Vehicle driving track determination method, system, electronic device and vehicle

Abstract

This application provides a method, system, electronic device, and vehicle for determining a vehicle's driving trajectory. The method includes: acquiring a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located during vehicle operation; determining an initial driving trajectory of the target vehicle based on the trajectory point sequence, wherein the initial driving trajectory indicates the driving of the environmental object at the current moment; adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the executability index of the adjusted initial driving trajectory is greater than the executability index of the initial driving trajectory before adjustment; and determining a target driving trajectory of the vehicle from at least one adjusted initial driving trajectory, wherein the target driving trajectory indicates the vehicle's driving. This application solves the technical problem of low accuracy in determining vehicle driving trajectories.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and more specifically, to a method, system, electronic device, and vehicle for determining the driving trajectory of a vehicle. Background Technology

[0002] Currently, vehicle trajectory determination mainly relies on high-precision maps to extract lane centerlines, navigation maps to plan global paths, or visual perception of lane lines. However, in scenarios with no or weak maps (such as temporary construction zones or rural roads), map gaps or invisible lane lines can lead to trajectory determination failure. Therefore, the technical problem of low accuracy in vehicle trajectory determination still exists.

[0003] There is currently no good solution to the above problems. Summary of the Invention

[0004] This application provides a method, system, electronic device, and vehicle for determining the driving trajectory of a vehicle, in order to at least solve the technical problem of low accuracy in determining the driving trajectory of a vehicle.

[0005] According to one aspect of the embodiments of this application, a method for determining the driving trajectory of a vehicle is provided. The method may include: acquiring a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located during vehicle operation; determining an initial driving trajectory of the target vehicle based on the trajectory point sequence of the target vehicle, wherein the initial driving trajectory is used to indicate the driving of the environmental object at the current moment; adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the executability index of the adjusted initial driving trajectory is greater than the executability index of the initial driving trajectory before adjustment; and determining a target driving trajectory of the vehicle from at least one adjusted initial driving trajectory, wherein the target driving trajectory is used to indicate the driving of the vehicle.

[0006] Furthermore, during the vehicle's movement, the trajectory point sequence of at least one target vehicle in the environment where the vehicle is located is obtained, including: during the vehicle's movement, obtaining the initial trajectory point sequence of the target vehicle; adjusting the initial trajectory point sequence using the coordinate type of the target vehicle to obtain the trajectory point sequence, wherein the coordinate type is used to indicate the type of coordinate system to which the target vehicle's location belongs, and the accuracy of the trajectory point sequence is greater than the accuracy of the initial trajectory point sequence.

[0007] Furthermore, the initial trajectory point sequence is adjusted using the target vehicle's coordinate type to obtain a new trajectory point sequence. This includes: in response to a geographic coordinate system, transforming the initial trajectory point sequence from the geographic coordinate system to the projected coordinate system to obtain a transformed initial trajectory point sequence; and in response to a relative coordinate system of the target vehicle, transforming the initial trajectory point sequence from the relative coordinate system to the vehicle's own coordinate system to obtain a transformed initial trajectory point sequence. The own coordinate system is constructed based on the vehicle's driving state information. The transformed trajectory point sequence is then adjusted using the vehicle's driving scenario. The initial trajectory point sequence is corrected to obtain a trajectory point sequence, wherein the adaptability of the trajectory point sequence to the driving scene is greater than that of the adaptability of the initial trajectory point sequence after coordinate system transformation to the driving scene; and / or, during the vehicle's driving process, the initial trajectory point sequence of the target vehicle is obtained, including: during the vehicle's driving process, the trajectory points of the target vehicle are collected according to the sampling strategy corresponding to the driving scene, wherein the sampling strategy is used to represent the rules for controlling the vehicle to collect trajectory points according to the sampling frequency, and different driving scenes correspond to different sampling frequencies; the target number of trajectory points are determined as the initial trajectory point sequence.

[0008] Furthermore, based on the trajectory point sequence of the target vehicle, the initial driving trajectory of the target vehicle is determined, including: calling a fitting strategy adapted to the density of the trajectory point sequence to fit the trajectory point sequence and obtain the initial driving trajectory, wherein the density is used to represent the number of trajectory points per unit distance in the trajectory point sequence, and the fitting strategy is used to represent the rules for fitting the trajectory point sequence.

[0009] Furthermore, the fitting strategy includes a first fitting strategy, which represents the rules for fitting the trajectory point sequence according to a preset spline curve. The first fitting strategy calls a fitting strategy adapted to the density of the trajectory point sequence to fit the trajectory point sequence and obtain an initial driving trajectory. This includes: in response to a density greater than or equal to a density threshold, determining nodes based on the order of the preset spline curve and the number of trajectory points, and determining control vertices based on the coordinates of the trajectory points; calling the first fitting strategy to fit the nodes and control vertices according to the preset spline curve to obtain the initial driving trajectory; the method further includes: in response to a fitting error of the initial driving trajectory determined according to the first fitting strategy exceeding a fitting error threshold, adjusting the interval between nodes; and refitting the initial driving trajectory using the nodes with adjusted intervals. The process continues until the fitting error is less than or equal to a fitting error threshold; and / or, the fitting strategy includes a second fitting strategy, which represents a rule for fitting the trajectory point sequence according to a preset polynomial. The method calls a fitting strategy adapted to the density of the trajectory point sequence to fit the trajectory point sequence and obtain an initial driving trajectory. This includes: determining the weights of the trajectory points in response to a density less than a density threshold; determining the polynomial coefficients of the preset polynomial based on the weights; calling the second fitting strategy and fitting the initial driving trajectory according to the polynomial coefficients; the method further includes: adjusting the weights in response to a fitting error of the initial driving trajectory determined according to the second fitting strategy being greater than a fitting error threshold; and refitting the initial driving trajectory using the adjusted weights until the fitting error is less than or equal to the fitting error threshold.

[0010] Furthermore, the executability indicators include at least one of the following: smoothness indicator, safety indicator, dynamic indicator, and consistency indicator. The consistency indicator is used to represent the degree of difference between the initial driving trajectory and the vehicle's driving trend. Adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory includes at least one of the following: responding to the curvature of the initial driving trajectory being greater than the curvature threshold corresponding to the vehicle's driving scenario, adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein curvature is used to represent the smoothness of the initial driving trajectory, the curvature of the adjusted initial driving trajectory is less than or equal to the curvature threshold, and the smoothness indicator of the adjusted initial driving trajectory is greater than the smoothness indicator of the initial driving trajectory before adjustment; responding to the distance between the initial driving trajectory and obstacles around the vehicle being less than a distance threshold, adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the distance between the adjusted initial driving trajectory and obstacles is greater than or equal to the distance threshold, and the safety indicator of the adjusted initial driving trajectory is greater than or equal to the distance threshold. The initial driving trajectory is adjusted based on the following criteria: First, the initial driving trajectory is adjusted to a target driving scenario, where the target driving scenario is used to indicate the dynamic range of the vehicle's movement. Second, the dynamic range of the adjusted initial driving trajectory is greater than the target driving scenario, and the dynamic range of the driving scenario is greater than the dynamic range of the initial driving scenario before adjustment. Third, the initial driving trajectory is adjusted to a differentiability index than the differentiability threshold, where the driving trend represents the vehicle's future driving state. Fourth, the difference between the adjusted initial driving trajectory and the driving trend is less than or equal to the differentiability threshold, and the consistency index of the adjusted initial driving trajectory is greater than the consistency index of the initial driving trajectory before adjustment.

[0011] Furthermore, the method further includes at least one of the following: in response to the target vehicle being in a failed state, extending the initial driving trajectory based on the trajectory points of the target vehicle before it became failed, to obtain an extended initial driving trajectory, wherein the driving length of the extended initial driving trajectory is greater than the driving length of the initial driving trajectory before it was extended, and / or the driving time of the extended initial driving trajectory is greater than the driving time of the initial driving trajectory before it was extended; in response to the target vehicle satisfying one of the following conditions, generating a target driving trajectory based on the vehicle's map information and the vehicle's perception information: the number of target vehicles is less than one; the target vehicle is in a failed state, and there are no target vehicles in a normal state within a preset time period; in response to the number of target vehicles in a normal state being greater than one within a preset time period, determining the initial driving trajectory using the trajectory point sequence of target vehicles in a normal state.

[0012] According to another aspect of the embodiments of this application, a vehicle driving trajectory determination system is also provided. The system may include: a client for acquiring a vehicle trajectory determination instruction; a server for responding to the trajectory determination instruction during vehicle operation, acquiring a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located; determining an initial driving trajectory of the target vehicle based on the trajectory point sequence of the target vehicle, wherein the initial driving trajectory is used to indicate the driving of an environmental object at the current moment; adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the executability index of the adjusted initial driving trajectory is greater than the executability index of the initial driving trajectory before adjustment; and determining a target driving trajectory of the vehicle from at least one adjusted initial driving trajectory, wherein the target driving trajectory is used to indicate the vehicle's driving.

[0013] According to another aspect of the embodiments of this application, a vehicle driving trajectory determination device is also provided. The device may include: an acquisition module, configured to acquire a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located during the vehicle's driving process; a first determination module, configured to determine an initial driving trajectory of the target vehicle based on the sequence of trajectory points of the target vehicle, wherein the initial driving trajectory is used to indicate the driving of the environmental object at the current moment; an adjustment module, configured to adjust the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the executability index of the adjusted initial driving trajectory is greater than the executability index of the initial driving trajectory before adjustment; and a second determination module, configured to determine the target driving trajectory of the vehicle from at least one adjusted initial driving trajectory, wherein the target driving trajectory is used to indicate the driving of the vehicle.

[0014] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0015] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0016] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0017] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0018] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the methods in various embodiments of this application.

[0019] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0020] In this embodiment, if it is necessary to determine the vehicle's driving trajectory, the trajectory point sequence of at least one target vehicle in the environment where the vehicle is located can be obtained during the vehicle's movement. The initial driving trajectory of the target vehicle can be determined based on the aforementioned trajectory point sequence. To ensure the executability of the final determined target driving trajectory, the initial driving trajectory can be adjusted to obtain an adjusted initial driving trajectory. The target driving trajectory of the vehicle can be determined from the adjusted initial driving trajectory. In other words, in this embodiment, by introducing the initial driving trajectory of the target vehicle in the environment, the deficiencies of high-precision maps and visual lane lines in map-less / weak-map scenarios are compensated for. The initial driving trajectory is directly constructed based on the trajectory point sequence of at least one target vehicle, and the executability index is used to optimize the initial driving trajectory, improving its executability. The optimal target driving trajectory for the vehicle is output from multiple optimized initial driving trajectories, without relying on maps or lane lines, thus ensuring the reliability of the driving trajectory in map-less / weak-map scenarios such as construction areas and rural roads. This achieves the technical effect of improving the accuracy of vehicle driving trajectory determination and solves the technical problem of low accuracy in vehicle driving trajectory determination. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0022] Figure 1 This is a schematic diagram of an application scenario for determining the driving trajectory of a vehicle according to an embodiment of this application;

[0023] Figure 2 This is a flowchart of a method for determining the driving trajectory of a vehicle according to an embodiment of this application;

[0024] Figure 3 This is a flowchart of a method for constructing a trajectory reference line for an intelligent vehicle according to an embodiment of this application;

[0025] Figure 4 This is a schematic diagram of a vehicle trajectory determination system according to an embodiment of this application;

[0026] Figure 5 This is a schematic diagram of a vehicle trajectory determination device according to an embodiment of this application;

[0027] Figure 6 This is a schematic diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] According to an embodiment of this application, a method for determining the driving trajectory of a vehicle is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] Figure 1 This is a schematic diagram illustrating an application scenario for determining the driving trajectory of a vehicle according to an embodiment of this application, such as... Figure 1 As shown, the scenario described above may include terminal device 10, network 20, and vehicle 30. Terminal device 10 can be used to obtain trajectory determination instructions from vehicle users (e.g., drivers, passengers) regarding whether the vehicle needs to determine its own driving trajectory. The terminal device can be a mobile phone, computer, or other device used by the driver in the vehicle, or an interactive interface used for user interaction. The trajectory determination instructions can be sent to vehicle 30 via network 20. At this point, vehicle 30 needs to execute steps S102 to S108 to realize the vehicle's driving trajectory determination process.

[0032] The following steps can be performed by vehicle 30: Step S102, during the vehicle's driving process, acquire the trajectory point sequence of at least one target vehicle in the environment where the vehicle is located; Step S104, based on the trajectory point sequence of the target vehicle, determine the initial driving trajectory of the target vehicle; Step S106, adjust the initial driving trajectory to obtain the adjusted initial driving trajectory; Step S108, determine the target driving trajectory of the vehicle from at least one adjusted initial driving trajectory.

[0033] In this embodiment, through steps S102 to S108, the initial driving trajectory of the target vehicle in the environment is introduced to compensate for the lack of high-precision maps and visual lane lines in map-less / weak-map scenarios. An initial driving trajectory is directly constructed based on the trajectory point sequence of at least one target vehicle, and the initial driving trajectory is optimized using executability metrics to improve its executability. The optimal target driving trajectory for the vehicle is output from multiple optimized initial driving trajectories, without relying on maps or lane lines, thus ensuring the reliability of the driving trajectory in map-less / weak-map scenarios such as construction zones and rural roads. This achieves the technical effect of improving the accuracy of vehicle driving trajectory determination and solves the technical problem of low accuracy in vehicle driving trajectory determination.

[0034] This embodiment provides a method for determining the driving trajectory of a vehicle. Figure 2 This is a flowchart of a method for determining the driving trajectory of a vehicle according to an embodiment of this application, as shown below. Figure 2 As shown, the method may include the following steps.

[0035] Step S202: During the vehicle's movement, acquire the trajectory point sequence of at least one target vehicle in the environment where the vehicle is located.

[0036] In the technical solution provided by step S202 of this application embodiment, the target vehicle can refer to other vehicles that can be identified and continuously tracked in the vehicle driving environment. The driving behavior of the target vehicle is representative and stable, and can reflect the reasonable traffic pattern under the current road conditions. The target vehicle can serve as an environmental reference source to indirectly infer drivable paths in scenarios without lane lines or maps. The number of target vehicles can be 1 to 5, without specific limitations, and can be determined based on screening conditions such as driving continuity, speed adaptability, heading consistency, and relative distance to other vehicles. For example, the target vehicle can be a vehicle traveling in the same direction as the vehicle, without violations, and with a stable driving trajectory.

[0037] Optionally, the trajectory point sequence can refer to a set of spatial position coordinates of the target vehicle continuously recorded at a sampling frequency in the time dimension. The trajectory point sequence can include multiple trajectory points. Each trajectory point can contain three-dimensional spatial coordinates, a timestamp, a heading angle, and optional longitudinal and lateral velocity information. These trajectory points can be used to characterize the actual motion trajectory of the target vehicle on the current road. The trajectory point sequence can be generated by fusing data from LiDAR, cameras, or millimeter-wave radar. For example, the trajectory point sequence can contain at least 30 consecutive trajectory points to fully depict the motion characteristics of the target vehicle during straight-line or turning processes. The aforementioned three-dimensional spatial coordinates can be Universal Transverse Mercator (UTM) projection or x and y coordinates in the vehicle's coordinate system.

[0038] In this embodiment, during the vehicle's movement, target vehicles in the environment can be detected to obtain a sequence of trajectory points of the target vehicles.

[0039] Optionally, during vehicle operation, environmental perception data can be collected through the vehicle's onboard multi-sensor fusion system. This environmental perception data may include LiDAR point clouds, camera images, and millimeter-wave radar target echoes. By combining target detection and tracking algorithms, dynamic vehicle targets in the vehicle's environment can be identified from the aforementioned environmental perception data, and the position, speed, heading angle, and tracking confidence information of each dynamic vehicle target can be output.

[0040] Optionally, stability and behavioral compliance screening can be implemented for the detected dynamic vehicle targets. For example, dynamic vehicle targets that have been continuously tracked for no less than 3 seconds, whose speed meets the scene constraints (straight-going ≤ 80km / h, turning ≤ 30km / h), whose absolute acceleration value does not exceed 2m / s², whose heading angle change rate is below the threshold (straight-going ≤ 4° / s, turning ≤ 20° / s), who do not engage in reverse driving or abnormal lane changing behavior, whose distance from the vehicle is between 2 and 60 meters, and whose driving intention matches the current scene (e.g., vehicles with the same turning intention are prioritized in intersection scenes) can be identified as target vehicles. For each target vehicle, a preset number of trajectory points can be continuously collected at a scene-adaptive frequency (e.g., 10Hz for straight-going scenes, 20Hz for turning scenes), and the spatial coordinates, timestamps, heading angles, and corresponding vehicle state parameters of the trajectory points can be obtained to form a structured trajectory point sequence.

[0041] It should be noted that the process and method for filtering out target vehicles from dynamic vehicle targets described in this embodiment of the application are merely illustrative examples. The above process is mainly to ensure the accuracy of target vehicle identification and the accuracy of vehicle trajectory determination, and no specific limitations are imposed here.

[0042] In the embodiments of this application, the above method ensures that the acquired trajectory point sequence has good spatiotemporal continuity, behavioral representativeness and scene adaptability by screening dynamic vehicle targets and using a high-precision trajectory point sampling mechanism. This provides a highly reliable and low-noise data input source for subsequent multi-trajectory fitting and optimization, effectively supports the autonomous generation of reference trajectories in scenarios without or with weak maps, avoids trajectory deviations caused by false detections or drifting targets, and significantly improves the robustness and engineering feasibility of subsequent trajectory construction.

[0043] Step S204: Determine the initial driving trajectory of the target vehicle based on the trajectory point sequence of the target vehicle.

[0044] In the technical solution provided by step S204 of this application embodiment, the initial driving trajectory is used to indicate the driving of an environmental object at the current moment. The initial driving trajectory can refer to a continuous geometric curve generated by mathematical fitting based on the trajectory point sequence of a single target vehicle, representing the actual driving path of the target vehicle in the current road environment. The initial driving trajectory can also be called an initial reference line. The initial driving trajectory does not rely on high-precision maps or lane line information, but is constructed based on the measured motion data of the target vehicle in the environment. The geometric shape of the initial driving trajectory can be used to reflect the actual traffic pattern of the target vehicle in unstructured or weakly structured roads, and can be used to indirectly infer a safe driving path that the vehicle can follow. The initial driving trajectory consists of a set of smooth, continuous coordinate points, and can be expressed as a 3rd-order B-spline curve or a cubic polynomial function. The fitting algorithm is adaptively selected based on the trajectory point density, and the fitting error is controlled within 0.3 meters, ensuring that the curve maintains trajectory details while meeting the requirements of vehicle dynamics continuity. Each initial driving trajectory corresponds to a target vehicle. The start and end range of the initial driving trajectory can cover the complete motion range of the target vehicle within the sampling period, including position, tangent direction and curvature change information.

[0045] In this embodiment, after obtaining the trajectory point sequence of the target vehicle in the environment where the vehicle is located, the initial driving trajectory of the target vehicle can be determined based on the trajectory point sequence of the target vehicle.

[0046] Optionally, after obtaining the trajectory point sequence, density analysis can be performed on the trajectory point sequence of the target vehicle. For example, a 1-meter sliding window can be used to calculate the number of trajectory points in the trajectory point sequence within a unit length to determine the spatial sampling density of the target vehicle. If the density of trajectory points in the trajectory point sequence is not less than 2 points / meter, a 3rd-order B-spline curve can be used for fitting to obtain the initial driving trajectory, ensuring that the generated initial driving trajectory has G2 continuity (position, first derivative, and second derivative are all continuous), meeting the smoothness requirements of vehicle driving.

[0047] Optionally, if the density of trajectory points in the trajectory point sequence is less than 2 points / meter, the initial driving trajectory can be obtained by fitting a cubic polynomial function using the weighted least squares method.

[0048] It should be noted that the above process and method for determining the initial driving trajectory based on the trajectory point sequence are only illustrative examples and are not specifically limited here. Any process and method that can plan the vehicle's own driving trajectory in the future by using the driving trajectory of the target vehicle in the vehicle's environment is within the protection scope of the embodiments of this application.

[0049] In this embodiment, the mechanism of obtaining the initial driving trajectory through density-driven adaptive fitting achieves accurate modeling of trajectory point sequences under different conditions. Under the conditions of no lane lines and no high-precision map, it effectively transforms the measured behavior of dynamic environmental targets into structured and computable geometric reference lines, providing high-fidelity and low-bias initial input for multi-trajectory parallel optimization, and significantly improving the accuracy and scene adaptability of subsequent trajectory generation.

[0050] Step S206: Adjust the initial driving trajectory to obtain the adjusted initial driving trajectory.

[0051] In the technical solution provided by step S206 of this application embodiment, the executability index of the adjusted initial driving trajectory is greater than that of the original initial driving trajectory. The adjusted initial driving trajectory can refer to an improved geometric path that conforms to vehicle safety driving specifications, generated after modification of the executability index based on the original initial driving trajectory. The adjusted initial driving trajectory can retain the topological structure and semantic meaning of the original initial driving trajectory, but through the constraints of multi-dimensional executability indicators, it can eliminate curvature abrupt changes, avoid static obstacles, and meet the dynamic limits of steering angle and lateral acceleration, ensuring that the adjusted initial driving trajectory has engineering feasibility for trajectory planning and direct vehicle invocation in real traffic environments. The adjusted initial driving trajectory can be expressed as a continuous curve, and its mathematical representation can be the same as the initial driving trajectory (e.g., a 3rd-order B-spline or a cubic polynomial), but the control parameters of the adjusted initial driving trajectory have been re-optimized according to the constraints of the executability index.

[0052] The aforementioned feasibility indicators refer to comprehensive performance benchmarks used to quantitatively evaluate whether a driving trajectory can enable a vehicle to execute safely, stably, and continuously. These benchmarks encompass three core dimensions: safety, feasibility, and continuity. Safety refers to maintaining a minimum clearance of at least 2 meters between the initial driving trajectory and surrounding static obstacles (such as curbs, traffic signs, and concrete pillars) to ensure no collision risk. Feasibility refers to the achievability of the initial driving trajectory within the vehicle's dynamic capabilities, including hard constraints such as a steering angle not exceeding ±35°, lateral acceleration not exceeding ±1.5 m / s² in straight-line scenarios, not exceeding ±2.5 m / s² in turning scenarios, and a minimum turning radius of at least 5 meters (calculated based on a 2.8-meter wheelbase). Continuity refers to the smoothness of the initial driving trajectory in terms of spatial and curvature changes, requiring a curvature change rate not exceeding 0.15 rad / m, with no sharp turns or discontinuous derivatives, to ensure stable vehicle response and the comfort of passengers.

[0053] It should be noted that the dimensions included in the above-mentioned executable indicators in the embodiments of this application are merely illustrative examples and are not specifically limited here. Any dimension that can correct the initial driving trajectory to ensure the safety and rationality of vehicle driving, as well as the parameters under that dimension, are within the scope of the executable indicators in the embodiments of this application.

[0054] In this embodiment, after determining the initial driving trajectory of the target vehicle, the initial driving trajectory can be adjusted according to the feasibility index to obtain the adjusted initial driving trajectory.

[0055] Optionally, adjusting the initial driving trajectory to generate a revised initial driving trajectory with higher executability indicators is a key transformation step in realizing the driving decision path from environmental behavior perception to safe execution. This embodiment introduces constraints of three types of executability indicators—smoothness, static safety, and large dynamic scene dynamics—to verify and iteratively correct the original initial driving trajectory item by item, ensuring that the initial driving trajectory meets the engineering requirements for stable vehicle operation in terms of geometry, spatial obstacle avoidance, and vehicle dynamics capabilities.

[0056] Optionally, the initial driving trajectory can be optimized for smoothness to improve the executability index of its continuity dimension. By calculating the curvature of each point on the trajectory and the rate of change of the initial driving trajectory along the path direction, local abrupt changes in curvature exceeding 0.15 rad / m are identified. These local abrupt changes correspond to path characteristics in vehicle control that are prone to yaw fluctuations or steering jitter. To eliminate such discontinuities, without disrupting the overall trajectory direction, a quadratic programming algorithm can be used to locally modify the initial driving trajectory by adjusting the control vertices or polynomial coefficients of the affected trajectory segments. This ensures that the rate of change of curvature does not exceed a preset threshold throughout the entire trajectory length, guaranteeing the continuity of the first and second derivatives of the trajectory. This satisfies the smooth response requirements of the vehicle's longitudinal and lateral control systems, enabling the continuity executability index of the initial driving trajectory to meet the required level.

[0057] Optionally, static safety verification of the initial driving trajectory can improve its executability in terms of safety. The position and geometric dimensions of static obstacles such as curbs, traffic signs, construction barriers, and concrete pillars output by the vehicle are obtained in a unified coordinate system. Spatial distance calculations are then performed between this position and geometric dimension information and each initial driving trajectory. If the shortest normal distance between any trajectory point and any static obstacle in the initial driving trajectory is less than 2 meters, the executability of the initial driving trajectory in terms of safety is deemed unsatisfactory. For the trajectory segments formed by these unsatisfactory trajectory points, fine-tuning is performed within a permissible lateral adjustment range (e.g., ±1.5 meters) while ensuring the steering intent of the initial driving trajectory remains unchanged, maintaining a safe clearance of at least 2 meters between the initial driving trajectory and each static obstacle. This adjustment can be performed locally in areas with gentle changes in path curvature and no conflict during turning, avoiding distortion of the initial driving trajectory due to excessive offset and ensuring the adjusted trajectory has engineering executability in terms of spatial obstacle avoidance.

[0058] Optionally, dynamic verification of the initial driving trajectory in high-dynamic scenarios is performed to improve its feasibility in terms of feasibility. For high-dynamic scenarios such as left and right turns at intersections, the change in heading angle corresponding to each trajectory point on the initial driving trajectory is extracted, the vehicle steering angle corresponding to the trajectory point is calculated, and it is verified whether it exceeds the maximum steering capability threshold of ±35°. At the same time, combined with the trajectory curvature and vehicle speed, the lateral acceleration is calculated and checked whether it exceeds the dynamic constraint upper limit of ≤±1.5m / s² for straight-ahead scenarios or ≤±2.5m / s² for turning scenarios. In addition, based on the vehicle wheelbase of 2.8 meters, the minimum turning radius of the trajectory is calculated backward through geometric relationships to ensure that it is not less than 5 meters, so as to avoid the vehicle being unable to complete the turning action due to the turning radius being too small. For any trajectory segment where any dynamic parameter exceeds the constraint range, a segmented correction strategy is adopted to maintain the starting and ending targets of the initial driving trajectory and the steering intention. Figure 1 Under the premise of consistency, by adjusting the control vertices or polynomial coefficients, the curvature distribution is redistributed within the dynamic feasible region until all dynamic indicators meet the constraints, thus completing the executability improvement in the feasibility dimension.

[0059] It should be noted that the three adjustment operations described above in this application embodiment are executed independently but work synergistically. Each operation can be directed to an executable indicator under a corresponding dimension. After adjusting the initial driving trajectory, the constraints of the executable indicators of each dimension can be re-verified on the adjusted initial driving trajectory to ensure that the adjusted initial driving trajectory does not introduce new violations.

[0060] In this embodiment, the method described above, through a multi-dimensional and verifiable adjustment process, achieves an initial driving trajectory that is significantly superior to the original initial driving trajectory in terms of safety, feasibility, and continuity—three executability indicators. This process does not rely on external high-precision maps or lane line priors; it can be constructed based on the environmental target vehicle behavior and the vehicle's own dynamic capabilities. This significantly improves the robustness and engineering feasibility of reference trajectory generation for intelligent vehicles in complex scenarios such as those without maps / weak maps or lane lines, providing a stable, safe, and continuous geometric foundation for subsequent multi-trajectory selection and decision control, effectively ensuring the continuity and safety of intelligent driving functions in extreme environments.

[0061] Step S208: Determine the target driving trajectory of the vehicle from at least one adjusted initial driving trajectory.

[0062] In the technical solution provided by step S208 of the embodiments of this application, the target driving trajectory is used to indicate vehicle movement. The target driving trajectory can refer to a single or a set of geometric paths selected from multiple adjusted initial driving trajectories, based on the scenario requirements of the road the vehicle is currently traveling on, its driving intention, and system redundancy strategies, as the current driving control benchmark for the vehicle. The target driving trajectory can be used to instruct the vehicle to complete longitudinal speed adjustment, lateral steering execution, and path tracking. Essentially, it is an engineering reference path with the highest executability and scenario adaptability, output after multiple rounds of optimization of environmental vehicle behavior information. The target driving trajectory can consist of one or more adjusted initial driving trajectories, and the number and selection strategy of the target driving trajectories depend on the road structure and operating scenario. For example, in a single-lane, interference-free scenario, a target driving trajectory that matches the vehicle's current driving intention can be selected; in scenarios involving left turns, right turns, or multiple parallel lanes at intersections, multiple adjusted initial driving trajectories corresponding to different steering intentions or lane positions can be retained simultaneously, forming a candidate trajectory set for dynamic switching by the vehicle.

[0063] In this embodiment, after adjusting the initial driving trajectory to obtain the adjusted initial driving trajectory, the target driving trajectory of the vehicle can be determined from at least one adjusted initial driving trajectory.

[0064] Optionally, scenario intent matching can be performed on each adjusted initial driving trajectory to determine the semantic consistency between the adjusted initial driving trajectory and the vehicle's current driving intent. For example, in a left or right turn scenario at an intersection, the trend of the heading angle change of each adjusted initial driving trajectory can be analyzed to identify whether it exhibits a continuous left or right veer characteristic. If the vehicle plans to perform a left turn, the initial driving trajectory with a continuously negative heading angle change rate and a change range consistent with the turning intent can be retained, while the initial driving trajectory with a heading angle tending towards straight or right veer can be excluded. In straight or rural road scenarios, the initial driving trajectory with a heading angle change rate of less than 4° / s and a trajectory offset of less than 0.8 meters can be retained to ensure consistency with the main direction of the road.

[0065] It should be noted that the process of scene intent matching for the initial driving trajectory described in the embodiments of this application can be based on semantic alignment between the initial driving trajectory and the vehicle's planned path, rather than simply relying on geometric similarity.

[0066] Optionally, the initial driving trajectories that meet the scenario intent matching are ranked by trajectory quality to evaluate the comprehensive performance of the initial driving trajectories' executability indicators. Each initial driving trajectory can be verified to meet three types of executability indicators: smoothness, static safety, and dynamic feasibility. Since different initial driving trajectories perform differently on different executability indicators, a normalized score can be generated for each initial driving trajectory's rate of curvature change, minimum clearance from static obstacles, maximum lateral acceleration, and minimum turning radius based on a preset weighting system, forming a comprehensive quality score for each initial driving trajectory. In a multi-lane environment, if multiple initial driving trajectories correspond to different lane positions and all pass verification, the initial driving trajectory that matches the vehicle's current lane position can be further prioritized based on lane availability information. In scenarios without lane lines, the initial driving trajectory with the smallest rate of curvature change, the largest average distance from obstacles, and the highest dynamic constraint margin is prioritized as the target driving trajectory.

[0067] Optionally, based on vehicle redundancy mechanisms and update cycles, a dynamic selection and switching strategy for target driving trajectories can be implemented. If the target vehicle corresponding to the current target driving trajectory is detected to have lost tracking or the trajectory has failed, and there are at least one other verified adjusted initial driving trajectory, the system can switch to an initial driving trajectory with a suboptimal quality score that meets all executability indicators, achieving uninterrupted continuous path supply. If there are no other available initial driving trajectories, a redundancy extension mechanism can be activated, performing linear extrapolation based on the trajectory points of the last 15 frames before the failure, with a delay of no more than 1 second. Simultaneously, the target selection and trajectory optimization process is re-executed during this buffer period to quickly restore the ability to supply multiple initial driving trajectories. The update cycle of the target driving trajectory can be dynamically adjusted according to the scenario. For example, in a straight-ahead scenario, evaluation and selection can be performed at a cycle of 100ms, while in an intersection turning scenario, this can be shortened to 50ms, ensuring that the selection of the target driving trajectory can respond to real-time requirements of environmental changes. Each switching process can employ a smooth transition strategy of 0.3 seconds to avoid control oscillations caused by sudden changes in the target driving trajectory.

[0068] Optionally, if the vehicle has high-precision positioning capabilities, the target driving trajectory can be output as a continuous curve in the UTM coordinate system; if there is no absolute positioning, it can be output as a relative coordinate sequence in the vehicle's own coordinate system, and the coordinate origin can be updated synchronously to maintain the consistency of the target driving trajectory.

[0069] In this embodiment, the method employs a multi-layered decision-making process involving scene intent matching, trajectory quality ranking, redundancy switching, and dynamic updates. This ensures the precise selection of the target driving trajectory from multiple adjusted initial driving trajectories, achieving a good balance between semantic rationality, engineering feasibility, and system continuity. This mechanism not only avoids functional interruptions caused by the failure of a single trajectory but also significantly improves the robustness and safety of intelligent vehicles' path decision-making in complex scenarios such as mapless, laneless, and highly dynamic intersections through a multi-trajectory parallel evaluation and optimization mechanism.

[0070] In steps S202 to S208 of this embodiment, if it is necessary to determine the vehicle's driving trajectory, the trajectory point sequence of at least one target vehicle in the environment where the vehicle is located can be obtained during the vehicle's driving process. The initial driving trajectory of the target vehicle can be determined based on the trajectory point sequence. To ensure the executability of the final determined target driving trajectory, the initial driving trajectory can be adjusted to obtain an adjusted initial driving trajectory. The target driving trajectory of the vehicle can be determined from the adjusted initial driving trajectory. That is, in this embodiment, by introducing the initial driving trajectory of the target vehicle in the environment, the lack of high-precision maps and visual lane lines in map-less / weak map scenarios is compensated for. The initial driving trajectory is directly constructed based on the trajectory point sequence of at least one target vehicle, and the executability index is used to optimize the initial driving trajectory, thereby improving the executability of the initial driving trajectory. The target driving trajectory adapted to the vehicle is output from multiple optimized initial driving trajectories without relying on maps or lane lines, thus ensuring the reliability of the driving trajectory in map-less / weak map scenarios such as construction areas and rural roads. This achieves the technical effect of improving the accuracy of vehicle trajectory determination and solves the technical problem of low accuracy in vehicle trajectory determination.

[0071] The embodiments of this application will be described in detail below with reference to the steps described above.

[0072] As an optional implementation, step S202, during the vehicle's movement, acquires a trajectory point sequence of at least one target vehicle in the environment where the vehicle is located, including: during the vehicle's movement, acquiring an initial trajectory point sequence of the target vehicle; adjusting the initial trajectory point sequence using the coordinate type of the target vehicle to obtain a trajectory point sequence, wherein the coordinate type is used to indicate the type of coordinate system to which the target vehicle's location belongs, and the accuracy of the trajectory point sequence is greater than the accuracy of the initial trajectory point sequence.

[0073] In this embodiment, during the process of acquiring the trajectory point sequence of the target vehicle in its environment, an initial trajectory point sequence of the target vehicle can be obtained. The initial trajectory point sequence can be adjusted using the target vehicle's coordinate type to obtain the final trajectory point sequence. The initial trajectory point sequence can refer to the set of spatial location points of the target vehicle observed in continuous time frames, directly output by the vehicle and not processed by a coordinate system. The aforementioned coordinate type can be used to identify the classification attribute of the coordinate reference system upon which the spatial location information of the target vehicle is based. Its core function is to determine whether the target trajectory points possess global geolocation capabilities, thereby determining the processing strategy to be adopted for subsequent coordinate transformation. The coordinate type can be a binary classification state, divided into "with absolute positioning" and "without absolute positioning," both used to achieve spatial consistency and uniformity of the trajectory point sequence.

[0074] If the coordinate type is "absolute positioning," it indicates that the target vehicle's position information comes from a comprehensive positioning system with centimeter-level accuracy. The original coordinates of the target vehicle are latitude and longitude in the World Geodetic System 1984 (WGS84) or the original UTM coordinates. In this case, the goal of coordinate transformation is to unify the above coordinates to the UTM Cartesian coordinate system to eliminate the influence of Earth's curvature and projection distortion on trajectory fitting accuracy and ensure the alignment of multiple vehicle trajectories in global space. If the coordinate type is "no absolute positioning," it indicates that the target vehicle is not connected to a high-precision external positioning source. The target vehicle's position information can be calculated based on kinematic models and relative sensing data (e.g., ranging, angle measurement). In this case, the goal of coordinate transformation is to dynamically project the relative observations of each target vehicle through real-time pose (position, heading angle) to unify them into the vehicle's own coordinate system, forming absolute relative coordinates synchronized with the vehicle's pose.

[0075] Optionally, during vehicle operation, an initial trajectory point sequence of the target vehicle is acquired. This step relies on an onboard multi-sensor fusion system (including LiDAR, cameras, and millimeter-wave radar) to continuously detect and track moving targets in the surrounding environment. The system associates and maintains the identified target vehicle's trajectory across multiple consecutive perception frames, retaining stable targets with continuous tracking time of at least five frames to eliminate false detections and short-term interference. For each stable target, the system outputs its position coordinates in the current perception coordinate system (e.g., radar point cloud coordinates, image pixel projection coordinates, or polar coordinates of the millimeter-wave radar) and the corresponding timestamp, forming a set of discrete points arranged in chronological order, i.e., the initial trajectory point sequence. The coordinate representation of this sequence is not aligned with the vehicle's global pose; the values ​​can exist in the form of sensor local coordinates, vehicle-to-vehicle relative coordinates, or original geographic coordinates, without unified normalization, thus exhibiting spatial heterogeneity. As the original input for trajectory construction, the numerical accuracy of the initial trajectory point sequence is affected by sensor noise, occlusion, multipath effects, etc., and can only reflect the relative motion trend of the target vehicle, lacking direct engineering usability for trajectory fitting and optimization.

[0076] Optionally, the initial trajectory point sequence is adjusted using the target vehicle's coordinate type to obtain a new trajectory point sequence. The coordinate type is a binary classification identifier used to clarify the positioning reference system upon which the target vehicle's position information depends, divided into two categories: "with absolute positioning" and "without absolute positioning." When the coordinate type is "with absolute positioning," it indicates that the target vehicle has the capability of fusion positioning using the Global Navigation Satellite System (GNSS) and the Inertial Measurement Unit (IMU), with a positioning accuracy ≤0.1 meters and an update frequency ≥10Hz. In this case, the position information in the initial trajectory point sequence comes from the absolute position estimation of the target vehicle in the WGS84 coordinate system by the sensing system. The system uniformly transforms these absolute coordinates to the UTM Cartesian coordinate system to eliminate the influence of Earth's curvature and projection distortion on the trajectory geometry, ensuring that multiple target trajectories have accurate spatial alignment capabilities in global space.

[0077] When the coordinate type is "no absolute positioning," it indicates that no high-precision external positioning source is connected. In this case, the position information in the initial trajectory point sequence is the relative coordinates (e.g., distance, azimuth angle) output by the sensing system relative to the vehicle. This is based on the real-time speed, yaw rate, and heading angle obtained from the vehicle's chassis controller area network (CAN) bus, and is then analyzed using a kinematic model (x... t =x t-1 +v t ×cos(θ t )×Δt,y t =y t-1 +v t ×sin(θ t The system calculates the vehicle's pose change in the global environment using a method that calculates x × Δt. This transforms the relative observations of all target vehicles into absolute coordinates in the vehicle's coordinate system, with the rear axle center as the origin and the X-axis along the forward direction. This enables synchronous updates of trajectory points and vehicle pose. The x-axis is then used to calculate the vehicle's pose. t It can be used to represent the lateral position coordinates of the target vehicle in the vehicle coordinate system at time t; x t-1 It can be used to represent the lateral position coordinates of the target vehicle in the self-vehicle coordinate system at time t-1; v t It can be used to represent the speed of the target vehicle traveling in the direction of travel at time t; θ t It can be used to represent the heading angle of the target vehicle at time t; y t It can be used to represent the longitudinal position coordinates of the target vehicle in the vehicle coordinate system at time t; y t-1It can be used to represent the longitudinal position coordinates of the target vehicle in the vehicle coordinate system at time t-1; Δt can be used to represent the time interval between two consecutive trajectory point samplings.

[0078] The above process compensates for the vehicle's attitude drift and cumulative motion errors through dynamic coordinate correction, ensuring that the trajectory point sequence remains consistent with the vehicle's motion even without absolute positioning. The adjusted trajectory point sequence has a unified coordinate reference in spatial representation, and its position error is significantly lower than that of the initial trajectory point sequence, thus meeting the stringent data accuracy requirements of subsequent multi-trajectory fitting and optimization.

[0079] In this embodiment, through a three-level process of "raw data acquisition—type discrimination—scenario-based conversion," the trajectory point sequence of the target vehicle is systematically refined from the raw perception output into engineering-usable data with a unified coordinate reference and high spatial consistency, significantly improving the input quality for trajectory construction. This method, without increasing hardware costs, fully utilizes the differences in capabilities of existing perception and positioning systems, achieving adaptive support for both "absolute positioning" and "non-absolute positioning" scenarios, ensuring that the accuracy of the trajectory point sequence is superior to the original data in different environments. This process provides a stable, reliable, and high-precision input foundation for subsequent multi-trajectory fitting, optimization, and target driving trajectory generation, effectively supporting the continuous generation of vehicle reference lines in scenarios without maps or lane lines, enhancing the perception robustness and functional continuity of the intelligent driving system in complex road environments, and providing solid data support for achieving safe and smooth trajectory planning in all scenarios.

[0080] As an optional implementation, the initial trajectory point sequence is adjusted using the coordinate type of the target vehicle to obtain a trajectory point sequence, including: in response to the coordinate type being a geographic coordinate system, the initial trajectory point sequence is transformed from the geographic coordinate system to a projected coordinate system to obtain a transformed initial trajectory point sequence; in response to the coordinate type being the target vehicle's relative coordinate system, the initial trajectory point sequence is transformed from the relative coordinate system to the vehicle's own coordinate system to obtain a transformed initial trajectory point sequence, wherein the own coordinate system is constructed based on the vehicle's driving state information, and the transformed initial trajectory point sequence is corrected using the vehicle's driving scenario to obtain a trajectory point sequence, wherein the fit between the trajectory point sequence and the driving scenario is greater than the fit between the transformed initial trajectory point sequence and the driving scenario.

[0081] In this embodiment, during the adjustment of the initial trajectory point sequence using the target vehicle's coordinate type, if the coordinate type is a geographic coordinate system, the initial trajectory point sequence can be transformed from the geographic coordinate system to a projected coordinate system to obtain the initial trajectory point sequence after coordinate transformation. Conversely, if the coordinate type is the target vehicle's relative coordinate system, the initial trajectory point sequence can be transformed from the relative coordinate system to the vehicle's own coordinate system to obtain the initial trajectory point sequence after coordinate transformation. The driving scenario of the vehicle can be used to correct the initial trajectory point sequence after coordinate transformation to obtain the trajectory point sequence. The geographic coordinate system can refer to a spatial reference system defined based on the Earth ellipsoid model, using latitude and longitude as the basic unit of measurement, used to express the absolute geographical location of the target on the Earth's surface. This geographic coordinate system is implemented using WGS84, with its origin located at the Earth's center of mass. Longitude, latitude, and altitude constitute a three-dimensional spatial coordinate system, suitable for global positioning and navigation. A geographic coordinate system can be a geodetic coordinate system or a spherical coordinate system. Its core function is to provide a standardized spatial representation framework for the raw observations output by satellite positioning systems (such as GNSS).

[0082] Optionally, a projected coordinate system can refer to a system that maps spherical coordinates in a geographic coordinate system to a two-dimensional Cartesian coordinate system through mathematical projection transformation. This eliminates the interference of the Earth's curvature on the geometric relationships of local areas, achieving linear measurements of distance, angle, and area. This projected coordinate system is implemented using UTM (Ultra-Mercurimetric Mapping), which divides the Earth into 60 longitude zones. Within each zone, a transverse Mercator projection is used, forming a Cartesian coordinate system with east and north axes, measured in meters (m). A projected coordinate system can also be called a Cartesian coordinate system or an engineering coordinate system.

[0083] Optionally, a relative coordinate system can refer to a local coordinate system established with the sensing sensor or the vehicle as the origin, describing the relative position of the target vehicle. The origin and orientation of the relative coordinate system depend on the instantaneous pose of the sensing device or the vehicle and are not globally consistent. The relative coordinate system can use the measurement output of radar, camera, or millimeter-wave radar as a reference, expressing the position of the target vehicle using polar coordinates (distance, azimuth) or local Cartesian coordinates (x, y). The relative coordinate system can also be called a sensing local coordinate system or a sensor reference coordinate system, and its core function is to provide the instantaneous spatial relationship of the target vehicle relative to other vehicles.

[0084] Optionally, the vehicle coordinate system can refer to a local Cartesian coordinate system based on the vehicle itself and updated synchronously with the vehicle's motion state. The origin of the vehicle coordinate system is fixed at the center of the rear axle of the vehicle. The X-axis is positive along the vehicle's forward direction, the Y-axis is positive perpendicular to the X-axis pointing to the left, and the Z-axis is perpendicular to the ground and pointing upwards, forming a right-handed coordinate system. This coordinate system serves as a unified reference framework for vehicle dynamics modeling, trajectory planning, and control algorithms. Its direction and position change in real time with the vehicle's heading angle, velocity, and acceleration. The vehicle coordinate system can also be called the vehicle body coordinate system or motion reference coordinate system. Its core function is to unify the observation data of all external sensing targets (such as environmental vehicles) into a local space synchronized with the vehicle's motion, achieving spatial consistency between sensing data and control commands.

[0085] Optionally, the aforementioned driving status information can refer to a set of key physical parameters characterizing the vehicle's current motion state, collected in real time by the vehicle chassis CAN bus. This driving status information may include longitudinal speed, yaw rate, steering angle, acceleration, etc. The aforementioned driving scenario can refer to driving situation types with specific geometric and dynamic constraints, categorized based on road environment characteristics and vehicle dynamic behavior, used to guide the post-processing and adaptation correction of trajectory point sequences. Driving scenarios include, but are not limited to, straight-ahead scenarios, left-turn scenarios at intersections, right-turn scenarios at intersections, construction zone passage scenarios, and rural unmarked road scenarios. Each driving scenario has different trajectory geometric characteristics (such as rate of curvature change, steering angle range, and lateral acceleration requirements) and safety constraints (such as minimum turning radius and static obstacle distance threshold). Driving scenarios can be referred to as dynamic environment classification or task modes; their core function is to provide semantic context for trajectory optimization, enabling the converted trajectory point sequence to be specifically corrected according to the physical laws and safety requirements of the current scenario. In this technical solution, the driving scenario serves as the decision basis for the post-processing of the trajectory point sequence. By identifying the current scenario type, the corresponding dynamic verification rules are activated (such as relaxing the lateral acceleration threshold in the turning scenario and enhancing the trajectory smoothness in the straight-line scenario), thereby improving the adaptability of the trajectory point sequence to the actual driving needs and ensuring that the generated reference line meets the scenario-based control objectives in terms of geometric rationality, dynamic feasibility, and safety.

[0086] Optionally, in response to the coordinate type being geographic coordinate system, the initial trajectory point sequence is transformed from the geographic coordinate system to the projected coordinate system, resulting in the initial trajectory point sequence after coordinate transformation. The geographic coordinate system refers to a three-dimensional spatial representation system of latitude, longitude, and altitude based on WGS84. The data originates from an onboard system with GNSS+IMU combined positioning capabilities. The original trajectory point sequence consists of the absolute geographical location of the target vehicle output by the sensing module, with units of degrees, minutes, seconds, and meters. Since the geographic coordinate system is a spherical coordinate system, it lacks the linear distance and angle characteristics of a Cartesian coordinate system. Directly using it for trajectory fitting will introduce Earth curvature and projection distortion errors, affecting the smoothness of subsequent trajectory and the accuracy of curvature calculation. To eliminate this effect, the system uses the Universal Transverse Mercator (UTM) projection algorithm based on WGS84 coordinates to uniformly convert the latitude, longitude, and altitude coordinates of the target vehicle into easting and northing Cartesian coordinates in meters, forming the initial trajectory point sequence after coordinate transformation. This transformation process automatically selects the corresponding projection parameters based on the UTM zone number where the vehicle is located, ensuring that the length, angle, and area ratio of the local region maintain minimal distortion. This gives the trajectory point sequence geometric consistency in Euclidean space, providing a stable mathematical foundation for subsequent spline interpolation, curvature analysis, and dynamic verification. This step is only initiated when the system determines the coordinate type to be "absolutely positioned," and the output is a planar trajectory sequence that has eliminated the influence of Earth's curvature and has uniform units and coordinate axis directions.

[0087] Optionally, in response to the target vehicle's relative coordinate system, the initial trajectory point sequence is transformed from the relative coordinate system to the vehicle's own coordinate system, resulting in the transformed initial trajectory point sequence. The relative coordinate system refers to the set of local coordinates output by the onboard perception system (LiDAR, camera, millimeter-wave radar), with the sensor or the vehicle as the reference origin. It is expressed as distance, azimuth, or relative Cartesian coordinates and is not aligned with the vehicle's global pose, exhibiting instability due to drift with vehicle movement. Without GNSS positioning, the system cannot obtain absolute position and must perform dynamic coordinate calculations based on the vehicle's driving status information. Driving status information includes longitudinal vehicle speed and yaw rate, collected in real-time by the chassis CAN bus. The measurement errors of these two parameters are controlled within ±0.1 m / s and ±0.1° / s, respectively, with a sampling frequency of no less than 10 Hz, constituting the input parameters of the vehicle's motion model. The system is based on a vehicle coordinate system (origin at the center of the rear axle, X-axis along the forward direction, Y-axis perpendicular to the left). Using a kinematic integral model, it converts the relative observations of the target vehicle frame-by-frame into its absolute position in the vehicle coordinate system. This process uses the vehicle's attitude angle θ. t Using the direction reference, with vehicle speed v tTo determine the motion intensity, the sampling interval Δt is used as the integration step size to achieve synchronous evolution of the target trajectory point and the vehicle's motion. Although the resulting trajectory point sequence has a unified coordinate reference, its spatial representation is still a purely kinematic calculation result, without considering the influence of road geometric constraints, steering intentions, or dynamic obstacles. The shape of the initial driving trajectory may deviate from the actual road behavior, requiring further scene semantic correction.

[0088] Optionally, the initial trajectory point sequence after coordinate system transformation is corrected using the vehicle's driving scenario to obtain a new trajectory point sequence. The driving scenario refers to the dynamic behavior category of the vehicle in the current road environment, such as a straight-ahead scenario, a left-turn scenario at an intersection, a right-turn scenario at an intersection, a scenario of passing through an unmarked rural road, or a scenario of passing through a temporary construction area. The classification is based on the perception system's comprehensive judgment of road structure, traffic signs, obstacle distribution, and the trend of the target vehicle's heading angle change. In the straight-ahead scenario, the driving scenario constraints require the trajectory point sequence to have high smoothness and a low rate of curvature change. The system performs local curvature detection on the transformed trajectory, identifies abnormal jitter points, and performs smoothing correction through weighted averaging within a sliding window or low-pass filtering to suppress high-frequency fluctuations caused by sensor noise. In turning scenarios, driving scenario constraints require the trajectory to adapt to the vehicle's maximum turning capability. Based on a preset heading angle change rate threshold (≤20° / s) and minimum turning radius constraint (≥5m), the system identifies inflection point regions in the trajectory. Through piecewise interpolation and curvature redistribution strategies, it corrects local abrupt changes in the trajectory, ensuring vehicle dynamics feasibility while preserving the geometric features of the turning intention. In scenarios without road markings, such as rural roads or construction areas, driving scenarios emphasize trajectory avoidance of road edges or static obstacles. The system correlates the position and size of static obstacles such as curbs, concrete pillars, and cones output by the perception module, sets a minimum safe distance threshold (≥2m), and performs lateral offset correction on the trajectory point sequence to ensure it always remains within the safe passage area. The corrected trajectory point sequence is geometrically closer to real road traffic behavior, dynamically satisfies vehicle control constraints, and meets scenario-based obstacle avoidance requirements in terms of safety. Its adaptability to the driving scenario is significantly higher than the uncorrected initial trajectory point sequence after coordinate system transformation.

[0089] In this embodiment, the method described above achieves a systematic leap from raw perception data to highly adaptable engineering data through a three-level progressive process of "coordinate type discrimination—coordinate system transformation—scene semantic correction." The method autonomously selects either geographic-projection or relative-vehicle conversion paths based on coordinate type, ensuring a unified, stable, and distortion-free spatial representation regardless of whether absolute positioning is available or not. Furthermore, by incorporating semantic constraints of the driving scenario, the method performs dynamic geometric and dynamic corrections to the trajectory, significantly improving the physical rationality and engineering usability of the trajectory point sequence in actual driving behavior.

[0090] As an optional implementation, during the vehicle's operation, the initial trajectory point sequence of the target vehicle is obtained, including: during the vehicle's operation, the trajectory points of the target vehicle are collected according to the sampling strategy corresponding to the driving scenario, wherein the sampling strategy is used to represent the rules for controlling the vehicle to collect trajectory points according to the sampling frequency, and different sampling frequencies correspond to different driving scenarios; the target number of trajectory points are determined as the initial trajectory point sequence.

[0091] In this embodiment, during the acquisition of the initial trajectory point sequence of the target vehicle, trajectory points of the target vehicle can be collected according to the sampling strategy corresponding to the driving scenario. A target number of trajectory points can be determined as the initial trajectory point sequence. The sampling strategy refers to a predefined systematic rule used to control the timing and density of trajectory point collection during vehicle operation, based on the dynamic characteristics of the current driving scenario and the trajectory data requirements. This sampling strategy can dynamically adjust the periodic frequency of trajectory point collection using the driving scenario as input, ensuring sufficient data points representing the trajectory's geometric features and dynamic behavior are obtained under different driving situations, while avoiding increased computational load due to redundant sampling. The sampling strategy can include the setting of the sampling frequency and the scenario mapping relationship. The sampling frequency refers to the number of times the target vehicle's position and attitude information is collected per unit time, measured in Hertz (Hz), i.e., the number of trajectory points collected per second. Different driving scenarios correspond to different sampling frequency requirements to match the dynamic change rate and trajectory complexity. The target number can be a pre-set number, for example, 30.

[0092] Optionally, during vehicle operation, trajectory points of the target vehicle are collected according to the sampling strategy corresponding to the driving scenario. The sampling strategy is a systematic rule used to control the vehicle to collect trajectory points at a specific sampling frequency. Essentially, it is a dynamic response mechanism based on driving scenario classification. Its core lies in automatically matching the optimal trajectory point collection frequency based on the complexity and dynamic change rate of the current driving scenario. Driving scenarios include straight-ahead scenarios, left-turn scenarios at intersections, right-turn scenarios at intersections, rural unmarked roads, and temporary construction zone passage scenarios, etc. Each type of scenario has different geometric and dynamic characteristics, and the trajectory point change rate and information density requirements differ significantly. The system uses a perception module to jointly analyze the target vehicle's heading angle change rate, lateral acceleration, trajectory curvature evolution trend, and environmental semantics (such as missing lane lines, traffic signs, and road boundary structures) to output the current driving scenario type.

[0093] For example, when a driving scenario is identified as a straight-ahead scenario, it is determined that the scenario has a low rate of dynamic change and the trajectory tends to be linear. In this case, the sampling strategy adopts a low-frequency acquisition mode, setting the sampling frequency to 10 Hz, that is, acquiring the position and heading angle information of the target vehicle once every 0.1 seconds. This frequency is sufficient to fully represent the trajectory of uniform or gradually changing motion, while avoiding redundant data occupying computing resources. When a driving scenario is identified as a left or right turn at an intersection, the system determines that the scenario has significant rapid heading angle shift, abrupt change in trajectory curvature, and lateral displacement acceleration. In this case, the sampling strategy switches to a high-frequency acquisition mode, increasing the sampling frequency to 20 Hz, that is, acquiring trajectory points once every 0.05 seconds, to ensure that the trajectory geometric details of the turning start, inflection point, and completion stages are completely preserved. This frequency adjustment mechanism is not a manually preset hard-coded logic, but a conditional response behavior triggered in real time based on the driving scenario classification results. Its execution depends on the linkage between the output of the scene recognition module and the sampling scheduling module, ensuring that the acquisition frequency is always consistent with the dynamic needs of the driving scenario, thereby achieving intelligent, adaptive, and resource-optimized data acquisition.

[0094] Optionally, the target number of trajectory points can be determined as the initial trajectory point sequence. The target number refers to the minimum number of effective trajectory points set to ensure the integrity of trajectory modeling and the stability of fitting. The value is determined comprehensively based on the geometric complexity of the driving scenario and the data requirements of the fitting algorithm.

[0095] For example, in straight-ahead scenarios, due to the gentle trajectory changes, the system sets a target of at least 30 continuous trajectory points. This number is sufficient to support the requirements of 3rd-order B-splines or 3rd-order polynomial fitting for curve smoothness and convergence. In intersection turning scenarios, due to the high curvature and nonlinear characteristics of the trajectory, the system also sets a target of at least 30 continuous trajectory points. However, the acquisition time must cover the entire turning action from start to finish to ensure that the trajectory includes the complete turning arc and avoid fitting anomalies or loss of inflection points due to insufficient points. During the acquisition process, the system continuously monitors the tracking status of the target vehicle, retaining only trajectory point sequences with continuous tracking time of no less than 3 seconds, without interruption or loss, and discarding breakpoint data caused by occlusion, false detection, or tracking failure. When the number of acquired trajectory points reaches the target number and meets the continuity and stability conditions, the system marks this sequence as the initial trajectory point sequence, which serves as the input for subsequent multi-trajectory fitting modules. This determination process does not rely on arbitrary truncation or random sampling, but uses the minimum data volume constraint and time continuity as the sole criterion to ensure that each initial trajectory point sequence has a complete motion evolution process and sufficient geometric information density, providing a reliable data foundation for subsequent adaptive fitting and dynamic verification.

[0096] In this embodiment, the above method dynamically adjusts the sampling frequency according to the driving scenario, ensuring resource conservation in low-dynamic scenarios and complete information in high-dynamic scenarios, significantly improving the efficiency and effectiveness of trajectory data collection; at the same time, by setting clear target quantity and continuity constraints, it ensures that each initial trajectory point sequence has sufficient data volume and completeness, avoiding fitting failure due to insufficient points or discontinuity.

[0097] As an optional implementation, step S204, determining the initial driving trajectory of the target vehicle based on the trajectory point sequence of the target vehicle, includes: calling a fitting strategy adapted to the density of the trajectory point sequence to perform fitting processing on the trajectory point sequence to obtain the initial driving trajectory, wherein the density is used to represent the number of trajectory points per unit distance in the trajectory point sequence, and the fitting strategy is used to represent the rules for fitting the trajectory point sequence.

[0098] In this embodiment, during the process of determining the initial driving trajectory based on the trajectory point sequence, a fitting strategy adapted to the density of the trajectory point sequence can be invoked to fit the trajectory point sequence and obtain the initial driving trajectory. Here, density refers to a quantitative indicator of the number of trajectory points per unit distance in the trajectory point sequence, used to characterize the spatial density of the trajectory points. The aforementioned fitting strategy can refer to a systematic rule for selecting and executing a specific mathematical modeling method for the trajectory point sequence, its function being to transform the discrete trajectory point sequence into a continuous, smooth, and differentiable curve expression, i.e., the initial driving trajectory. This strategy has adaptive characteristics, automatically matching a suitable fitting algorithm based on the density characteristics of the trajectory point sequence. The aforementioned fitting strategy can be an adaptive fitting algorithm.

[0099] Optionally, a fitting strategy adapted to the density of the trajectory point sequence is invoked to fit the trajectory point sequence. The density of the trajectory point sequence is defined as the number of trajectory points contained within a unit distance. It is calculated by sliding a window with a step size of 1 meter along the trajectory path, counting the number of trajectory points in each window, and finally obtaining the density distribution of the entire trajectory. This density is a core indicator for evaluating the spatial sampling quality of trajectory data and can be used to reflect the complexity of the target vehicle's motion state and the richness of information in the perceived data.

[0100] In the embodiments of this application, the above method dynamically selects two optimal fitting methods, namely 3rd-order B-spline or weighted least squares, based on the objective and quantifiable physical index of trajectory point density, to deal with high-density and low-density trajectory scenarios respectively, effectively avoiding underfitting, overfitting or computational instability problems that occur when a single algorithm has uneven data distribution.

[0101] As an optional implementation, the fitting strategy includes a first fitting strategy, which represents the rule for fitting the trajectory point sequence according to a preset spline curve. The first fitting strategy calls a fitting strategy adapted to the density of the trajectory point sequence to fit the trajectory point sequence and obtain an initial driving trajectory. This includes: in response to a density greater than or equal to a density threshold, determining nodes based on the order of the preset spline curve and the number of trajectory points, and determining control vertices based on the coordinates of the trajectory points; calling the first fitting strategy to fit the nodes and control vertices according to the preset spline curve to obtain the initial driving trajectory; the method further includes: in response to a fitting error of the initial driving trajectory determined according to the first fitting strategy being greater than a fitting error threshold, adjusting the interval between nodes; and refitting the initial driving trajectory using the nodes with adjusted intervals until the fitting error is less than or equal to the fitting error threshold.

[0102] In this embodiment, during the process of fitting the trajectory point sequence to obtain the initial driving trajectory, if the density is greater than or equal to a density threshold, nodes can be determined based on the order of the preset spline curve and the number of trajectory points. Control vertices can also be determined based on the coordinates of the trajectory points. A first fitting strategy can be invoked to fit the nodes and control nodes according to the preset spline curve to obtain the initial driving trajectory. If the fitting error of the initial driving trajectory determined by the first fitting strategy is greater than a fitting error threshold, the interval between nodes can be adjusted, and the initial driving trajectory can be refitted using the nodes with the adjusted intervals until the fitting error is less than or equal to the fitting error threshold.

[0103] The first fitting strategy can refer to the mathematical modeling rules used to transform a sequence of trajectory points into a continuous smooth curve, with the core being a parametric curve fitting method based on B-splines. This first fitting strategy can be a B-spline fitting strategy. This first fitting strategy can be used in high-density trajectory point scenarios. By controlling the collaborative construction of vertices and node vectors, it achieves a G2 continuous expression of the trajectory, ensuring a smooth transition of position, velocity, and acceleration. This is a key modeling method for ensuring vehicle driving comfort and control stability. The preset spline curve can refer to a predefined 3rd-order B-spline curve, mathematically a piecewise polynomial function, possessing local controllability and high-order continuity. This preset spline curve can be jointly defined by node vectors and control vertices. The density threshold can refer to a quantitative standard used to determine whether the spatial distribution of trajectory points is sufficient, set to 2 points / meter, meaning that each meter of trajectory length contains at least two trajectory points. This threshold is the dividing line between high-density and low-density trajectories. When the trajectory density meets or exceeds this value, it is considered to have sufficient spatial sampling capability, enough to support fine modeling of curvature changes, thereby triggering a high-precision B-spline fitting strategy.

[0104] Optionally, the aforementioned nodes can refer to the ordered parameter sequence defining the segmented intervals and basis function support domains in the B-spline curve. Their number is determined by the number of trajectory points and the curve order, calculated as: Number of nodes = Number of trajectory points + Order + 1. The node spacing determines the local flexibility of the curve; a uniform distribution ensures computational stability. Their function is to divide the curve into segmented control regions, providing a mathematical framework for the interpolation and solution of control vertices. The aforementioned control vertices can refer to the geometric control points in the B-spline curve used to define the overall shape of the curve. The coordinates of the control vertices can be obtained by inversely calculating the measured coordinates of the trajectory points using the least squares method. The control vertices are not directly located on the trajectory, but through coupling with the node weight functions, they jointly generate a smooth curve that approximates the original trajectory.

[0105] Optionally, in response to a density greater than or equal to a density threshold, nodes are determined based on the order of the preset spline curve and the number of trajectory points, and control vertices are determined based on the coordinates of the trajectory points. The density threshold is 2 nodes / meter. When the spatial density of the trajectory point sequence meets or exceeds this standard, it is determined that it has sufficiently dense geometric sampling capability and meets the prerequisite for performing high-precision B-spline fitting. At this time, the system determines the total number of nodes according to the preset spline curve order of 3, combined with the total number of trajectory points, and according to the node number calculation formula (number of nodes = number of trajectory points + order + 1). The nodes are the parameterized support sequence of the B-spline curve, which are uniformly distributed in the parameter space and used to divide the local domain of the curve. Their distribution density is directly bound to the number of trajectory points, ensuring that each trajectory point has a corresponding basis function to participate in the interpolation. At the same time, the system collects the measured coordinate values ​​of the trajectory points in a unified coordinate system as constraints, and solves for the set of control vertices that minimizes the fitting error by constructing a least-squares optimization equation, using the trajectory point coordinates as observations and the control vertices as variables to be solved.

[0106] Optionally, the first fitting strategy is invoked to fit the nodes and control vertices to obtain the initial driving trajectory according to a preset spline curve. The first fitting strategy is a 3rd-order B-spline fitting method, the mathematical essence of which is to generate a continuous curve by using the node vector and the set of control vertices through a linear combination of B-spline basis functions. The curve is divided into segments in the parameter space according to the nodes, and each segment is jointly influenced by four adjacent control vertices, ensuring that the curve achieves continuity in position, first derivative, and second derivative at the connection points, i.e., G2 continuity. Based on the node sequence and the coordinates of the control vertices, the system calculates the coordinate values ​​corresponding to each parameter on the curve point by point, generating a continuous and differentiable parametric curve. This curve is the initial driving trajectory of the target vehicle, which realistically approximates the original trajectory points in space, while satisfying the smoothness requirements of vehicle dynamics.

[0107] Optionally, in response to the fitting error of the initial driving trajectory determined according to the first fitting strategy exceeding the fitting error threshold, the interval between nodes is adjusted. Using the nodes with the adjusted interval, the initial driving trajectory is refitted until the fitting error is less than or equal to the fitting error threshold. The fitting error threshold is set to 0.3 meters to constrain the geometric accuracy of trajectory reconstruction. If the root mean square error of the initial fitting result exceeds this threshold, the system determines that the current node interval (default 0.5 meters) fails to adequately adapt to local curvature changes in the trajectory, resulting in suboptimal fitting. At this time, the system triggers an error correction mechanism, reducing the node interval from 0.5 meters to 0.4 meters, making the nodes denser in the parameter domain, thereby enhancing the resolution and responsiveness of the B-spline basis function to local trajectory features. After adjusting the node interval, the system recalculates the node vector, keeping the total number of nodes unchanged, compressing only the spacing between adjacent nodes, and then reconstructs the basis function matrix. Based on the same initial control vertex values ​​or re-optimizing the control vertices, B-spline fitting is performed again. The above process can be executed cyclically, with each adjustment focusing solely on reducing the node interval, until the fitting error converges to within 0.3 meters.

[0108] In this embodiment, the above method uses a density threshold as the basis for algorithm switching to ensure that the high-precision B-spline method is enabled when there is sufficient data; it uses a node number formula and control vertex optimization to ensure the mathematical standardization of modeling; and it uses dynamic adjustment of node interval as an error compensation means to break through the fitting limitations of fixed node distribution and significantly improve the reconstruction capability of complex curvature segments.

[0109] As an optional implementation, the fitting strategy includes a second fitting strategy, which represents the rule for fitting the trajectory point sequence according to a preset polynomial. The second fitting strategy calls a fitting strategy adapted to the density of the trajectory point sequence to fit the trajectory point sequence and obtain an initial driving trajectory. This includes: determining the weights of the trajectory points in response to a density less than a density threshold; determining the polynomial coefficients of the preset polynomial based on the weights; and calling the second fitting strategy to fit the initial driving trajectory according to the polynomial coefficients. The method further includes: adjusting the weights in response to a fitting error of the initial driving trajectory determined according to the second fitting strategy exceeding a fitting error threshold; and refitting the initial driving trajectory using the adjusted weights until the fitting error is less than or equal to the fitting error threshold.

[0110] In this embodiment, during the process of fitting the trajectory point sequence to obtain the initial driving trajectory, if the density is less than a density threshold, the weights of the trajectory points can be determined. Based on these weights, the polynomial coefficients of a preset polynomial can be determined. A second fitting strategy can be invoked to fit the initial driving trajectory based on the polynomial coefficients. If the fitting error of the initial driving trajectory determined by the second fitting strategy is greater than a fitting error threshold, the weights can be adjusted, and the adjusted weights can be used to refit the initial driving trajectory until the fitting error is less than or equal to the fitting error threshold. The second fitting strategy can refer to a trajectory modeling method based on the weighted least squares principle, used in sparse scenarios where the trajectory point density is lower than the density threshold; it can be called a weighted least squares fitting strategy. The core of the second fitting strategy is to strengthen the fitting contribution of key regions and weaken the interference of outliers or low-confidence points by assigning non-uniform weights to each trajectory point, thereby generating a smooth and reasonable trajectory expression even under conditions of insufficient data. The aforementioned predefined polynomial can refer to a cubic polynomial function (e.g., y=ax³+bx²+cx+d) as the basic fitting model. This model is concise, computationally efficient, and possesses sufficient degrees of freedom to characterize nonlinear trajectory trends. This polynomial does not rely on higher-order terms or complex basis functions; the curve shape is determined by the polynomial coefficients a, b, c, and d.

[0111] Optionally, in response to a density less than a density threshold, the weights of trajectory points are determined; based on these weights, the polynomial coefficients of a preset polynomial are determined. The density threshold is 2 points / meter. When the number of points per unit distance in the trajectory point sequence is lower than this standard, the system determines that the data is sparse and cannot support high-order curve fitting, and switches to a second fitting strategy. In this case, the system first calculates the vertical distance from each point to the initial trajectory based on the initial linear fit or coarse geometric inference, using this distance as the basis for weight allocation. The weight allocation rule is in the form of the inverse of distance, that is, the weight of a trajectory point is equal to 1 divided by the sum of the distance from that point to the initial trajectory and a small constant ε (ε=0.01), ensuring that points closer to the initial trajectory have higher weights and points farther away have lower weights, thus prioritizing the retention of observation points closer to the actual motion path. After the weights are determined, the system uses the coordinates of all trajectory points and their corresponding weights as input to construct a weighted least squares optimization model. The objective function is to minimize the weighted sum of squared residuals. A pre-defined polynomial (y=ax³+bx²+cx+d) is used as the fitting function. The polynomial coefficients a, b, c, and d are solved through matrix operations to make the generated curve best approximate the original trajectory points in a weighted sense. This coefficient calculation process involves the inversion of a linear system, and the result is the unique output of this strategy, forming the mathematical basis for subsequent trajectory reconstruction.

[0112] Optionally, a second fitting strategy is invoked to fit the initial driving trajectory based on the polynomial coefficients. The second fitting strategy uses the solved four polynomial coefficients as unique parameters, substitutes them into a preset polynomial function, and calculates the corresponding ordinate values ​​within the horizontal coordinate interval covered by the trajectory points at fixed step sizes, generating a continuous, differentiable three-dimensional spatial curve projection. This curve is a univariate cubic polynomial expression in a two-dimensional plane, possessing overall smoothness and local adaptability. Under sparse point conditions, it can effectively suppress oscillations and overfitting. Its shape is determined by the coefficients, does not depend on interpolation points, and only reflects the overall trend of the trajectory.

[0113] Optionally, in response to the fitting error of the initial driving trajectory determined according to the second fitting strategy exceeding the fitting error threshold, the weights are adjusted; using the adjusted weights, the initial driving trajectory is refitted until the fitting error is less than or equal to the fitting error threshold. The fitting error threshold is set to 0.3 meters to constrain the geometric accuracy of trajectory reconstruction. If the root mean square error of the first fitting result exceeds this value, the system determines that the current weight allocation has failed to effectively focus on the key trajectory region, possibly due to initial estimation bias leading to inaccurate weights. At this time, the system uses the current fitted curve as a reference to recalculate the vertical distance from each trajectory point to the curve, and recalculates the weights based on the updated distances, assigning higher weights to points closer to the new curve, forming a closed-loop iterative mechanism of "fitting—error analysis—weight update—refitting". This process is repeated, with each iteration using the updated weights as input to resolve the polynomial coefficients until the fitting error converges to within 0.3 meters.

[0114] In the embodiments of this application, the above method effectively solves the problems of inaccurate trajectory modeling, oscillation, and extrapolation distortion under low sampling density. Its core innovation lies in introducing a distance inverse weighting mechanism, which makes the fitting process naturally focus on high-confidence observation points, significantly improving the robustness of the trajectory in scenarios with limited perception information, such as rural roads and construction areas.

[0115] As an optional implementation, the executability index includes at least one of the following: a smoothness index, a safety index, a dynamic index, and a consistency index. The consistency index is used to represent the degree of difference between the initial driving trajectory and the vehicle's driving trend. Adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory includes at least one of the following: responding to the curvature of the initial driving trajectory being greater than a curvature threshold corresponding to the driving scenario in which the vehicle is located, adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein curvature is used to represent the smoothness of the initial driving trajectory, the curvature of the adjusted initial driving trajectory is less than or equal to the curvature threshold, and the smoothness index of the adjusted initial driving trajectory is greater than the smoothness index of the initial driving trajectory before adjustment; responding to the distance between the initial driving trajectory and obstacles around the vehicle being less than a distance threshold, adjusting the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the distance between the adjusted initial driving trajectory and obstacles is greater than or equal to the distance threshold, and the smoothness index of the adjusted initial driving trajectory is greater than the smoothness index of the initial driving trajectory before adjustment. The safety index is greater than that of the initial driving trajectory before adjustment. In response to the target driving scenario, the dynamic constraints allowed by the target driving scenario are invoked to adjust the initial driving trajectory, resulting in an adjusted initial driving trajectory. The target driving scenario indicates the dynamic range of vehicle movement, which is greater than that of other driving scenarios. The fit between the adjusted initial driving trajectory and the target driving scenario is greater than the fit between the initial driving scenario before adjustment and the target driving scenario. Furthermore, the dynamic index of the adjusted initial driving trajectory is greater than that of the initial driving trajectory before adjustment. In response to a difference greater than a difference threshold, the initial driving trajectory is adjusted again, resulting in an adjusted initial driving trajectory. The driving trend represents the vehicle's driving state at future times. The difference between the adjusted initial driving trajectory and the driving trend is less than or equal to the difference threshold. Finally, the consistency index of the adjusted initial driving trajectory is greater than that of the initial driving trajectory before adjustment.

[0116] In this embodiment, during the adjustment of the initial driving trajectory, if the curvature of the initial driving trajectory is greater than the curvature threshold corresponding to the driving scene in which the vehicle is located, the initial driving trajectory is adjusted. If the distance between the initial driving trajectory and obstacles around the vehicle is less than a distance threshold, the initial driving trajectory can be adjusted. If the driving scene is a target driving scene, the dynamic constraints allowed by the target driving trajectory can be invoked to adjust the initial driving trajectory. If the difference is greater than a difference threshold, the initial driving trajectory can be adjusted to obtain an adjusted initial driving trajectory.

[0117] Optionally, in response to the curvature of the initial driving trajectory exceeding the curvature threshold corresponding to the driving scenario, the initial driving trajectory is adjusted to obtain an adjusted initial driving trajectory. Curvature characterizes the degree of bending of the trajectory in space, and its rate of change reflects the smoothness of the driving path, being a core component of the smoothness index. When the system detects that the curvature value of a certain segment of the trajectory exceeds a preset threshold (e.g., the rate of change of curvature exceeds 0.15 rad / m in a turning scenario), it determines that the segment has excessive bending, which may lead to sudden changes in vehicle lateral acceleration, decreased ride comfort, or unstable control. The system then initiates local smoothing optimization in that area, employing a quadratic programming algorithm with the minimization of the rate of change of curvature as the objective function and the trajectory point position as the constraint condition. Within the allowable geometric offset range, the control vertices or polynomial coefficients are fine-tuned to gradually reduce the curvature of that segment until the curvature threshold requirement is met. After adjustment, the overall curvature continuity of the trajectory is improved, the rate of change of curvature is reduced, and the smoothness index increases accordingly, ensuring that the acceleration changes smoothly when the vehicle travels along the trajectory, meeting the engineering specifications for driving comfort and control stability.

[0118] Optionally, in response to the initial driving trajectory being less than a distance threshold between it and obstacles around the vehicle, the initial driving trajectory is adjusted to obtain an adjusted initial driving trajectory. Obstacles include static structures such as curbs, concrete pillars, construction barriers, and traffic signs, whose positions and dimensions are output in real time by the perception system and converted to a unified coordinate system. The distance threshold is set to 2 meters as the minimum safe distance between the vehicle and static obstacles. When the system detects that the distance between a point on the trajectory and any static obstacle is less than this threshold, it determines that there is a collision risk and the safety index fails to meet the standard. The system then, within the allowable lateral adjustment range of the trajectory, uses the obstacles as constraints and locally offsets the trajectory in a direction perpendicular to the trajectory to ensure that the closest distance between the trajectory and all obstacles is greater than or equal to 2 meters. The offset process follows the principle of minimum disturbance, making fine adjustments only in necessary areas to avoid introducing new curvature abrupt changes. After adjustment, the safe distance between the trajectory and obstacles is restored to the compliant range, the safety index is improved, and the vehicle's driving safety is guaranteed in environments without lane markings or complex environments.

[0119] Optionally, in response to the target driving scenario, the system invokes the dynamic constraints allowed by the target driving scenario to adjust the initial driving trajectory, resulting in an adjusted initial driving trajectory. The target driving scenario refers to a high-dynamic action area where the vehicle is turning left or right at an intersection. Such scenarios have a higher tolerance for lateral acceleration, steering angle, and steering radius than straight-ahead scenarios. The system identifies the current scenario as a high-dynamic scenario by recognizing the rate of change of heading angle, trajectory curvature trend, and environmental semantic information. At this time, the system invokes preset dynamic constraints: the maximum steering angle does not exceed ±35°, the maximum lateral acceleration is relaxed to ±2.5m / s², and the minimum steering radius is not less than 5 meters (calculated based on a wheelbase of 2.8 meters). If the steering angle change of the initial trajectory exceeds this range, or the lateral acceleration exceeds the limit, or the radius corresponding to the curvature is less than 5 meters, the system performs segmented reconstruction within the target segment of the trajectory. By adjusting the control vertices or polynomial coefficients, the system reduces the curvature gradient and slows down the steering rate without disrupting the trajectory continuity, so that the trajectory meets the dynamic constraints of the target scenario. After the adjustment, the adaptability of the trajectory to high-dynamic scenarios is significantly improved, the dynamic performance indicators are enhanced, and the vehicle can complete safe and controllable steering operations at complex intersections.

[0120] Optionally, in response to a difference exceeding a difference threshold, the initial driving trajectory is adjusted to obtain an adjusted initial driving trajectory. The difference is used to quantify the degree of deviation between the initial driving trajectory and the actual driving trend of the target vehicle. The driving trend refers to the direction of motion, speed, and steering intention of the environmental vehicle in the following seconds, inferred from the rate of change of heading angle, speed trend, and trajectory direction of its continuously tracked trajectory. The difference threshold is set as follows: heading angle deviation not exceeding 8°, and lateral trajectory offset not exceeding 0.8 meters. When the system detects that the initial trajectory and the driving trend of any target vehicle exceed the above thresholds, it determines that the trajectory fails to effectively reflect the reasonable path selection of the environmental vehicle, and the consistency index is too low. The system then uses the trajectory trend of the target vehicle as a reference benchmark and performs direction correction within a local range of the trajectory. By translating or rotating the trajectory segment, it ensures that the heading angle change trend is consistent with the target vehicle, and the lateral offset is compressed to within the threshold. After adjustment, the consistency between the trajectory and the behavior trend of the environmental vehicle is improved, the consistency index is enhanced, ensuring that the generated reference trajectory has realistic followability and improving the selection reliability of the subsequent trajectory planning module.

[0121] In this embodiment, the method significantly enhances the engineering usability of the initial driving trajectory in complex, unstructured scenarios by independently verifying and collaboratively improving four types of executability indicators: smoothness, safety, dynamics, and consistency. Each adjustment is based on a clear physical threshold and aims to improve the indicators, avoiding blind optimization and ensuring that each correction is interpretable and controllable.

[0122] As an optional implementation, the method further includes at least one of the following: in response to the target vehicle being in a failed state, extending the initial driving trajectory based on the trajectory points of the target vehicle before it became failed, to obtain an extended initial driving trajectory, wherein the driving length of the extended initial driving trajectory is greater than the driving length of the initial driving trajectory before it was extended, and / or, the driving time of the extended initial driving trajectory is greater than the driving time of the initial driving trajectory before it was extended; in response to the target vehicle satisfying one of the following conditions, generating a target driving trajectory based on the vehicle's map information and the vehicle's perception information: the number of target vehicles is less than one; the target vehicle is in a failed state, and there are no target vehicles in a normal state within a preset time period; in response to the number of target vehicles in a normal state being greater than one within a preset time period, determining the initial driving trajectory using the trajectory point sequence of the target vehicles in a normal state.

[0123] In this embodiment, if the target vehicle is in a failed state, the initial driving trajectory can be extended based on the trajectory points of the target vehicle before the failed state to obtain an extended initial driving trajectory. If the target vehicle falls under any of the following conditions, a target driving trajectory can be generated based on the vehicle's map and perception information: the number of target vehicles sensed by the vehicle is less than one; the target vehicle is in a failed state, and no target vehicle in a normal state exists within a preset time period. If the number of target vehicles in a normal state within the preset time period is greater than one, the initial driving trajectory can be determined using the trajectory point sequence of the target vehicles in a normal state. Here, a failed state can refer to an operational state where the continuity and reliability of the trajectory data of an environmental vehicle are lost due to tracking interruption, abnormal movement, or exceeding the perception field of view. This failed state can also be called a tracking failed state. The extended processing can refer to an emergency operation where, after the target vehicle enters a failed state, the initial driving trajectory is geometrically continued using a linear extrapolation method based on the trajectory points of the last 15 frames before the target vehicle's failure. The preset time period can refer to a fixed time window, set to 1 second, allowing waiting for new, valid target vehicles to recover after all target vehicles are in a failed state.

[0124] Optionally, in response to a target vehicle being in a failed state, the initial driving trajectory is extended based on the trajectory points of the target vehicle before it became failed, resulting in an extended initial driving trajectory. A failed state refers to a state where the target vehicle's trajectory data continuity is interrupted due to tracking loss, unstable movement, or deviation from the sensor's field of view, rendering it unusable as a valid reference source. When the system determines that a target vehicle has entered this state, a short-term trajectory extension mechanism is immediately activated. This mechanism uses linear extrapolation based on the last 15 frames of trajectory points before the vehicle's failure, geometrically extending along its final movement direction and velocity trend. The extension operation does not introduce new sensing data; it only relies on the linear trend at the end of the historical trajectory to ensure that the trajectory extends continuously in space, with an extension distance not exceeding 5 meters and an extension time not exceeding 1 second. The extended initial driving trajectory has a longer driving length and duration than the original trajectory, forming a brief buffer period. This provides a time window for the system to re-select new targets, avoiding a precipitous interruption of the reference trajectory due to the instantaneous failure of a single target.

[0125] Optionally, in response to a target vehicle meeting one of the following conditions, a target driving trajectory is generated based on the vehicle's map information and perception information: the number of target vehicles is less than one, or the target vehicle is in a disabled state and no target vehicle in a normal state exists within a preset time period. The preset time period is 1 second, which is the decision threshold for the system to determine that environmental vehicles are unavailable for an extended period. When the system detects that all environmental vehicles are in a disabled state and fails to re-identify any target vehicle in a normal state that meets stable conditions within 1 second, the system determines that the primary reference source has completely failed and immediately activates an emergency redundancy strategy. This strategy calls on the vehicle's map information and static structural information such as road boundaries, curbs, and traffic signs output by the perception module, combined with the vehicle's positioning status, to construct a safe driving channel under geometric constraints in a scenario without lane lines. The system generates a reference trajectory that conforms to driving safety regulations based on the road topology and obstacle distribution. Its form is generated based on the fusion of map semantics and perception boundaries, and does not depend on the behavior of environmental vehicles. It is a function maintenance means in the degraded mode, ensuring that the intelligent driving system still has an executable trajectory benchmark in extreme situations where there are no reference targets.

[0126] Optionally, in response to the number of target vehicles in a normal state being greater than one within a preset time period, the initial driving trajectory is determined using the trajectory point sequence of the target vehicles in a normal state. A normal state refers to a stable tracking target that meets the following conditions: continuous tracking time ≥ 3 seconds, speed and heading angle change rate within the scene adaptation threshold, distance to the target vehicle within 2 to 60 meters, and no reverse driving or abnormal behavior. If ≥ 2 target vehicles in a normal state are detected again within 1 second, it is determined that the main reference source has been restored, the backup trajectory generation process is immediately terminated, and the main process reconstruction is initiated. The system processes the trajectory point sequence of each target vehicle in a normal state according to preset sampling rules and coordinate transformation strategies, and autonomously selects a weighted least squares or B-spline fitting strategy based on the trajectory density to generate multiple initial driving trajectories. This process fully reuses the fitting and optimization modules of the main process, ensuring that the newly generated trajectory is highly consistent with environmental behavior and has realistic followability, thereby achieving a smooth switch from redundant mode to high-precision main mode of the reference trajectory.

[0127] In this embodiment, the method described above systematically addresses the risk of reference line interruption caused by environmental vehicle tracking fluctuations in scenarios without or with weak maps through a three-stage collaborative approach of "short-term extension—map backup—target recovery." The extension process achieves millisecond-level continuity maintenance, avoiding abrupt control changes; the backup trajectory generated by map and perception fusion provides underlying safety redundancy, ensuring no functional degradation; and the rapid reconstruction mechanism after new target recovery guarantees that system performance returns to optimal levels. This mechanism requires no additional hardware, is implemented based on existing perception and positioning capabilities, and possesses engineering deployability and scenario generalization capabilities.

[0128] The technical solutions of the embodiments of this application will be illustrated below with reference to preferred embodiments.

[0129] Currently, intelligent driving reference lines, as the core benchmark for trajectory planning and control, have formed a multi-source data-supported generation technology system. Their core applications include providing trajectory planning benchmarks, coordinate transformation and environmental modeling, decision-making and control support, and multi-system collaborative integration. Existing reference line generation mainly relies on four technical paths: First, extracting and smoothing lane centerlines based on high-precision maps, suitable for structured roads such as highways; second, extracting global paths based on navigation maps, adapting to long-distance scenarios without high-precision maps; third, generating virtual centerlines based on visual perception of lane lines, used in scenarios without maps or with temporary road changes; and fourth, integrating multi-source data such as maps, perception, and navigation to improve adaptability to complex environments. Spline interpolation and quadratic programming smoothing algorithms are commonly used in the generation process, combined with dynamic updates of the reference lines based on vehicle status to ensure trajectory continuity. These technologies have been widely applied in the trajectory planning modules of various levels of autonomous driving systems.

[0130] Among related technologies, intelligent driving reference line technology still has many defects and shortcomings: First, single-source solutions have significant limitations. High-precision map-dependent solutions are limited by coverage and have high update costs, making them unsuitable for dynamic scenarios such as road construction and rerouting; navigation map-based solutions lack accuracy and cannot support high-precision trajectory planning; visual perception-based solutions are easily affected by lighting and weather, resulting in poor stability. Second, multi-source fusion solutions face a trade-off between efficiency and robustness. Complex data verification and optimization algorithms lead to insufficient real-time performance, making it difficult to match the dynamic response requirements of high-speed driving scenarios. Third, they lack adaptability to special scenarios. In scenarios such as unstructured roads without lane lines and complex intersections, reference line extraction or adjustment is prone to deviation. Fourth, the conflict handling mechanism is imperfect. When facing obstacles or sudden road changes, reference line adjustments can easily disrupt smoothness, affecting driving comfort and safety, and making it difficult to balance obstacle avoidance requirements and trajectory continuity.

[0131] To address the technical pain points of existing intelligent driving reference line technologies, such as poor adaptability in scenarios without maps or weak maps, and without lane lines, such as intersections, temporary construction areas, and rural roads, which easily lead to reference line construction failures and consequently interruptions in intelligent driving functions and reduced safety, this application aims to provide a method for constructing autonomous vehicle trajectory reference lines based on environmental vehicle information. This method, as a supplement to the existing "map + perception (lane lines, road boundaries)" fusion scheme, solves the problem of difficulty in accurately constructing reference lines in scenarios without maps or weak maps, and without lane lines, while also considering adaptability to scenarios with and without absolute positioning, ensuring the continuity of intelligent driving vehicle functions and driving safety in the aforementioned special scenarios.

[0132] The methods of the embodiments of this application will be further illustrated below.

[0133] In this embodiment, addressing the insufficient adaptability of existing "map + perception (lane lines, road boundaries)" fusion solutions in scenarios without maps / weak maps and without lane lines, such as intersections, temporary construction areas, and rural roads, environmental vehicle trajectories are incorporated into the fusion system to form a supplementary "map + perception (lane lines, road boundaries, vehicle trajectories)" fusion solution. Through a full-process design of "perception-filtering-multi-trajectory construction-separate optimization-dynamic updating," stable driving trajectories of multiple environmental vehicles are collected and filtered to generate multiple initial reference lines. Each reference line is optimized, ultimately outputting multiple reference trajectories adapted to multiple lanes (if any), adapting to large dynamic scenarios such as left and right turns at intersections, and simultaneously compatible with both scenarios with and without absolute positioning. This further improves the scenario coverage capability of the fusion solution and ensures the continuity and safety of intelligent driving functions.

[0134] This module supplements and improves upon the existing "map + perception" fusion framework, comprising five functional modules. The collaborative logic of each module is as follows: First, the environmental perception and multi-target initial screening module relies on the existing vehicle perception system to complete the basic identification and tracking of vehicles in the environment. It does not involve any new perception fusion algorithm, but only adds multi-target screening logic. Second, the coordinate transformation and preprocessing module adopts differentiated coordinate processing strategies based on the differences in vehicle positioning capabilities and designs trajectory point sampling rules in combination with the needs of large dynamic scenarios. Third, the multi-trajectory reference line fitting module generates multiple initial reference lines based on the selected trajectories of multiple vehicles in the environment. Fourth, the reference line optimization module optimizes each initial reference line from three dimensions: smoothness, static safety, and dynamic verification of large dynamic scenarios, and finally outputs multiple reference trajectories adapted to multiple lanes (if any). Fifth, the dynamic update and redundancy backup module adapts to real-time changes in dynamic scenarios such as intersection turning, and ensures the continuous supply of multiple reference trajectories through periodic updates and redundancy design.

[0135] The embodiments described in this application achieve the following: First, they supplement existing fusion systems by incorporating environmental vehicle trajectories into the "map + perception" fusion framework, overcoming the adaptation bottlenecks in scenarios without maps / weak maps and without lane lines. Second, they construct a multi-vehicle, multi-trajectory optimization mechanism, generating and optimizing multiple reference trajectories through the trajectories of multiple environmental vehicles, directly outputting solutions adapted to multiple lanes, and improving the scenario adaptability and selection flexibility of reference trajectories. Third, they adapt to highly dynamic scenarios by integrating dynamic requirements such as left and right turns at intersections into the entire process of target selection, trajectory sampling, fitting, and dynamic verification. Fourth, they take into account both multi-positioning scenarios and engineering applications, requiring no additional hardware and can be directly integrated into existing intelligent driving systems. As a supplement to the "map + perception (lane lines, road boundaries)" solution, it incorporates vehicle trajectory to form a multi-source fusion system, effectively filling the adaptation gaps in scenarios without maps / weak maps and without lane lines; improving the adaptability of reference trajectories: by generating and optimizing multiple reference trajectories through the trajectories of multiple vehicles in the environment, it directly outputs solutions adapted to multiple lanes (if any), providing more options for subsequent trajectory planning and improving the flexibility of scenario adaptation; adapting to the needs of large dynamic scenarios: the entire process is optimized to adapt to large dynamic scenarios such as left and right turns at intersections, ensuring the rationality and safety of each reference trajectory during dynamic driving; accurately ensuring static safety: focusing on the obstacle avoidance needs of static obstacles such as curbs and roadblocks, each reference trajectory is verified to improve driving safety in special scenarios; strengthening functional continuity: the redundant backup mechanism can seamlessly switch with existing solutions, avoiding interruptions in intelligent driving functions, and taking into account both the universality of the solution and its engineering application value.

[0136] Figure 3 This is a flowchart of a method for constructing a trajectory reference line for an intelligent vehicle according to an embodiment of this application, such as... Figure 3 As shown, the method may include the following steps.

[0137] Step S301: Environmental perception and screening of multiple stable reference targets.

[0138] In this embodiment, relying on existing in-vehicle multi-sensor fusion systems (LiDAR, camera, millimeter-wave radar), environmental perception data and basic vehicle identification results are acquired. Existing perception and tracking algorithms are reused without involving any new design, outputting basic information such as the position, speed, and heading angle of the vehicles in the environment. Existing trajectory tracking algorithms are reused to achieve continuous tracking of vehicles in the environment, retaining only stable vehicle targets that have been continuously tracked for ≥5 frames, and eliminating abnormal targets that are frequently lost during tracking.

[0139] Select 1-5 stable reference targets from the tracked vehicles. The selection criteria include: ① Driving continuity: continuous tracking time ≥ 3 seconds; ② Speed ​​adaptability: turning speed ≤ 30km / h, straight-going speed ≤ 80km / h, absolute acceleration ≤ 2m / s²; ③ Attitude adaptability: heading angle change rate ≤ 4° / s (straight-going) and / or ≤ 20° / s (turning left or right at intersections), adapting to attitude changes in turning scenarios, with no obvious reverse driving or illegal driving behavior; ④ Scene matching: in intersection scenarios, prioritize vehicles with the same turning intention (judged by the heading angle change trend, such as a left-turning vehicle whose heading angle continuously shifts to the left), and in rural roads / construction areas, prioritize vehicles traveling along the main road; ⑤ Distance rationality: distance from the vehicle 2-60m, avoiding targets that are obscured or have low trajectory reference value.

[0140] Step S302: Multi-vehicle trajectory point acquisition and preprocessing.

[0141] In this embodiment, multi-vehicle trajectory point differential sampling is adopted: for each selected stable reference target, a dynamic sampling strategy is adopted: the sampling frequency is 10Hz for straight-going scenarios and the sampling frequency is increased to 20Hz for intersection turning scenarios; during sampling, the position, timestamp, heading angle and vehicle status data (speed, yaw rate) of the trajectory points are recorded simultaneously. At least 30 continuous trajectory points are collected for each target vehicle to ensure complete capture of the turning trajectory.

[0142] The differentiated coordinate transformation strategy, when absolute positioning (GNSS+IMU) is available, collects the target vehicle's WGS84 coordinates and converts them to the UTM coordinate system, verifying that the threshold for abrupt changes in trajectory point differences during steering scenarios is ≤8m, adapting to large displacement changes during steering; when absolute positioning is unavailable, coordinates are transformed based on the vehicle's kinematic model, and the vehicle's attitude angle error is corrected in real time during steering. Specifically, it is divided into two scenarios:

[0143] Scenario 1 (with absolute positioning): The vehicle is equipped with a GNSS+IMU combined positioning module (positioning accuracy ≤0.1m, update frequency ≥10Hz), which directly collects the absolute coordinates of the reference target in the WGS84 coordinate system, and at the same time converts the absolute trajectory points of the reference target into the UTM coordinate system to reduce the influence of the Earth's curvature on subsequent fitting calculations.

[0144] Scenario 2 (No Absolute Positioning): Real-time status parameters are acquired based on the vehicle chassis CAN bus (speed error ≤ 0.1m / s, yaw rate error ≤ 0.1° / s). A vehicle coordinate system is established (origin is the rear axle center, X-axis is along the vehicle's forward direction, Y-axis is perpendicular to the X-axis pointing to the left, and Z-axis is perpendicular to the ground and upward). The distance (distance accuracy ≤ 0.1m) and azimuth angle (angle accuracy ≤ 0.1°) of the reference target relative to the vehicle are collected by sensors. This is combined with the vehicle's kinematic model (the formula can be: x...). t =x t-1 +v t ×cos(θ t )×Δt,y t =y t-1 +v t ×sin(θ t ()×Δt, where x and y are coordinates in the vehicle coordinate system, v is the vehicle speed, θ is the vehicle heading angle, and Δt is the sampling time interval) calculates the vehicle attitude change, converts the relative trajectory points of the reference target into absolute coordinates in the vehicle coordinate system, and completes coordinate unification.

[0145] Step S303, multiple initial reference line fitting.

[0146] In this embodiment, the density (number of trajectory points per unit distance) of the trajectory point sequence of each target vehicle is analyzed separately. The density threshold is set to 2 points / meter. The trajectory point sequence is traversed through a sliding window of 1m length to distinguish the trajectory characteristics of straight-going and turning scenarios.

[0147] For each target vehicle, trajectory points are selected independently: 3rd-order B-spline fitting is used when the point density is ≥2 points / meter, which can better preserve trajectory details and ensure smoothness; weighted least squares fitting is used when the point density is <2 points / meter, which improves fitting accuracy through weight allocation and avoids excessive trajectory fluctuations.

[0148] For each target vehicle's trajectory points, an initial reference line is fitted to generate a separate line, with the fitting error constrained to ≤0.3m. This results in multiple initial reference lines corresponding to the number of target vehicles. The specific fitting process is shown below.

[0149] During B-spline fitting, the order can be selected; specifically, a 3rd-order B-spline is used to ensure the continuity of the fitted curve G2 (continuity of position, first derivative, and second derivative), satisfying the requirements for smooth vehicle driving. Node vectors can be set; specifically, uniform node distribution with a node interval of 0.5m (balancing computational load and fitting accuracy) is used, and the number of nodes equals the number of trajectory points plus the order plus 1. Control vertex solving can be performed; specifically, the control vertices are solved by establishing least-squares optimization equations using the trajectory point coordinates as constraints. Initial line generation can be performed; specifically, B-spline curves are generated based on the control vertices and node vectors, and if the fitting error exceeds the standard, the node interval is adjusted and refitting is performed.

[0150] In the weighted least squares fitting process, the function form can be a cubic polynomial (y=ax³+bx²+cx+d), balancing fitting accuracy and computational efficiency. Weight allocation can be performed, specifically using a distance-inverse weighting strategy: the weight of a trajectory point = 1 / (the initial distance from the point to the fitted curve + ε) (ε=0.01 to avoid infinite weights when the distance is 0). Trajectory points closer to each other have higher weights, improving the accuracy of key trajectory segments. Coefficients can be solved, specifically using weighted least squares as the objective function and solving for the polynomial coefficients through matrix inversion. Initial line generation can be performed, specifically generating the fitted curve based on the polynomial coefficients; if the fitting error does not meet the requirements, the weighting strategy is adjusted and refitting is performed.

[0151] Step S304, multi-reference line optimization.

[0152] In this embodiment, the reference line can be optimized for smoothness, verified for static security, verified for dynamics in large dynamic scenes, and verified for path consistency.

[0153] Optionally, during the smoothness optimization process, smoothness optimization can be performed on each initial reference line separately, the curvature of each reference line can be calculated, the threshold for the rate of change of curvature in the steering scenario can be set to 0.15 rad / m, the curvature abrupt change points on each reference line can be identified and locally corrected through a quadratic programming algorithm, so as to ensure that each reference trajectory (especially the steering trajectory) is smooth and continuous and meets the requirements of vehicle driving comfort.

[0154] Optionally, during the static safety verification process, verification can be carried out on static obstacles such as curbs, roadblocks, traffic signs, and cement pillars. The specific process is as follows: Static obstacle information association: obtain the position and size information of static obstacles output by the perception module and convert it to a coordinate system consistent with the reference line; safety distance verification: set the minimum safety distance of static obstacles to ≥2m; trajectory correction: if the distance between the reference line and the static obstacle does not meet the standard, fine-tune the trajectory within the allowable turning range to ensure that the static obstacle is avoided.

[0155] Optionally, during the dynamics verification process for large dynamic scenarios, the focus can be on adapting to large dynamic scenarios such as left and right turns at intersections, and optimizing the dynamic constraints: maximum steering angle constraint, specifically, the steering angle corresponding to the change in the reference line heading angle is ≤±35° (the maximum steering angle threshold of the vehicle), and short-term 35° is allowed in steering scenarios to complete the steering action; maximum lateral acceleration constraint, specifically, ≤±1.5m / s² for straight-going scenarios, and relaxed to ≤±2.5m / s² for steering scenarios, taking into account both steering requirements and driving safety; steering radius constraint, specifically, based on the vehicle's wheelbase (assuming 2.8m), ensuring that the minimum steering radius of the steering reference line is ≥5m to avoid understeer; verification and correction, specifically, a segmented adjustment strategy is adopted for trajectory segments that exceed the constraints, correcting the curvature of the reference line within the constraint range and recalculating the trajectory until all dynamic characteristics requirements are met.

[0156] Optionally, during the path consistency verification process, each optimized reference line can be verified separately to check the consistency between each reference line and the driving trend of the corresponding target vehicle, with a heading angle deviation of ≤8° and a trajectory offset of ≤0.8m in the turning scenario. All reference lines that pass the verification are retained as valid reference trajectories, and finally, multiple reference trajectories adapted to multiple lanes (if any) are output and incorporated into the existing "map + perception" fusion system for selection and use by the subsequent trajectory planning module.

[0157] Step S305: Dynamic updates and redundant backups.

[0158] In this embodiment, the update cycle is dynamically adjusted according to the scenario: 100ms for straight-ahead scenarios and 50ms for intersection turning scenarios. In each cycle, the trajectory points of multiple target vehicles are re-collected, and steps S302 to S304 are repeated to generate and optimize multiple new reference trajectories. A 0.3s smooth transition strategy is used to switch between the old and new reference trajectories to avoid control fluctuations caused by sudden changes in trajectory during turning.

[0159] Step S306: Is the target sufficient?

[0160] In this embodiment, if there are sufficient targets, that is, if there are stable reference targets in the environment where the vehicle is located, the process can return to step S304. Conversely, if there are insufficient targets, that is, if there are no stable reference targets in the environment where the vehicle is located, step S307 can be executed.

[0161] Optionally, the operating status of all stable reference targets is monitored in real time. If any target fails (lost / unstable / out of view), a reference line can still be generated normally if there are ≥1 remaining target; if there are <1 remaining target, it is determined to be a multi-target failure.

[0162] Step S307: Switch to the existing trajectory.

[0163] In this embodiment, a redundancy mechanism is immediately activated when multiple targets fail: short-term trajectory extension, specifically, for each valid reference trajectory, linear extrapolation is performed based on the trajectory points of the last 15 frames before the corresponding target vehicle fails, with an extension distance ≤ 5m and an extension time ≤ 1s, providing a buffer for re-selecting reference targets; new reference target search, specifically, during the redundant trajectory extension period, the environmental perception and target selection process of step S301 is re-executed to select new stable reference targets from vehicles in the surrounding environment; emergency switching, specifically, if ≥ 2 new stable reference targets are found within 1s, multiple new reference trajectories are generated and optimized based on the new target trajectory points, and the transition is smooth; if no new target is found within 1s, an emergency trajectory strategy is activated to switch the reference trajectory to the trajectory output by the existing "map + perception (lane lines / road boundaries)" scheme, ensuring that the intelligent driving function is not interrupted.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0165] According to another aspect of the embodiments of this application, corresponding to the embodiments of the above-described vehicle trajectory determination method, this specification also provides a vehicle trajectory determination system.

[0166] Figure 4 This is a schematic diagram of a vehicle trajectory determination system according to an embodiment of this application, as shown below. Figure 4 As shown, the vehicle trajectory determination system 40 may include: a client 401 for obtaining a trajectory determination instruction for the vehicle; a server 402 for responding to the trajectory determination instruction during vehicle travel, obtaining a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located; determining the initial travel trajectory of the target vehicle based on the trajectory point sequence of the target vehicle, wherein the initial travel trajectory is used to indicate the travel of the environmental object at the current moment; adjusting the initial travel trajectory to obtain an adjusted initial travel trajectory, wherein the executability index of the adjusted initial travel trajectory is greater than the executability index of the initial travel trajectory before adjustment; and determining the target travel trajectory of the vehicle from at least one adjusted initial travel trajectory, wherein the target travel trajectory is used to indicate the travel of the vehicle.

[0167] According to another aspect of the embodiments of this application, corresponding to the embodiments of the above-described method for determining the driving trajectory of a vehicle, this specification also provides a device for determining the driving trajectory of a vehicle.

[0168] Figure 5 This is a schematic diagram of a vehicle trajectory determination device according to an embodiment of this application, as shown below. Figure 5 As shown, the vehicle trajectory determination device 50 may include: an acquisition module 502, a first determination module 504, an adjustment module 506, and a second determination module 508. The acquisition module 502 is used to acquire a sequence of trajectory points of at least one target vehicle in the environment in which the vehicle is located during vehicle operation. The first determination module 504 is used to determine the initial trajectory of the target vehicle based on the sequence of trajectory points of the target vehicle, wherein the initial trajectory indicates the movement of the environmental object at the current moment. The adjustment module 506 is used to adjust the initial trajectory to obtain an adjusted initial trajectory, wherein the executability index of the adjusted initial trajectory is greater than the executability index of the original initial trajectory. The second determination module 508 is used to determine the target trajectory of the vehicle from at least one adjusted initial trajectory, wherein the target trajectory indicates the movement of the vehicle.

[0169] Figure 6 This is a schematic diagram of an electronic device according to an embodiment of this application, such as... Figure 6 As shown, an electronic device 60 is also provided, including: a memory 601 storing an executable program; and a processor 602 for running the program, wherein the program executes the methods in various embodiments of this application when it runs.

[0170] Embodiments of this application also provide a vehicle, including: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods described in various embodiments of this application when it runs.

[0171] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0172] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0173] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0174] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0175] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0176] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0177] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0178] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0179] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0180] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for determining the driving trajectory of a vehicle, characterized in that, include: During the vehicle's operation, a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located is acquired. Based on the trajectory point sequence of the target vehicle, the initial driving trajectory of the target vehicle is determined, wherein the initial driving trajectory is used to indicate the driving of the environmental object at the current moment; The initial driving trajectory is adjusted to obtain an adjusted initial driving trajectory, wherein the executability index of the adjusted initial driving trajectory is greater than the executability index of the initial driving trajectory before adjustment. From at least one adjusted initial driving trajectory, a target driving trajectory of the vehicle is determined, wherein the target driving trajectory is used to indicate the driving of the vehicle.

2. The method according to claim 1, characterized in that, During the vehicle's operation, the trajectory point sequence of at least one target vehicle in the environment in which the vehicle is located is acquired, including: During the vehicle's movement, the initial trajectory point sequence of the target vehicle is acquired; The initial trajectory point sequence is adjusted using the coordinate type of the target vehicle to obtain the trajectory point sequence. The coordinate type is used to indicate the type of coordinate system to which the target vehicle's location belongs. The accuracy of the trajectory point sequence is greater than the accuracy of the initial trajectory point sequence.

3. The method according to claim 2, characterized in that, Using the coordinate type of the target vehicle, the initial trajectory point sequence is adjusted to obtain the trajectory point sequence, including: In response to the coordinate type being a geographic coordinate system, the initial trajectory point sequence is transformed from the geographic coordinate system to the projected coordinate system to obtain the initial trajectory point sequence after the coordinate system transformation; In response to the coordinate type being the relative coordinate system of the target vehicle, the initial trajectory point sequence is transformed from the relative coordinate system to the vehicle's own coordinate system, resulting in the transformed initial trajectory point sequence. The own coordinate system is constructed based on the vehicle's driving state information. Using the vehicle's driving scenario, the initial trajectory point sequence after coordinate system transformation is corrected to obtain the trajectory point sequence, wherein the fit between the trajectory point sequence and the driving scenario is greater than the fit between the initial trajectory point sequence after coordinate system transformation and the driving scenario. And / or, During the vehicle's movement, the initial trajectory point sequence of the target vehicle is acquired, including: During the vehicle's operation, trajectory points of the target vehicle are collected according to the sampling strategy corresponding to the driving scenario. The sampling strategy represents the rules for controlling the vehicle to collect trajectory points at a sampling frequency, and different driving scenarios correspond to different sampling frequencies. The target number of trajectory points are determined as the initial trajectory point sequence.

4. The method according to claim 1, characterized in that, Based on the trajectory point sequence of the target vehicle, the initial driving trajectory of the target vehicle is determined, including: A fitting strategy adapted to the density of the trajectory point sequence is invoked to fit the trajectory point sequence to obtain the initial driving trajectory. Here, the density represents the number of trajectory points per unit distance in the trajectory point sequence, and the fitting strategy represents the rules for fitting the trajectory point sequence.

5. The method according to claim 4, characterized in that, The fitting strategy includes a first fitting strategy, which represents the rules for fitting the trajectory point sequence according to a preset spline curve. The first fitting strategy calls a fitting strategy that matches the density of the trajectory point sequence to perform fitting processing on the trajectory point sequence, thereby obtaining the initial driving trajectory, including: In response to the density being greater than or equal to a density threshold, nodes are determined based on the order of the preset spline curve and the number of trajectory points, and control vertices are determined based on the coordinates of the trajectory points; The first fitting strategy is invoked, and the nodes and control vertices are fitted according to the preset spline curve to obtain the initial driving trajectory; The method further includes: In response to the fact that the fitting error of the initial driving trajectory determined according to the first fitting strategy is greater than the fitting error threshold, the interval between the nodes is adjusted; Using the nodes after adjusting the interval, the initial driving trajectory is refitted until the fitting error is less than or equal to the fitting error threshold. And / or, The fitting strategy includes a second fitting strategy, which represents a rule for fitting the trajectory point sequence according to a preset polynomial. The second fitting strategy calls a fitting strategy adapted to the density of the trajectory point sequence to perform fitting processing on the trajectory point sequence, obtaining the initial driving trajectory, including: In response to the density being less than the density threshold, the weight of the trajectory point is determined; Based on the weights, the polynomial coefficients of the preset polynomial are determined; The second fitting strategy is invoked to fit the initial driving trajectory based on the polynomial coefficients; The method further includes: In response to the fact that the fitting error of the initial driving trajectory determined according to the second fitting strategy is greater than the fitting error threshold, the weights are adjusted; Using the adjusted weights, the initial driving trajectory is refitted until the fitting error is less than or equal to the fitting error threshold.

6. The method according to claim 1, characterized in that, The executability indicators include at least one of the following: smoothness indicator, safety indicator, power indicator, and consistency indicator. The consistency indicator is used to represent the degree of difference between the initial driving trajectory and the driving trend of the vehicle. The initial driving trajectory is adjusted to obtain an adjusted initial driving trajectory, which includes at least one of the following: In response to the curvature of the initial driving trajectory being greater than the curvature threshold corresponding to the driving scenario in which the vehicle is located, the initial driving trajectory is adjusted to obtain the adjusted initial driving trajectory. The curvature is used to represent the smoothness of the initial driving trajectory. The curvature of the adjusted initial driving trajectory is less than or equal to the curvature threshold, and the smoothness index of the adjusted initial driving trajectory is greater than the smoothness index of the initial driving trajectory before adjustment. In response to the initial driving trajectory being less than a distance threshold between the distance between the initial driving trajectory and the obstacles around the vehicle, the initial driving trajectory is adjusted to obtain an adjusted initial driving trajectory, wherein the distance between the adjusted initial driving trajectory and the obstacles is greater than or equal to the distance threshold, and the safety index of the adjusted initial driving trajectory is greater than the safety index of the initial driving trajectory before adjustment. In response to the driving scenario being a target driving scenario, the dynamic constraints allowed by the target driving scenario are invoked to adjust the initial driving trajectory, resulting in an adjusted initial driving trajectory. The target driving scenario indicates the dynamic range of the vehicle's movement, which is greater than the dynamic range of any driving scenario other than the target driving scenario. The adaptation degree between the adjusted initial driving trajectory and the target driving scenario is greater than the adaptation degree between the initial driving scenario before adjustment and the target driving scenario. Furthermore, the dynamic performance index of the adjusted initial driving trajectory is greater than the dynamic performance index of the initial driving trajectory before adjustment. In response to the difference being greater than a difference threshold, the initial driving trajectory is adjusted to obtain the adjusted initial driving trajectory, wherein the driving trend is used to represent the driving state of the vehicle at the future time, the difference between the adjusted initial driving trajectory and the driving trend is less than or equal to the difference threshold, and the consistency index of the adjusted initial driving trajectory is greater than the consistency index of the initial driving trajectory before adjustment.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes at least one of the following: In response to the target vehicle being in a failed state, the initial driving trajectory is extended based on the trajectory points of the target vehicle before it is in the failed state to obtain the extended initial driving trajectory, wherein the driving length of the extended initial driving trajectory is greater than the driving length of the initial driving trajectory before the extension, and / or the driving time of the extended initial driving trajectory is greater than the driving time of the initial driving trajectory before the extension. In response to the target vehicle meeting one of the following conditions, the target driving trajectory is generated based on the vehicle's map information and the vehicle's perception information: the number of target vehicles is less than one; the target vehicle is in the failed state, and there is no target vehicle in a normal state within a preset time period; In response to the fact that the number of target vehicles in the normal state within the preset time period is greater than one, the initial driving trajectory is determined using the trajectory point sequence of the target vehicles in the normal state.

8. A system for determining the driving trajectory of a vehicle, characterized in that, include: The client is used to obtain the trajectory determination instruction for the vehicle; The server is configured to, during the vehicle's movement, respond to the trajectory determination instruction and acquire a sequence of trajectory points of at least one target vehicle in the environment where the vehicle is located; determine an initial driving trajectory of the target vehicle based on the sequence of trajectory points of the target vehicle, wherein the initial driving trajectory is used to indicate the movement of the environmental object at the current moment; adjust the initial driving trajectory to obtain an adjusted initial driving trajectory, wherein the executability index of the adjusted initial driving trajectory is greater than the executability index of the initial driving trajectory before adjustment; and determine the target driving trajectory of the vehicle from at least one adjusted initial driving trajectory, wherein the target driving trajectory is used to indicate the movement of the vehicle.

9. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 7.

10. A vehicle, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 7.

Images