dynamics constraints, ensuring driving comfort. This method enables autonomous vehicles to perceive potential conflict risks and reasonably avoid them, providing a theoretical reference for trajectory planning of autonomous vehicles in complex traffic scenarios such as those without lane markings.

[0052] The iTransformer-LSTM model of this invention introduces an LSTM decoder module on the basis of the original iTransformer structure, which enhances the ability to model short-time local dynamic features, thereby improving the overall prediction accuracy and stability. Furthermore, the short-time Fourier transform, by converting the vehicle's lateral dynamic signal into a time-spectrum graph, can accurately capture the start and end times and frequency evolution patterns of transient behavior, thereby improving the noise resistance and interpretability of behavior recognition. Using the vehicle's lateral offset behavior extracted by the short-time Fourier transform as the input of the iTransformer-LSTM model, it is possible to detect the frequency domain precursors of dangerous offsets earlier, improving the prediction accuracy and interpretability.

[0053] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0054] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0055] Figure 1 This is a flowchart of the vehicle trajectory planning method for the toll station diversion area considering vehicle interaction in Example 1;

[0056] Figure 2 This is a schematic diagram of the distribution of reference points within the diversion zone in Example 1;

[0057] Figure 3 This is a schematic diagram of converting the planned trajectory of a vehicle based on the Frenet coordinate system into longitudinal distance and lateral offset relative to a reference driving trajectory in Example 1;

[0058] Figure 4 This is the ST diagram describing the spatiotemporal interaction between the autonomous vehicle and the obstacle vehicle in Example 1;

[0059] Figure 5 This is a schematic diagram of the longitudinal trajectory planning of the autonomous vehicle under the overtaking and following strategies in Example 1;

[0060] Figure 6 This is a schematic diagram of all planned trajectories of the autonomous vehicle in the Cartesian coordinate system in Example 1;

[0061] Figure 7 This is a three-dimensional spatiotemporal occupancy map between the planned trajectory and the predicted trajectory of the obstacle vehicle in Example 1;

[0062] Figure 8 This is a schematic diagram of the vehicle distribution at the initial moment of autonomous driving trajectory planning in simulation case 1;

[0063] Figure 9 This is a schematic diagram of trajectory planning for an autonomous vehicle considering interactions within the toll station diversion area in simulation case one.

[0064] Figure 10 This is a schematic diagram of the trajectory planning of an autonomous vehicle in the toll station diversion area without considering the interaction in simulation case 1.

[0065] Figure 11 This is a schematic diagram illustrating the speed changes of an autonomous vehicle under simulation case one, considering and not considering interaction.

[0066] Figure 12 This is a schematic diagram illustrating the acceleration changes of an autonomous vehicle considering and not considering interactions in simulation case one.

[0067] Figure 13 This is a schematic diagram of the trajectory planning of the autonomous vehicle throughout the entire process in simulation case 1;

[0068] Figure 14 This is a flowchart of the background vehicle trajectory prediction process in the diversion zone in Example 2. Detailed Implementation

[0069] To facilitate understanding of the present invention, a more complete description is provided below, along with preferred embodiments. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the present invention.

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0071] Example 1:

[0072] See Figure 1 This embodiment provides a vehicle trajectory planning method for toll station diversion areas that considers vehicle interaction, as detailed below:

[0073] S1. Determine the initial target toll lane based on vehicle and environmental information at the toll station;

[0074] S2. Determine the reference driving trajectory of the autonomous vehicle in the Cartesian coordinate system and transform the Cartesian coordinate system to the Frenet coordinate system;

[0075] S3. Sample along the lateral and longitudinal directions within the trajectory planning area in front of the autonomous vehicle;

[0076] S4, Predicting the Future Whether any obstructing vehicles enter the trajectory planning area at any given time;

[0077] If the future When an unobstructed vehicle enters the trajectory planning area, the autonomous vehicle is set to cruise mode, generating a set of lateral trajectories based on lateral sampling points and a set of longitudinal trajectories based on longitudinal sampling points.

[0078] If the future If an obstacle vehicle enters the trajectory planning area, longitudinal sampling is performed again based on the ST diagram according to overtaking and following strategies. A lateral trajectory set is generated based on the lateral sampling points, and a longitudinal trajectory set is generated based on the resampled longitudinal sampling points. Here, S represents the longitudinal displacement along the reference driving trajectory, and T represents the travel time. This represents any point in time between the current sample and the next sample.

[0079] S5. Based on the fitting of the lateral trajectory set and the longitudinal trajectory set, all planned trajectories of the autonomous vehicle during the period from the current sampling to the next sampling are obtained. Based on the predicted trajectory of the obstacle vehicle, the planned trajectories with collision risk are eliminated using the three-dimensional spatiotemporal occupancy map, and the trajectory selection set composed of optional trajectories is obtained.

[0080] S6. Find the optimal driving trajectory from the current sampling to the next sampling from the trajectory selection set. If the optimal driving trajectory is found, output it directly. If the optimal driving trajectory is not found, return to step S3.

[0081] In this embodiment, the toll station diversion area refers to the transition area where vehicles gradually merge from the main lanes into various toll lanes. The vehicle trajectory planning method in this embodiment will be described in detail below:

[0082] Preferably, before the autonomous vehicle enters the toll plaza diversion area, in order to determine the endpoint of the autonomous vehicle's planned trajectory, it is necessary to select the toll lane with the least delay under the current traffic conditions as the initial target toll lane. This embodiment uses toll lane selection entropy. To achieve the initial target toll lane selection. It is a quantitative indicator of selection behavior based on toll station lane configuration and traffic flow dynamics, depending on the toll station's average service rate, the autonomous vehicle's longitudinal position relative to the toll lane, and the position and number of vehicles ahead of the autonomous vehicle. Autonomous Vehicle Selection The toll lane with the highest value is the initial target toll lane before entering the diversion zone; the toll lane selection entropy is used. Represented as:

[0083]

[0084] in, The longitudinal distance from the autonomous vehicle to the toll lane (unit: meters); For toll lanes Service rate (unit: vehicles / hour); For toll lanes The number of vehicles in the queue (unit: vehicles).

[0085] Preferably, the reference driving trajectory aims to provide a global driving trajectory for autonomous vehicles, and needs to comprehensively consider the geometric features of the scene and road boundary information, including the position of each main lane, the drivable area of ​​the diversion zone, the curvature of the diversion zone curves, the width of the diversion zone, and the location distribution of toll lanes, etc. Specifically, determining the reference driving trajectory of the autonomous vehicle in the Cartesian coordinate system in step S2 is as follows:

[0086] The diversion zone is divided into several sub-regions along the longitudinal direction of the road. Multiple reference points are set at intervals along the transverse direction of the road on the boundary lines between adjacent sub-regions. Reference points are also set at the center of each main lane and each toll lane. Figure 2 As shown, the lateral distance between adjacent reference points in the diversion zone can be set to be the same as the width of a lane;

[0087] Starting from the reference point corresponding to the initial mainline lane where the autonomous vehicle is currently located, and ending at the reference point corresponding to the initial target toll lane, a reference point is selected on each sub-region boundary to construct a set. ;

[0088] For sets The set is obtained by interpolating between any two adjacent reference points. Based on sets The reference driving trajectory is obtained by fitting the trajectory points in the data.

[0089] Furthermore, when selecting reference points on the boundary line between two adjacent sub-regions, the resulting set is required to be... The slope of the line connecting any two adjacent reference points satisfy:

[0090]

[0091] in, Indicates adjacent reference points and The absolute value of the slope of the line connecting them. Indicates lane width, This represents the total number of selected reference points, i.e., the set. The number of reference points in the data.

[0092] Preferably, interpolation between adjacent reference points can solve the problem of only... Having too many reference points can easily lead to an uneven trajectory, resulting in excessive lateral swaying and reduced vehicle comfort and safety. This embodiment specifically increases the density of trajectory points through quadratic spline interpolation, improving the stability of the autonomous vehicle and ensuring driving comfort and safety within the toll station diversion area. The polynomial expression for quadratic spline interpolation is:

[0093]

[0094]

[0095] in, For reference point and Interpolation polynomial between , and All of them are coefficients to be determined. Indicates the first Reference points Substitute coordinate values ​​into the reference point and The function value obtained by the interpolation polynomial between them. Indicates the first Reference points Coordinate values; Indicates the first Reference points Coordinate values Indicates the first Reference points Coordinate values; Indicates reference point and any point between Coordinate values;

[0096] To ensure the continuity and smoothness of the generated curve, the following continuity and boundary conditions must be met:

[0097]

[0098]

[0099]

[0100]

[0101] in, Indicates the first Reference points Substitute coordinate values ​​into the reference point and The function value obtained by the interpolation polynomial between them. express The first derivative; Indicates the first Reference points Substitute coordinate values ​​into the reference point and The function value obtained by the interpolation polynomial between them. express The first derivative; This represents the second derivative value at the reference point at the beginning of the reference driving trajectory; This represents the second derivative value at the reference point at the end of the reference driving trajectory.

[0102] Preferably, the method for generating the reference driving trajectory is not limited to that listed in this embodiment. Those skilled in the art can also use other well-known methods in the field to generate the reference driving trajectory; of course, those skilled in the art can also use other methods to set and select reference points.

[0103] Preferably, the vehicle's motion state changes dynamically according to the driving environment. The Cartesian coordinate system used in traditional trajectory planning methods is difficult to accurately describe the relative positional relationship between the vehicle and the road. In order to simplify the computational complexity of the planning process and accurately express the relative position between the vehicle and the road, the global Cartesian coordinate system needs to be converted to the Frenet coordinate system.

[0104] Figure 3 This describes the transformation of a vehicle's planned trajectory into longitudinal distance and lateral offset relative to a reference trajectory, based on the Frenet coordinate system. Figure 3 middle This indicates the vehicle's current longitudinal displacement. The unit vector representing the normal vector at the current trajectory point. The unit vector representing the tangent line at the current trajectory point. Indicates the projection point on the reference driving trajectory The unit vector of the normal vector at that location. Indicates the projection point on the reference driving trajectory The unit vector of the tangent at that point. Indicates a point on the reference driving trajectory The azimuth at that location, This represents the projection of the current trajectory point of the autonomous vehicle onto the reference driving trajectory. Represents the current vehicle trajectory point and projection point The horizontal distance between them This represents the position vector of the autonomous vehicle in the global coordinate system (i.e., the Cartesian coordinate system). Indicates the azimuth angle of the current trajectory point. Indicates global coordinate system The unit vector of direction. Indicates global coordinate system The unit vector of direction; further, the coordinates of the current operating state of the autonomous vehicle are represented as:

[0105]

[0106] The Cartesian coordinate system typically represents the parameters of a vehicle's motion state as follows: ,in For trajectory points The direction angle, For curvature, For trajectory points acceleration, Represents trajectory points speed, This indicates the coordinates of a point on the reference driving trajectory.

[0107] The Frenet coordinate system uses parameters To express the vehicle's driving status, among which Indicates longitudinal displacement. Let be the first derivative of the vertical axis with respect to time, representing the vehicle's velocity along the reference trajectory. Let be the second derivative of the longitudinal axis with respect to time, representing the acceleration of the vehicle along the reference trajectory. Indicates lateral displacement. Let be the first derivative of the lateral displacement, and represent the velocity perpendicular to the reference trajectory. Let be the second derivative of the lateral displacement, and let represent the acceleration perpendicular to the reference trajectory. This represents the first derivative of the horizontal position with respect to the vertical position. This represents the second derivative of the horizontal position with respect to the vertical position.

[0108] Preferably, in step S2, the transformation from the Cartesian coordinate system to the Frenet coordinate system is expressed as follows:

[0109]

[0110] in, This indicates the vehicle's current longitudinal displacement. This represents the difference between the azimuth angle of the current trajectory point and the azimuth angle of the reference trajectory point. Represents the curvature of the reference trajectory point. This represents the curvature of the current vehicle trajectory point. The derivative of the curvature of the reference trajectory point. Indicates the current trajectory point of the vehicle coordinate, Represents the reference trajectory point coordinate, Indicates the azimuth angle of the reference trajectory point. Indicates the azimuth angle of the current trajectory point. This represents the rate of change of the azimuth angle of the reference trajectory point. This represents the rate of change of the azimuth angle of the current trajectory point. Indicates the current trajectory point of the vehicle coordinate, Represents the reference trajectory point coordinate, Indicates the vehicle's current speed. This represents the distance between the vehicle's trajectory point and the reference trajectory point. This indicates the vehicle's current acceleration.

[0111] Preferably, based on the vehicle kinematics model and the Frenet coordinate system, the trajectory planning of an autonomous vehicle can be decomposed into lateral and longitudinal trajectory planning. Therefore, it is necessary to sample along the lateral and longitudinal directions within the trajectory planning area in front of the autonomous vehicle. In this embodiment, the time interval between two adjacent samples is... The planned trajectory generated from the current sampling is used to guide autonomous vehicles in the future. How to travel within a certain time period, by every The system plans a trajectory over time, ensuring the autonomous vehicle eventually enters the toll lane. Typically 2 to 3 seconds, if the sampling time interval If the time is too long, it will not be able to meet the real-time response requirements for changes in the surrounding environment.

[0112] Preferably, by laterally offsetting on both sides of the reference driving trajectory The lateral sampling range on both sides of the reference driving trajectory is described by the meter, and the lateral sampling range is represented as... The interval between horizontal sampling points is set to Meters. In this embodiment, the lateral sampling range is set to an area offset laterally by 5 meters on each side of the reference driving trajectory, with a sampling interval of 1 meter.

[0113] Preferably, the range of longitudinal sampling depends on How far can an autonomous vehicle travel longitudinally within a given time period? What is the interval between longitudinal sampling points within the longitudinal sampling range? Meters. This can be based on the current speed of the autonomous vehicle and the sampling time interval. The longitudinal sampling range can also be determined based on the average speed of vehicles within the diversion zone and the sampling time interval. To determine the longitudinal sampling range, or other methods may be used by those skilled in the art to determine the longitudinal sampling range.

[0114] Preferably, the trajectory planning area refers to the area covered by the horizontal sampling range and the vertical sampling range. The sampling points within the trajectory planning area are distributed in a matrix, and the horizontal interval between the sampling points is [missing information]. Meters, with longitudinal intervals of rice.

[0115] As mentioned above, trajectory planning for autonomous vehicles can be decomposed into lateral and longitudinal trajectory planning, respectively. Therefore, it is necessary to construct polynomial equations for the lateral and longitudinal trajectories separately to fit the lateral and longitudinal trajectories. In the trajectory generation algorithm, the coefficients of the polynomial equations are determined by satisfying the position, velocity, and acceleration constraints at the previous and current sampling times.

[0116] Let the longitudinal state of the autonomous vehicle at the time of the last sampling be... The longitudinal state of the autonomous vehicle at the time of sampling is Then the polynomial equation of the longitudinal trajectory is expressed as:

[0117]

[0118] in, Indicates the longitudinal trajectory. and These are all coefficients of the longitudinal trajectory polynomial equation. Indicates time; This indicates the longitudinal displacement of the autonomous vehicle during the last sampling. express The first derivative with respect to time, express The second derivative with respect to time, This indicates the longitudinal displacement of the autonomous vehicle at the time of sampling. express The first derivative with respect to time, express The second derivative with respect to time;

[0119] Differentiating the polynomial equation of the longitudinal trajectory, we get:

[0120]

[0121]

[0122] in, for The first derivative, for The second derivative;

[0123] Therefore, the longitudinal trajectory planning solution can be represented by a matrix as follows:

[0124]

[0125] Vertical state and Substituting the matrix into the longitudinal trajectory planning solution, we get:

[0126]

[0127]

[0128] in, Representation matrix The inverse matrix, Representation matrix .

[0129] Let the lateral state of the autonomous vehicle at the time of the last sampling be... The lateral state of the autonomous vehicle at the time of sampling is Then the polynomial equation of the transverse trajectory is expressed as:

[0130]

[0131] in, Indicates the horizontal trajectory. and These are all coefficients of the transverse trajectory polynomial equation. Indicates time; This indicates the lateral displacement of the autonomous vehicle at the time of the last sampling. express The first derivative with respect to time, express The second derivative with respect to time, This indicates the lateral displacement of the autonomous vehicle at the time of sampling. express The first derivative with respect to time, express The second derivative with respect to time;

[0132] Differentiating the polynomial equation of the transverse trajectory, we get:

[0133]

[0134]

[0135] in, express The first derivative, express The second derivative;

[0136] Similarly, the solution for lateral trajectory planning can be represented by a matrix as follows:

[0137]

[0138] Horizontal state and Substituting the matrix into the lateral trajectory planning solution, we can obtain:

[0139]

[0140]

[0141] in, Representation matrix The inverse matrix, Representation matrix .

[0142] By determining the lateral trajectory polynomial equation and the lateral sampling points, the set of lateral trajectories of the autonomous vehicle from the current sampling time to the next sampling time can be calculated (i.e., the next sampling time interval is obtained). The set of lateral trajectories in the data (i.e., the set of longitudinal trajectories from the current sampling time to the next sampling time) can be calculated using the determined longitudinal trajectory polynomial equation and longitudinal sampling points. (The set of longitudinal trajectories in the data).

[0143] Because the toll plaza diversion area is a widened, gradually transitioning zone without lane markings, autonomous vehicles need to ensure continuity of trajectory, speed, and acceleration, and generate a stable and smooth trajectory curve during safe entry into the toll lane. To achieve this, the autonomous vehicle's driving state needs to switch between cruise, following, or overtaking modes depending on the presence of obstructing vehicles ahead. Specifically, based on the predicted future trajectories of background vehicles within the diversion area, the system can determine the background vehicles' positions in the next sampling time interval. Will the vehicle enter the trajectory area at some point in the future and become an obstacle, affecting the safe operation of autonomous vehicles?

[0144] when the future When an unobstructed vehicle enters the trajectory planning area, the autonomous vehicle is set to cruise mode. Based on the lateral sampling points and the lateral trajectory polynomial equation, the autonomous vehicle's trajectory in the next sampling time interval can be generated. The set of lateral trajectories within the range, based on longitudinal sampling points and the longitudinal trajectory polynomial equation, can generate the autonomous vehicle's trajectory for the next sampling time interval. The set of longitudinal trajectories within, where This represents any point in time between the current sampling and the next sampling.

[0145] Furthermore, in cruise mode, since it is not necessary to know the state of the autonomous vehicle at the current sampling moment, the longitudinal trajectory set can also be generated using a fourth-order polynomial, which is expressed as:

[0146]

[0147] in, Indicates the longitudinal trajectory. All are coefficients of a fourth-degree polynomial. Indicates time.

[0148] when the future If an obstacle vehicle enters the trajectory planning area at any time, it indicates that there is a risk of collision between the obstacle vehicle and the autonomous vehicle in the longitudinal direction. An ST diagram should be constructed based on the spatiotemporal interaction relationship between the autonomous vehicle and the obstacle vehicle entering the trajectory planning area. Based on the ST diagram, an overtaking sampling area and a following sampling area should be divided. Longitudinal sampling should be carried out again in the overtaking sampling area and the following sampling area to obtain the sampling points of the overtaking and following strategies.

[0149] like Figure 4 As shown, the ST diagram is used to describe the spatiotemporal interaction between autonomous vehicles and obstacle vehicles, where Represents the longitudinal displacement along the reference driving trajectory. Represents travel time. When the obstacle vehicle is in the future... to The longitudinal section occupying the trajectory planning area of ​​the autonomous vehicle within the time period [ When the obstacle is in motion, its movement will be mapped onto a rectangular projection area in the ST diagram. The height of this area represents the length of the obstacle vehicle, with the upper boundary being the front and the lower boundary being the rear. Based on the obstacle vehicle's projection area... endpoints and The ST map is divided into overtaking sampling areas. and vehicle sampling area The accompanying sampling area The upper boundary is The lower boundary is the starting point for trajectory planning of autonomous vehicles; overtaking sampling area. The lower boundary is defined by the endpoints. Subtract a safety interval The distance is used to determine the trajectory, and the upper boundary is the endpoint of the trajectory planning. Meanwhile, overtaking sampling area and vehicle sampling area The distance between the sampling points and the projected boundary of the obstacle vehicle is greater than or equal to the safety interval. , Figure 4 The red dots in the diagram are the sampling points.

[0150] Furthermore, Figure 5 The diagram illustrates the longitudinal trajectory planning of autonomous vehicles under overtaking and following strategies. Green trajectory lines indicate that the autonomous vehicle has adopted an overtaking strategy in the longitudinal trajectory, while red trajectory lines indicate that the autonomous vehicle has adopted a following strategy in the longitudinal trajectory.

[0151] like Figure 6 As shown, by combining each horizontal trajectory point in the horizontal trajectory set with each vertical trajectory point in the vertical trajectory set, all planned trajectories of the autonomous vehicle in the Cartesian coordinate system can be generated, as follows:

[0152] Calculating the Future Lateral offset at time With longitudinal offset ,in This represents any point in time between the current sampling and the next sampling (i.e., the next sampling time interval). (any time in the process)

[0153] Using the transformation formula from Frenet coordinates to Cartesian coordinates, the time points are... Corresponding trajectory points Transformed into a trajectory point in Cartesian coordinate system ;in, express The position coordinates of the autonomous vehicle in the Cartesian coordinate system at any given time;

[0154] Each time point between the current sample and the next sample Corresponding trajectory points Connect them sequentially according to time to form a complete planned trajectory.

[0155] Preferably, to reduce the dimensionality of trajectory planning and improve the real-time performance of the algorithm, autonomous vehicles typically use a two-dimensional spatial occupancy map to describe the spatial relative positional relationship between the autonomous vehicle and obstacle vehicles within the diversion zone. However, traditional two-dimensional spatial occupancy maps are insufficient to fully reflect the real-time interaction between autonomous vehicles and other obstacle vehicles, and cannot accurately capture the collision risk of autonomous vehicles. Therefore, this embodiment introduces a spatiotemporal occupancy grid, which introduces a time dimension to the traditional two-dimensional spatial occupancy map. By integrating the planned trajectory of the autonomous vehicle with the predicted trajectory of the obstacle vehicle, the occupancy status at different times is constructed, thereby eliminating planned trajectories with lateral collision risks.

[0156] Furthermore, based on the predicted trajectory of the obstacle vehicle, a three-dimensional spatiotemporal occupancy map is used to eliminate planned trajectories with collision risks. Specifically:

[0157] Future The predicted and planned trajectories of vehicles that will enter the trajectory planning area at any time are projected onto the spatiotemporal occupancy map to analyze the occupancy status at different times. The planned trajectories with conflict points are removed, and the planned trajectories without conflict points are added to the trajectory selection set as optional trajectories.

[0158] like Figure 7 As shown, Figure 7The red trajectory line represents the predicted trajectory of an obstacle vehicle that will enter the trajectory planning area in the future. The black trajectory line represents the planned trajectory that has a conflict point with the predicted trajectory of the obstacle vehicle. The green trajectory line represents the optional trajectory that does not have a conflict point with the predicted trajectory of the obstacle vehicle. In the figure, the X-axis represents the longitudinal distance (unit: meters), the Y-axis represents the lateral distance (unit: meters), and the Z-axis represents the time (unit: seconds). The three-dimensional spatiotemporal occupancy map can effectively identify the planned trajectory with collision risk.

[0159] Furthermore, although a set of optimal trajectories has been obtained, different optional trajectories exhibit varying performance in terms of safety, comfort, and traffic efficiency. Therefore, this embodiment constructs a comprehensive evaluation index to assess the quality of each trajectory, and this comprehensive evaluation index is expressed as follows:

[0160]

[0161] in, , , , and All are weighting factors. This represents the ratio of the length of the selectable trajectory to the length of the reference trajectory. This indicates the degree of curvature change in the selectable trajectory. This indicates the degree of deviation of the selected trajectory from the reference trajectory. This represents the ratio of the time corresponding to the output of the optional trajectory to the output time of the reference driving trajectory. This represents the collision cost of the optional trajectory.

[0162] Furthermore, to avoid evaluation indicators , , , and The large differences in values ​​between them cause the optimal driving trajectory solution to converge slowly, requiring further adjustments. , , , and Normalization is performed to limit the value range of each evaluation index to [0,1].

[0163] Furthermore, the collision cost of optional trajectories This represents the spatiotemporal collision risk between the available trajectory and the predicted trajectory of the obstacle vehicle. Based on the spatiotemporal occupancy map, the future collision risk between each trajectory point of the available trajectory and the predicted trajectory of the obstacle vehicle is calculated. The distance at time points is used as the collision cost for that trajectory point. The sum of the collision costs of all trajectory points on the selectable trajectory is obtained. :

[0164]

[0165] in, Indicates the future Trajectory points of the selectable trajectory at time t. The distance between the trajectory points and the predicted trajectory of the obstacle vehicle. This represents the number of trajectory points into which the optional trajectory is discretized. This indicates the preset maximum horizontal distance.

[0166] Furthermore, the degree of curvature change of the selectable trajectory This is used to measure the smoothness of the available trajectory. If the vehicle's trajectory is not smooth, it will cause unnecessary lateral deviation, leading to phenomena such as bumps and tilting. The smoothness of the available trajectory is related to curvature; the smaller the curvature value, the slower the trajectory change, and the higher the ride comfort and safety. The square of the ratio of the curvature at discrete points of the available trajectory to the maximum curvature is used as the curvature change cost. :

[0167]

[0168] in, The first of the optional trajectories The curvature at the location of each trajectory point This represents the maximum curvature of the preset trajectory points. This indicates the number of trajectory points into which the optional trajectory is discretized.

[0169] During vehicle operation, to prevent the selectable trajectory from deviating from the reference trajectory in the Cartesian coordinate system and thus affecting the vehicle's driving efficiency, the trajectory of the autonomous vehicle should closely match the reference trajectory. The degree of deviation of the selectable trajectory from the reference trajectory... Represented as:

[0170]

[0171] in, The first of the optional trajectories The lateral offset distance between each trajectory point and the reference driving trajectory. This indicates the preset maximum lateral offset distance. This indicates the number of trajectory points into which the optional trajectory is discretized.

[0172] The ratio of the length of the selectable trajectory to the length of the reference trajectory. Represented as:

[0173]

[0174] in, Indicates the length of the optional trajectory. Indicates the length of the reference driving trajectory;

[0175] The ratio of the time corresponding to the selectable trajectory to the output time of the reference trajectory. Represented as:

[0176]

[0177] in, This indicates the time corresponding to the output of the optional trajectory. This indicates the output time of the reference driving trajectory.

[0178] Furthermore, after completing the trajectory quality assessment of the vehicle's selectable trajectories, it is also necessary to ensure that the vehicle's motion trajectory conforms to vehicle dynamics constraints and handling limitations. Therefore, it is necessary to further screen the selectable trajectories in the trajectory optimization set, comprehensively considering vehicle dynamics constraints such as vehicle acceleration and curvature, to select the optimal driving trajectory for the autonomous vehicle. In this embodiment, the vehicle dynamics constraints are expressed as follows:

[0179]

[0180] in, This indicates the minimum permissible acceleration. Indicates the maximum permissible acceleration. This represents the acceleration of any point on the selectable trajectory. Indicates the minimum permitted driving speed. Indicates the maximum permissible driving speed. This represents the speed at any point on the selectable trajectory. This represents the minimum permissible curvature. Indicates the maximum permissible curvature. This represents the curvature of any point on the selectable trajectory.

[0181] Therefore, in this embodiment, the specific method for finding the optimal driving trajectory from the trajectory selection set is as follows:

[0182] Calculate the comprehensive evaluation index of each selectable trajectory in the optimal trajectory set. , according to each optional trajectory Sort from lowest to highest, according to the ranking (i.e., according to the comprehensive evaluation indicators). (In ascending order) check whether each optional trajectory meets the vehicle dynamics constraints, and output the first optional trajectory that meets the vehicle dynamics constraints as the optimal driving trajectory; if all optional trajectories in the trajectory selection set do not meet the vehicle dynamics constraints, return to step S3 to resample and plan the trajectory.

[0183] It should be noted that when returning to step S3 for resampling, there is no need to reselect the target toll lane. Ultimately, the toll lane from which the autonomous vehicle exits depends on the optimal driving trajectory selected in the trajectory planning before entering the toll lane. That is, the toll lane from which the optimal driving trajectory output by the last trajectory planning enters is the toll lane from which the vehicle exits.

[0184] Simulation Case 1:

[0185] To verify the effectiveness of the trajectory planning method considering vehicle interactions in this embodiment, a non-interactive trajectory planning method was constructed as a control for analysis. This method, also based on a trajectory sampling strategy in the Frenet coordinate system, generates a trajectory cluster that satisfies vehicle kinematic constraints according to a preset target speed and reference trajectory. In the non-interactive trajectory planning method, the vehicle lacks the ability to predict the future trajectories of surrounding vehicles, cannot perceive the positional changes and dynamic evolution of obstacle vehicles, and trajectory planning relies entirely on the target state and reference trajectory, lacking a response mechanism for other traffic participants in the environment.

[0186] In the constructed tollbooth diversion area scenario, the number of main lanes is set to three, from top to bottom as the upper lane, middle lane, and lower lane. The toll lanes are set to eight, numbered 1 to 8 from bottom to top, with lanes 1 to 5 being ETC lanes and lanes 6 to 8 being MTC lanes. A frame from the tollbooth scenario is selected as the initial starting point for autonomous driving trajectory planning, and the vehicle distribution is as follows: Figure 8 As shown.

[0187] like Figure 8 In the scenario shown, the black vehicles represent autonomous vehicles, the blue vehicles represent other road users, and the orange dashed line represents the trajectory prediction result for the blue vehicles. At the start of the experiment, the autonomous vehicles... The initial target toll lane for vehicles before entering the toll station diversion area was selected as lane 2, and a reference trajectory (green dashed line) was generated based on global trajectory planning.

[0188] Figure 9 This demonstrates the trajectory planning results of an autonomous vehicle considering interactions within a tollbooth diversion zone. At this point, the vehicle's environmental perception system successfully identifies the vehicle V1, which is a decelerating obstacle ahead, and generates a set of candidate trajectories that meet the constraints based on the trajectory planning method of this embodiment and the current environment. The green curve represents multiple selectable trajectories, while the red curve represents the optimal driving trajectory that is ultimately selected and executed, indicating that the vehicle adopted a following strategy.

[0189] In comparison, Figure 10The image shows the trajectory planning result without considering interactions. Because the vehicle failed to recognize and respond promptly to the deceleration behavior of the vehicle in front (V1) during the trajectory planning process, the final generated trajectory overlapped with the actual trajectory of V1, resulting in a collision.

[0190] See Figure 11 and Figure 12 The trajectory planning method in this embodiment predicts the deceleration behavior of the vehicle in front, ensuring that the vehicle maintains a safe longitudinal distance and reduces lateral distance, achieving stable following. In contrast, the non-interactive trajectory planning method does not perceive the behavior of the vehicle in front during the planning process; the vehicle only adjusts its speed based on the target speed. Although it initially reduces its speed to 10 m / s, due to the lack of a response mechanism to the dynamics of the vehicle in front, a collision occurs at approximately 11 seconds. The acceleration curve further shows that the vehicle experiences significant deceleration in the first 2 seconds, which then stabilizes. However, without considering the interaction, the acceleration exhibits a sudden change in the later stages.

[0191] Therefore, in complex and dynamic scenarios such as toll station diversion areas, trajectory planning methods must fully model the interaction behavior between vehicles in order to effectively improve the safety and feasibility of the planning. Figure 13 Based on the trajectory planning results of the vehicle throughout the entire process, it achieves stable following of the vehicle in front through longitudinal deceleration control, and does not change the target toll lane.

[0192] Example 2:

[0193] As shown in Example 1, the predicted trajectories of background vehicles within the diversion zone can be used to determine the background vehicles in the next sampling time interval. Will the vehicle enter the trajectory area and become an obstacle at some point in the future? For example... Figure 14 As shown, this embodiment provides a method for obtaining the predicted trajectory of an obstacle vehicle (i.e., a background vehicle) within a diversion zone, including:

[0194] A1. Obtain the vehicle Multi-source time-series variables, including vehicles Before the start of trajectory planning Frame trajectory data, vehicles Attribute information and vehicle Environment variables;

[0195] Preferably, the trajectory data is used to describe the movement state of the vehicle within the diversion zone, the vehicle The trajectory data includes vehicles forward Frame trajectory points ,along shaft speed ,along shaft speed and the direction of vehicle travel Angle between axes .

[0196] Preferably, the attribute information is for vehicles. Defined static inputs, including vehicles The initial mainline lane it belongs to Vehicle type and charging type Among them, vehicle type Including trucks and passenger cars, the toll types This includes ETC toll collection and MTC toll collection (i.e., whether vehicles pay via ETC or manually), the initial mainline lanes. Only vehicles participating in the prediction The trajectory of vehicles immediately upon entering the diversion zone is not included in the vehicle prediction. The trajectory after the diversion zone.

[0197] Preferably, the environmental variables are used to describe the vehicle. Surrounding dynamic traffic characteristics, including , , and ,in, Indicates vehicle Its left anterior area of ​​influence The minimum Euclidean distance between vehicles. Indicates vehicle Its right anterior area of ​​influence The minimum Euclidean distance between vehicles. Indicates vehicle Its left and rear influence area The minimum Euclidean distance between vehicles. Indicates vehicle Its right rear area of ​​influence The minimum Euclidean distance between vehicles.

[0198] Specifically, the left anterior affected area Right anterior affected area Left rear affected area and the right rear affected area The division method is as follows: construct the dividing line. and Vehicle The area surrounding it is divided into four regions, with a rectangular region set up in each region as the area of ​​influence; wherein, one length-direction side of each rectangular region is the dividing line. Overlapping, a side in the width direction and the dividing line Overlap, dividing line Parallel to the vehicle The direction of travel, the dividing line Perpendicular to the vehicle The direction of travel, and Intersection and vehicle Their centers of mass coincide.

[0199] Furthermore, the size of the rectangular area can be flexibly set by those skilled in the art; in this embodiment, the rectangular area is 30m × 5m.

[0200] A2. Regarding vehicles forward The trajectory data of the frame is smoothed and filtered to obtain preprocessed trajectory data;

[0201] Since trajectory data acquisition may be affected by various factors, in order to effectively reduce errors caused by noise, outliers or data instability, this embodiment uses a Savitzky-Golay filter to smooth the trajectory data.

[0202] Specifically, the Savitzky-Golay filter, also known as the convolutional smoothing filter, is a commonly used smoothing filtering technique in signal processing. Its core idea is to minimize noise interference while preserving key signal features (such as peak and slope variations) as much as possible. Unlike traditional moving average filters that use simple constant weighting coefficients, the Savitzky-Golay filter fits a polynomial of a given order to a sliding window of fixed length using the least squares method, and uses the fitted result to smooth the data at the center point. This method effectively removes high-frequency noise while maintaining the signal's frequency characteristics and local variations, making it particularly suitable for this embodiment where precise preservation of the local motion features of the vehicle trajectory is required. Its mathematical expression is shown below:

[0203]

[0204] in, This refers to the index of the data point (i.e., the trajectory point) within the sliding window. For the fitted data point values, For order, These are the polynomial coefficients.

[0205] The sum of squared residuals between the fitted data points and the original data points is expressed as:

[0206]

[0207] in, The number of points on one side that need to be fitted. For the original data points, express direction or The trajectory points of the direction.

[0208] To obtain the best fit, the sum of squared residuals must be minimized when using the least squares method. Regarding coefficients Taking the partial derivative and setting it to zero, we get:

[0209]

[0210] Find:

[0211]

[0212] in, Indicates when the coefficient When taking partial derivatives, In the formula The coefficients in.

[0213] Coefficient of determination Then obtained The expression, through The expression estimates the center point value within the window, and the expression for calculating the fitted output value through convolution is:

[0214]

[0215] in, The value of the fitted polynomial, These are the smoothing coefficients obtained by fitting the polynomial using the least squares method. This indicates that normalization is being performed. This indicates the index of the data point within the sliding window.

[0216] For the remaining data points, the above operations can be repeated by continuously moving the window to perform continuous fitting analysis on the remaining data points.

[0217] A3. Using short-time Fourier transform (STFT), the vehicle's trajectory in the preprocessed trajectory data is analyzed. shaft speed and the vehicle's direction of travel Angle between axes Perform frequency domain analysis to determine the vehicle Has a lateral offset occurred?

[0218] Furthermore, the short-time Fourier transform is a joint time-frequency analysis method for non-stationary signals. Its core idea is to use a fixed-length sliding window to segment the time-domain signal and perform a Fourier transform on the signal in each segment to obtain the local spectrum of that segment. The calculation formula is as follows:

[0219]

[0220] in, The original time-domain signal represents the vehicle's path along... shaft speed or included angle , Represents the window function. for Conjugate, The imaginary unit, Indicates frequency, Represents time-domain signal of Transformation, Indicates the center position of the window function. Pi Indicates time, Represents the natural constant.

[0221] In the Short Time Fourier Transform (STFT), the window function can effectively suppress spectral leakage and quickly capture high-frequency changes. The selection of its value has a significant impact on the time-frequency analysis results. The formula for calculating the Hamming window function is as follows:

[0222]

[0223] in, =1, 2, ..., , This refers to the total length of the Hanming window; Indicates the effective length of the window function; The value of the window function.

[0224] Furthermore, in response to the behavior of vehicles making sharp turns and frequently deviating laterally within the toll station diversion area, the total length of the window in this embodiment is set according to the typical duration of vehicle lateral deviation (i.e., the average value of vehicle lateral deviation) to cover the entire behavior cycle.

[0225] Further, determine the vehicle Whether lateral offset occurs specifically refers to:

[0226] If both frequency domain feature one and frequency domain feature two show peak values ​​at a certain moment, then the vehicle is considered to be at that moment... Lateral offset occurred; among which, frequency domain feature one is generated by the vehicle along... shaft speed After short-time Fourier transform, the second frequency domain feature is obtained from the vehicle's driving direction and... Angle between axes It is obtained through short-time Fourier transform.

[0227] A4. Vehicle The lateral offset recognition results, preprocessed trajectory data, attribute information, and environmental variables are input together into the iTransformer model for prediction. Then, the prediction results of the iTransformer model are input into the LSTM decoder module to generate the vehicle. The trajectory prediction results.

[0228] Preferably, to compensate for the iTransformer model's insufficient ability to capture short-term local dynamic changes, this embodiment introduces an LSTM decoder module (i.e., a Long Short-Term Memory network) to further model the high-dimensional spatiotemporal features output by the encoder. The LSTM decoder module uses the high-dimensional spatiotemporal features extracted by the iTransformer as input, and combines the hidden state and memory units from the previous time step, using a gating mechanism for information filtering and updating, finally generating the trajectory prediction result. In this embodiment, the iTransformer-LSTM model introduces an LSTM decoder module on top of the original iTransformer structure, enhancing its ability to model short-term local dynamic features, thereby improving the overall accuracy and stability of the prediction. Furthermore, the Short-Time Fourier Transform (SFT) converts the vehicle's lateral dynamic signal into a time-spectrum graph, accurately capturing the start and end times and frequency evolution patterns of transient behavior, improving the noise resistance and interpretability of behavior recognition. Using the vehicle's lateral offset behavior extracted by the SFT as input to the iTransformer-LSTM model enables earlier detection of frequency domain precursors of dangerous offsets, improving the accuracy and interpretability of the prediction.

[0229] Those skilled in the art will recognize that the iTransformer model is based on the Transformer network architecture and redefines the functions of the attention mechanism and the feedforward network. For any details regarding the iTransformer model not described in detail, please refer to the prior art.

[0230] Of course, in addition to the background vehicle (obstacle vehicle) trajectory prediction methods listed in this embodiment, those skilled in the art can also use other methods to predict the trajectory of the background vehicle, which will not be listed one by one in this embodiment.

[0231] Simulation Case 2:

[0232] In this simulation case, three mainstream prediction models commonly used in the field of deep learning are selected for comparison with the baseline model of the model proposed in this embodiment. The three mainstream prediction models are Transformer model, Informer model and CNN-LSTM model.

[0233] Within a framework of unified input feature variables, a comparative model set including Transformer, Informer, CNN-LSTM, iTransformer, and iTransformer-LSTM was constructed, and multi-model comparative experiments were conducted. With a fixed total time step of 10 s, vehicle trajectories at different time steps (1 s, 3 s, 4 s, 5 s, 6 s, 7 s) were predicted by adjusting different observation windows (9 s, 7 s, 6 s, 5 s, 4 s, 3 s). The trajectory prediction accuracy of each model in different prediction time domains was evaluated, and the specific comparison results are shown in Table 1.

[0234] Table 1 Comparison of evaluation metrics for each model

[0235]

[0236] Among them, evaluation indicators for and The loss function is composed of the comprehensive values, and the evaluation index FDE is the endpoint displacement error. FDE and FDE are represented as follows:

[0237]

[0238]

[0239] in, for The endpoint displacement error at time [time]. The mean square error between the actual trajectory and the predicted trajectory. The mean absolute error between the actual trajectory and the predicted trajectory. Representative vehicle in The actual horizontal coordinate of the moment Representative vehicle in The predicted horizontal coordinate at time. Representative vehicle in The true vertical coordinate of the moment, Representative vehicle in The predicted vertical coordinate at time.

[0240] Analyzing the comparative data in Table 1, it can be seen that the prediction accuracy of all five models decreases to varying degrees with increasing prediction duration. Furthermore, the loss value increases significantly when the prediction duration is 7 seconds, with the highest increase reaching 62.5%, far exceeding the other five prediction durations. The prediction model proposed in this embodiment performs best in prediction tasks with different time steps, especially in the 7-second prediction task, where its loss value of 8.87 is significantly better than Transformer (9.78) and CNN-LSTM (9.29), demonstrating the significant advantage of the prediction model in long-duration predictions. Specifically, iTransformer's performance is closest to that of the prediction model in this embodiment, but in predictions with short time steps (such as 1 second and 3 seconds), the loss value of iTransformer-LSTM is significantly lower than that of iTransformer, and its training performance is improved by an average of 26.76%. In long-step prediction (6s), the difference in loss values ​​between the two models narrowed, and training performance improved by 16.8%. However, in the 7s prediction task, iTransformer-LSTM maintained stable performance, while iTransformer's loss value was the highest among all models, reaching 9.98. iTransformer's prediction results for other time steps were better than Transformer's. Its loss value for the 7s prediction time was slightly higher than Transformer's by 0.2. This is because iTransformer relies solely on the encoder's output for prediction, resulting in incomplete information when generating predictions over longer time spans. The prediction model in this embodiment precisely compensates for this shortcoming. By fusing an LSTM decoder, iTransformer-LSTM enhances its ability to resolve local temporal patterns, making the model more stable in long-step trajectory prediction and effectively improving prediction accuracy.

[0241] Further analysis of the FDE (Final Determination) metric reveals that iTransformer-LSTM has the lowest FDE among all models (3.79), a 3% decrease compared to iTransformer (3.91), and a 14% improvement compared to the worst-performing CNN-LSTM (4.31), indicating smaller overall trajectory prediction bias and stronger stability. Furthermore, Informer and Transformer have similar FDE scores of 4.16 and 4.28 respectively, suggesting that Informer has slightly better trajectory stability in long-term predictions than Transformer. CNN-LSTM performs the worst across all metrics, exhibiting the highest FDE value, indicating a larger overall deviation between its predicted and actual trajectories. Overall, iTransformer-LSTM, leveraging the temporal modeling capabilities of the LSTM decoder, demonstrates optimal performance in both short- and long-term trajectory prediction, improving prediction accuracy and effectively reducing overall bias, thus ensuring the stability and accuracy of trajectory prediction.

[0242] Furthermore, to investigate the impact of lateral offset recognition, vehicle attribute information, and environmental variables on vehicle trajectory measurement results, this simulation case also compared the prediction performance of the STFT-iTransformer-LSTM model with input to the basic feature set and the enhanced feature set under the same hyperparameter configuration. The basic feature set contains only vehicle trajectory data, while the enhanced feature set includes lateral offset recognition results, vehicle attribute information, trajectory data, and environmental variables. The experimental results are shown in Table 2.

[0243] Table 2 Comparison of Indicators for Different Characteristic Variables

[0244]

[0245] As shown in Table 2, the enhanced feature set can effectively reduce [the impact of time step] in all prediction time steps. The enhanced feature set improved trajectory prediction accuracy. Compared to the basic feature set, the enhanced feature set reduced the loss value by an average of 5.64%, and across all prediction durations, its... The values ​​are all lower than those of the basic feature set, with the largest difference reaching 15% (at a prediction duration of 4 seconds), indicating a significant improvement in prediction accuracy. Especially under long-step prediction (such as 7 seconds), the loss value of the enhanced feature set is 8.87, compared to 9.19 of the basic feature set, representing a 3.5% reduction in error. This demonstrates that introducing other variables can effectively improve the stability of long-term vehicle prediction in the diversion zone.

[0246] The above analysis illustrates the crucial role of multi-source variables in trajectory prediction within toll station diversion areas, particularly the interaction characteristics between vehicles and surrounding traffic, as well as the significant impact of environmental factors on vehicle trajectory prediction results. This experiment further validates the effectiveness of lateral offset identification, vehicle attribute information, and environmental variables in improving trajectory prediction accuracy, demonstrating that fusing more feature information can enhance the model's time-series modeling capabilities, especially in long-step prediction tasks, effectively reducing errors.

[0247] Example 3:

[0248] This embodiment provides a vehicle trajectory planning system for a toll station diversion area that considers vehicle interaction, including a memory and a processor. The memory stores a computer program, and when the processor runs the computer program, it executes the trajectory planning method in Embodiment 1 and the obstacle vehicle trajectory prediction method in Embodiment 2.

[0249] Example 4:

[0250] This embodiment provides a storage medium storing a computer program, which, when run, executes the trajectory planning method in Embodiment 1 and the obstacle vehicle trajectory prediction method in Embodiment 2.

[0251] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A vehicle trajectory planning method for a toll station diversion area considering vehicle interaction, characterized in that, include: S1. Determine the initial target toll lane based on vehicle and environmental information at the toll station; S2. Determine the reference driving trajectory of the autonomous vehicle in the Cartesian coordinate system and transform the Cartesian coordinate system to the Frenet coordinate system; S3. Sample along the lateral and longitudinal directions within the trajectory planning area in front of the autonomous vehicle; S4, Predicting the Future Whether any obstructing vehicles enter the trajectory planning area at any given time; If the future When an unobstructed vehicle enters the trajectory planning area, the autonomous vehicle is set to cruise mode, generating a set of lateral trajectories based on lateral sampling points and a set of longitudinal trajectories based on longitudinal sampling points. If the future If an obstacle vehicle enters the trajectory planning area, longitudinal sampling is performed again based on the ST diagram according to overtaking and following strategies. A lateral trajectory set is generated based on the lateral sampling points, and a longitudinal trajectory set is generated based on the resampled longitudinal sampling points. Here, S represents the longitudinal displacement along the reference driving trajectory, and T represents the travel time. This represents any point in time between the current sample and the next sample. S5. Based on the fitting of the lateral trajectory set and the longitudinal trajectory set, all planned trajectories of the autonomous vehicle during the period from the current sampling to the next sampling are obtained. Based on the predicted trajectory of the obstacle vehicle, the planned trajectories with collision risk are eliminated using the three-dimensional spatiotemporal occupancy map, and the trajectory selection set composed of optional trajectories is obtained. S6. Find the optimal driving trajectory from the current sampling to the next sampling from the trajectory selection set. If the optimal driving trajectory is found, output it directly. If the optimal driving trajectory is not found, return to step S3. The predicted trajectory of the obstacle vehicle is obtained as follows: A1. Obtain the vehicle Multi-source time-series variables, including vehicles Before the start of trajectory planning Frame trajectory data, vehicles Attribute information and vehicle Environment variables; A2. Regarding vehicles forward The trajectory data of the frame is smoothed and filtered to obtain preprocessed trajectory data; A3. Using short-time Fourier transform, the vehicle's trajectory in the preprocessed trajectory data is analyzed. shaft speed and the vehicle's direction of travel Angle between axes Perform frequency domain analysis to determine the vehicle Has a lateral offset occurred? A4. Vehicle The lateral offset recognition results, preprocessed trajectory data, attribute information, and environmental variables are input together into the iTransformer model for prediction. Then, the prediction results of the iTransformer model are input into the LSTM decoder module to generate the vehicle. The trajectory prediction results; Among them, an ST graph is constructed based on the spatiotemporal interaction relationship between autonomous vehicles and obstacle vehicles entering the trajectory planning area, and an overtaking sampling area and a following sampling area are divided based on the ST graph. Longitudinal sampling is performed again in the overtaking sampling area and the following sampling area to obtain the sampling points of the overtaking and following strategies. The predicted trajectory based on obstacle vehicles uses a 3D spatiotemporal occupancy map to eliminate planned trajectories with collision risks. Specifically: Future The predicted and planned trajectories of vehicles that will enter the trajectory planning area at any time are projected onto the spatiotemporal occupancy map to analyze the occupancy status at different times; planned trajectories with conflict points are eliminated, and planned trajectories without conflict points are added to the trajectory selection set as optional trajectories; The specific method for finding the optimal driving trajectory from the trajectory selection set is as follows: Calculate the comprehensive evaluation index of each selectable trajectory in the optimal trajectory set. Each selectable trajectory is evaluated according to a comprehensive evaluation index. Sort the trajectories from low to high, and check whether each optional trajectory satisfies the vehicle dynamics constraints in turn. The first optional trajectory that satisfies the vehicle dynamics constraints is output as the optimal driving trajectory. If all optional trajectories in the trajectory selection set do not satisfy the vehicle dynamics constraints, return to step S3 to resample and plan the trajectory.

2. The vehicle trajectory planning method for toll station diversion areas considering vehicle interaction as described in claim 1, characterized in that, Comprehensive evaluation index of selectable trajectories Represented as: in, , , , and All are weighting factors. This represents the ratio of the length of the selectable trajectory to the length of the reference trajectory. This indicates the degree of curvature change in the selectable trajectory. This indicates the degree of deviation of the selected trajectory from the reference trajectory. This represents the ratio of the time corresponding to the output of the optional trajectory to the output time of the reference driving trajectory. This represents the collision cost of the optional trajectory.

3. The vehicle trajectory planning method for toll station diversion areas considering vehicle interaction according to claim 2, characterized in that, Collision cost of optional trajectories Represented as: in, Indicates the future Trajectory points of the selectable trajectory at time t. The distance between the trajectory points and the predicted trajectory of the obstacle vehicle. This represents the number of trajectory points into which the optional trajectory is discretized. This indicates the preset maximum horizontal distance; The degree of curvature change of the selectable trajectory Represented as: in, The first of the optional trajectories The curvature at the location of each trajectory point This represents the maximum curvature of the preset trajectory points; The degree of deviation of the selectable trajectory from the reference driving trajectory Represented as: in, The first of the optional trajectories The lateral offset distance between each trajectory point and the reference driving trajectory. This indicates the preset maximum lateral offset distance; The ratio of the length of the selectable trajectory to the length of the reference trajectory. Represented as: in, Indicates the length of the optional trajectory. Indicates the length of the reference driving trajectory; The ratio of the time corresponding to the selectable trajectory to the output time of the reference trajectory. Represented as: in, This indicates the time corresponding to the output of the optional trajectory. This indicates the output time of the reference driving trajectory.

4. The vehicle trajectory planning method for toll station diversion areas considering vehicle interaction according to claim 1, characterized in that, The vehicle dynamics constraints are expressed as follows: in, This indicates the minimum permissible acceleration. Indicates the maximum permissible acceleration. This represents the acceleration of any point on the selectable trajectory. Indicates the minimum permitted driving speed. Indicates the maximum permissible driving speed. This represents the speed at any point on the selectable trajectory. This represents the minimum permissible curvature. Indicates the maximum permissible curvature. This represents the curvature of any point on the selectable trajectory.

5. The vehicle trajectory planning method for toll station diversion areas considering vehicle interaction according to claim 1, characterized in that, Determine the vehicle Whether lateral offset occurs specifically refers to: If both frequency domain feature one and frequency domain feature two show peak values ​​at a certain moment, then the vehicle is considered to be at that moment... Lateral offset occurred; among which, frequency domain feature one is generated by the vehicle along... shaft speed After short-time Fourier transform, the second frequency domain feature is obtained from the vehicle's driving direction and... Angle between axes It is obtained through short-time Fourier transform.

6. A vehicle trajectory planning system for a toll station diversion area considering vehicle interaction, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the trajectory planning method as described in any one of claims 1-5 when running the computer program.

7. A storage medium, characterized in that, The storage medium stores a computer program, which, when run, executes the trajectory planning method as described in any one of claims 1-5.

Images