Vehicle control method and vehicle

Abstract

The application discloses a vehicle control method and a vehicle, and relates to the technical field of intelligent driving, and comprises the following steps: acquiring road boundary perception data and state data of a vehicle in a driving process; determining a current spatial pose relationship of the vehicle relative to a road boundary based on the road boundary perception data and the state data; dynamically tracking the road boundary based on the current spatial pose relationship, and generating a dynamic tracking result; generating a target driving trajectory according to the dynamic tracking result and the state data; determining a collision risk trend according to a real-time relative state of the vehicle and the road boundary in the process of the vehicle driving along the target driving trajectory; and controlling the vehicle to perform an avoidance action according to the collision risk trend. The method realizes dynamic tracking of the road boundary and cooperative generation of a trajectory by penetrating the whole process of perception, decision and control, overcomes the problems of information time delay and adaptive deviation, and improves driving safety.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a vehicle control method and a vehicle. Background Technology

[0002] With the continuous iteration of autonomous driving technology, navigation-on-autopilot (NOA) functions have been gradually applied in complex urban road scenarios. Among them, U-turn scenarios, as a high-frequency typical operating condition of urban NOA, place extremely high demands on the reliability of the perception system and the real-time performance of the decision-making system. The curb, as the most critical road boundary reference during the U-turn process, directly affects the safety of the vehicle's driving trajectory and the smoothness of traffic flow due to its perception accuracy and response time.

[0003] Currently, existing NOA systems generally adopt a modular architecture for U-turn scenarios, meaning that the perception, decision-making, and control components operate relatively independently. The perception module is responsible for outputting static environmental information such as the curb position, the decision-making module uses this information to plan the trajectory, and the control module executes corresponding commands. However, due to the lack of a high-frequency dynamic coordination mechanism between the modules throughout the entire U-turn process, the perception results cannot be adaptively adjusted according to the vehicle's real-time attitude and trajectory progress, resulting in significant time delays and adaptation deviations between the perceived information and the trajectory planning. Furthermore, existing NOA systems mostly employ a passive response-based prevention strategy, triggering adjustments only after a deviation or anomaly occurs. This lacks the ability to proactively predict collision risks based on time-series extrapolation, making it difficult to promptly identify and avoid potential curb contact risks during the U-turn process. Summary of the Invention

[0004] This application provides a vehicle control method and a vehicle, with the aim of improving the driving safety of navigation-assisted driving.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A vehicle control method, comprising:

[0007] Acquire road boundary perception data and vehicle status data during vehicle operation;

[0008] Based on the road boundary perception data and the state data, the current spatial pose of the vehicle relative to the road boundary is determined.

[0009] Based on the current spatial pose relationship, the road boundary is dynamically tracked to generate dynamic tracking results;

[0010] Based on the dynamic tracking results and the status data, the target driving trajectory of the vehicle is generated;

[0011] During the process of the vehicle traveling along the target driving trajectory, the collision risk trend is determined based on the real-time relative state between the vehicle and the road boundary.

[0012] Based on the collision risk trend, the vehicle is controlled to perform corresponding avoidance maneuvers.

[0013] A vehicle includes: a processor, a memory, and a bus; the processor and the memory are connected via the bus.

[0014] The memory is used to store a program, and the processor is used to run the program, wherein the program is executed by the processor to perform the vehicle control method described above.

[0015] The technical solution provided in this application includes: acquiring road boundary perception data and vehicle state data during vehicle operation; determining the vehicle's current spatial pose relative to the road boundary based on the road boundary perception data and state data; dynamically tracking the road boundary based on the current spatial pose relationship and generating dynamic tracking results; generating the vehicle's target driving trajectory based on the dynamic tracking results and state data; determining the collision risk trend based on the real-time relative state between the vehicle and the road boundary while the vehicle is traveling along the target driving trajectory; and controlling the vehicle to perform corresponding avoidance actions based on the collision risk trend. This application, by establishing a dynamic collaborative mechanism of perception-decision-control and a forward-looking risk prediction system, achieves adaptive tracking and proactive prevention of road boundaries by the vehicle in the target scenario, effectively improving driving safety and traffic flow. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a first flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0018] Figure 2 This is a second flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0019] Figure 3 A third flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0020] Figure 4 A fourth flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0021] Figure 5 A fifth flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0022] Figure 6 A sixth flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0023] Figure 7 A seventh flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0024] Figure 8 The eighth flowchart of a vehicle control method provided in this application embodiment;

[0025] Figure 9 A ninth flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0026] Figure 10 A tenth flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0027] Figure 11 This is an eleventh flowchart illustrating a vehicle control method provided in an embodiment of this application;

[0028] Figure 12 This is a schematic diagram of the twelfth step of a vehicle control method provided in an embodiment of this application;

[0029] Figure 13 This is a schematic diagram of the architecture of a vehicle provided in an embodiment of this application. Detailed Implementation

[0030] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0032] like Figure 1 The diagram shown is a first flowchart of a vehicle control method provided in an embodiment of this application, which can be applied to a vehicle control system and includes the following steps.

[0033] S101: Acquire road boundary perception data and vehicle status data during vehicle operation.

[0034] Among them, road boundary perception data refers to the perception information about road boundaries (such as road edges) collected by vehicles through on-board sensors (such as surround view cameras, lidar, millimeter-wave radar, etc.), including geometric attribute information such as the location, height, width, and direction of the road boundary.

[0035] Vehicle status data refers to the vehicle's own motion status information collected by the on-board positioning unit and body sensors, including the vehicle's current position, heading angle, speed, steering angle, and body attitude.

[0036] In a possible implementation, the vehicle control system acquires real-time sensing data from surround-view cameras, LiDAR, millimeter-wave radar, and the vehicle positioning unit via an onboard communication bus (such as a controller area network bus or onboard Ethernet). Due to differences in sampling frequencies and data output formats among the various sensors, the system needs to perform timestamp synchronization and coordinate system alignment on the aforementioned multi-source sensing data. Timestamp synchronization refers to associating data collected by each sensor at the same time with the same time reference, while coordinate system alignment refers to transforming the data output by each sensor from its respective sensor coordinate system to a unified vehicle coordinate system. Through these synchronization and alignment processes, the accuracy of subsequent multi-source data fusion processing can be ensured.

[0037] In some examples, the sampling frequency of multi-source sensing data (i.e., sensing information of road boundaries) can be set according to actual needs, such as 20Hz, that is, a set of data is collected every 50ms to ensure the real-time performance and accuracy of the data.

[0038] Optionally, before acquiring road boundary perception data and vehicle state data during vehicle operation, embodiments of this application may also utilize... Figure 12 The steps shown determine the prerequisites for performing this method.

[0039] S102: Based on road boundary perception data and state data, determine the current spatial pose relationship of the vehicle relative to the road boundary.

[0040] Specifically, road boundary area information can be generated based on road boundary perception data, and the current spatial pose of the vehicle relative to the road boundary can be determined based on the road boundary area information and state data. Road boundary area information refers to information that characterizes the spatial distribution range of the road boundary. This information represents the spatial distribution of the road boundary in the form of a region, which is different from the discrete point or line information output by traditional technologies.

[0041] The current spatial pose relationship refers to the position and attitude relationship of the vehicle relative to the road boundary at the current moment, including spatial geometric parameters such as the relative distance and relative angle between the vehicle and the road boundary.

[0042] In a possible implementation, the vehicle control system, based on the road boundary perception data acquired in step S101, integrates the discrete perception data into spatially continuous regional information, namely road boundary region information, through spatial continuity analysis. Spatial continuity analysis refers to the process of identifying spatially adjacent data points with consistent geometric features in the perception data and aggregating these data points into a continuous region. Subsequently, based on the road boundary region information and the vehicle's state data (including current position, heading angle, etc.), the system calculates the vehicle's current spatial pose relative to the road boundary, specifically including determining parameters such as the current distance between the vehicle and the road boundary, and the deviation of the vehicle's heading angle relative to the road boundary's direction.

[0043] Optionally, the process of determining the vehicle's current spatial pose relative to the road boundary based on road boundary perception data and state data can be found in [reference needed]. Figure 2 The steps are shown.

[0044] Optionally, embodiments of this application can also be implemented through... Figure 10 The steps shown improve the accuracy and robustness of road boundary recognition from the source of perception.

[0045] S103: Based on the current spatial pose relationship, dynamically track the road boundary and generate dynamic tracking results.

[0046] Dynamic tracking refers to the process of continuously tracking the position and direction of the road boundary based on the real-time spatial pose relationship between the vehicle and the road boundary.

[0047] It should be noted that the dynamic tracking is adjusted based on the vehicle's real-time status and the progress of the scene, rather than preset fixed parameters. This allows the tracking system to adapt to the dynamic changes in the relative positional relationship between the vehicle and the road boundary at different stages of the scene.

[0048] In a possible implementation, the vehicle control system determines the vehicle's driving progress in the current target scene based on the current spatial pose relationship determined in step S102, and dynamically adjusts the tracking strategy of the road boundary according to the driving progress.

[0049] For example, at different stages of a U-turn scenario (such as entering the U-turn area, the middle stage of the U-turn, and exiting the U-turn area), the vehicle's attitude and trajectory characteristics differ significantly. The system adaptively adjusts the tracking parameters (such as tracking sensitivity, tracking window size, filtering coefficients, etc.) according to the current stage to generate dynamic tracking results that are adapted to the current scenario, avoiding tracking lag or oversensitivity caused by fixed tracking logic.

[0050] Optionally, the process of dynamically tracking the road boundary based on the current spatial pose relationship and generating dynamic tracking results can be found in [reference needed]. Figure 4 The steps are shown.

[0051] S104: Generate the target driving trajectory of the vehicle based on the dynamic tracking results and status data.

[0052] The target driving trajectory refers to the spatiotemporal path that the vehicle is expected to travel along in the target scene. This trajectory consists of a series of path points and their corresponding time information.

[0053] The coupling of dynamic tracking results and state data refers to the fusion of dynamic tracking information of road boundaries (i.e., the position, direction and changing trend of road boundaries) with the vehicle's own state (i.e., position, heading angle, speed, etc.) to generate a driving trajectory that maintains an appropriate safe distance from the road boundaries.

[0054] In a possible implementation, the vehicle control system couples the dynamic tracking results generated in step S103 with the vehicle's state data to determine a safe driving corridor for the vehicle relative to the road boundary. A safe driving corridor refers to the spatiotemporal area within which the vehicle can safely travel without interfering with the road boundary. Based on this, the system combines the geometric features and kinematic constraints (such as the vehicle's minimum turning radius and maximum steering angular velocity) of the target scenario type (e.g., a U-turn scenario) to plan a target driving trajectory from the vehicle's current position to the scenario exit, maintaining a preset safe distance from the road boundary.

[0055] Optionally, the process of generating the vehicle's target driving trajectory based on dynamic tracking results and status data can be found in [reference needed]. Figure 3 The steps are shown.

[0056] S105: During the process of the vehicle traveling along the target driving trajectory, the collision risk trend is determined based on the real-time relative state between the vehicle and the road boundary.

[0057] Among them, real-time relative state refers to the real-time spatial relationship between the vehicle and the road boundary during the vehicle's travel along the target trajectory, including real-time distance and real-time relative attitude (such as heading deviation).

[0058] Collision risk trend refers to the degree of risk of a collision between a vehicle and a road boundary, and the direction of its change, determined based on the current real-time relative state. This trend is used to characterize the direction of collision risk changes over a future period of time.

[0059] In a possible implementation, as the vehicle travels along the target trajectory, the system continuously acquires the real-time distance and attitude deviation between the vehicle and the road boundary, and determines the collision risk trend based on these parameters. Specifically, the system determines a safety distance margin based on the real-time distance and a preset safety distance, and comprehensively assesses the collision risk level at the current moment and the risk change trend over a future period by combining the attitude deviation.

[0060] Optionally, the real-time relative state includes the real-time distance and attitude deviation between the vehicle and the road boundary. The process of determining the collision risk trend based on the real-time relative state between the vehicle and the road boundary can be found in [reference needed]. Figure 5 The steps are shown.

[0061] S106: Based on the collision risk trend, control the vehicle to perform corresponding avoidance actions.

[0062] Avoidance maneuvers refer to the control operations performed by the vehicle's control system to avoid the risk of collision with the road boundary, including deceleration, trajectory adjustment, and combinations thereof. The specific type and intensity of the avoidance maneuver are determined by the level of the collision risk trend.

[0063] In a possible implementation, the vehicle control system determines an avoidance maneuver corresponding to the collision risk trend determined in step S105, and controls the vehicle to execute the avoidance maneuver. When the collision risk trend is a first risk trend, the system controls the vehicle to perform emergency deceleration and adjust its driving trajectory to increase the distance between the vehicle and the road boundary. When the collision risk trend is a second risk trend, the system controls the vehicle to perform moderate deceleration and adjust its driving trajectory to increase the safety distance margin to a safe range. When the collision risk trend is a third risk trend, the system controls the vehicle to maintain its current driving state and continuously monitors the real-time relative status.

[0064] Optionally, the process of controlling the vehicle to perform corresponding avoidance actions based on the collision risk trend can be found in [reference needed]. Figure 7The steps are shown.

[0065] Optionally, after the vehicle performs the corresponding avoidance maneuver, combined with Figure 8 The steps shown in this application embodiment can also be achieved through... Figure 9 The steps shown modify the tracking parameters in the updated tracking strategy to improve the reliability of the tracking strategy.

[0066] The processes shown in S101-S106 above, by establishing a dynamic collaborative mechanism of perception-decision-control and a forward-looking risk prediction system, enable vehicles to adaptively track and actively prevent and control road boundaries in target scenarios, effectively improving driving safety and traffic flow.

[0067] like Figure 2 The diagram shown is a second flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0068] S201: Divide the sensing area corresponding to the road boundary sensing data into multiple sensing units.

[0069] The perception area refers to the spatial range covered by the road boundary perception data. This range is usually the field of view centered on the vehicle and extending in front of and to the sides of the vehicle.

[0070] A sensing unit refers to several sub-regions into which the sensing area is divided according to a preset size. Each sensing unit corresponds to a local spatial block within the sensing area. The purpose of dividing the sensing area into sensing units is to discretize a continuous space into basic units that can be independently evaluated.

[0071] In one possible implementation, the vehicle control system divides the sensing area corresponding to the road boundary perception data into multiple sensing units according to a preset unit size (e.g., 0.5m × 0.5m). These sensing units are arranged continuously in space, collectively covering the entire sensing area. Each of the divided sensing units can then be independently evaluated for confidence and validity.

[0072] S202: Obtain the boundary confidence level corresponding to each sensing unit.

[0073] Boundary confidence refers to the degree of confidence that a road boundary exists in each sensing unit. The boundary confidence is calculated by the sensing model based on the sensing data within each sensing unit. The value of boundary confidence ranges from 0 to 1, with a higher value indicating a higher confidence in the existence of a road boundary.

[0074] In a possible implementation, the system inputs the sensing data from each sensing unit into the sensing model, which then independently evaluates the confidence level of the boundary existence of each sensing unit and outputs the boundary confidence level corresponding to each sensing unit.

[0075] S203: Multiple sensing units whose boundary confidence meets the preset confidence conditions and whose locations are adjacent and uninterrupted are identified as effective region units.

[0076] The preset confidence condition refers to the confidence threshold used to determine whether a single sensing unit is valid, such as a boundary confidence level greater than or equal to a preset confidence threshold (e.g., 0.8). This confidence threshold can be dynamically adjusted according to the target scene type.

[0077] Adjacent and uninterrupted means that multiple sensing units are spatially adjacent to each other without gaps or breaks, that is, these sensing units together constitute a continuous and uninterrupted region.

[0078] In a possible implementation, the system first filters out sensing units whose boundary confidence meets the preset confidence conditions, and then further identifies multiple sensing units that are adjacent and uninterrupted from these valid sensing units.

[0079] For example, when multiple consecutive sensing units (e.g., three or more consecutive units) within a certain area meet the preset confidence level, the system collectively identifies these consecutive sensing units as valid area units. If the number of sensing units that consecutively meet the conditions is less than the preset threshold (e.g., less than three), it is determined to be an invalid area.

[0080] S204: Generate road boundary area information based on each valid area unit.

[0081] The system integrates all the sensing units that are identified as valid area units to generate road boundary area information.

[0082] In some examples, the road boundary area information represents the spatial distribution range of the road boundary in the form of a region, which is different from the detection results output in the form of points or lines in traditional technologies.

[0083] S205: Based on road boundary area information and status data, determine the current spatial pose relationship of the vehicle relative to the road boundary.

[0084] The current spatial pose relationship refers to the vehicle's position and attitude relative to the road boundary at the current moment, including spatial geometric parameters such as the relative distance and relative angle between the vehicle and the road boundary. This current spatial pose relationship is the data foundation for subsequent dynamic tracking and trajectory planning.

[0085] In a possible implementation, the system can determine the spatial distribution range and direction of the road boundary based on the road boundary area information, and calculate the relative distance between the vehicle and the road boundary, as well as the deviation of the vehicle's heading angle relative to the direction of the road boundary, by combining the vehicle's current position and heading angle in the state data, thereby determining the vehicle's current spatial pose relative to the road boundary.

[0086] The processes described in S201-S205 above, by dividing the sensing area into multiple sensing units and performing effective area determination based on boundary confidence and spatial continuity, upgrade from point / line detection to area detection, effectively improving the accuracy and robustness of road boundary perception. Based on this, the current spatial pose relationship is determined using road boundary area information and state data, providing accurate vehicle-road boundary relative positioning information for subsequent dynamic tracking and trajectory planning.

[0087] like Figure 3 The diagram shown is a third flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0088] S301: Determine the location and direction of the road boundary based on the dynamic tracking results.

[0089] The location of the road boundary refers to its spatial coordinates in the vehicle coordinate system or global coordinate system, representing its specific spatial position. The orientation of the road boundary refers to its spatial orientation along its extension direction, representing its extension trend and directional changes.

[0090] In a possible implementation, the system extracts the location and orientation information of the road boundary from the dynamic tracking results generated in step S103 (or step S403). The dynamic tracking results contain continuous spatial distribution information of the road boundary. The system extracts geometric parameters from this continuous distribution information to determine the spatial coordinates of each key point on the road boundary, and fits the orientation of the road boundary based on the spatial coordinates of each key point.

[0091] For example, the system can use the least squares method to perform straight line fitting or curve fitting on multiple location points on the road boundary to obtain the overall orientation of the road boundary.

[0092] S302: Determine the initial safe distance between the vehicle and the road boundary based on the location and direction of the road boundary and the current location in the status data.

[0093] The initial safe distance refers to the actual spatial distance between the vehicle and the road boundary in its current state, without any additional trajectory adjustments. The current position refers to the information in the state data that characterizes the vehicle's current spatial location.

[0094] In a possible implementation, the system calculates the initial safe distance between the vehicle and the road boundary based on the location and orientation of the road boundary determined in step S301, and the current location in the state data. Specifically, the system determines the shortest distance from the vehicle to the road boundary as the initial safe distance based on the spatial relationship between the vehicle's current location and the location of the road boundary, combined with the orientation of the road boundary.

[0095] For example, the system can calculate the vertical distance from the vehicle's current position to the fitted curve of the road boundary, which is the initial safe distance.

[0096] S303: Determine the target safety distance based on the target scenario type and the current state of the vehicle.

[0097] The target safety distance refers to the expected safe distance to ensure that the vehicle does not collide with the road boundary under the current target scenario type and vehicle state. The value of the target safety distance is dynamically determined based on the target scenario type and the current state of the vehicle; different target safety distances correspond to different types of scenarios and different vehicle states.

[0098] In some examples, vehicle status data includes information such as vehicle speed and vehicle attitude. Since vehicle speed directly affects the vehicle's braking distance and steering response, the target safety distance should vary at different vehicle speeds; that is, the higher the vehicle speed, the greater the required target safety distance. Furthermore, in different target scenario types (e.g., wide curb scenarios and narrow curb scenarios), the target safety distance should also be adaptively adjusted due to the different physical characteristics of road boundaries and the severity of collision consequences.

[0099] In a possible implementation, the system determines the type of target scene based on the identified target scene (see [reference]). Figure 12 Based on the steps shown and the vehicle's current state (e.g., current speed, vehicle posture), the corresponding target safety distance is determined from a preset scene-state mapping relationship. The preset scene-state mapping relationship specifies the target safety distance corresponding to different target scene types and different vehicle states.

[0100] For example, in wide curb scenarios, because the road boundary is relatively wide and rigid, the system can set a larger target safety distance (e.g., 0.4m or more); in narrow curb scenarios, where the road boundary is relatively narrow, the system can set a smaller target safety distance (e.g., 0.3m or more). The specific value of the target safety distance can be dynamically adjusted according to the actual scenario and vehicle status. The system can also set different target safety distances at different vehicle speeds, i.e., a larger target safety distance is set when driving at high speeds, and a smaller target safety distance is set when driving at low speeds.

[0101] S304: When the initial safe distance is less than the target safe distance, adjust the target driving trajectory to increase the distance between the vehicle and the road boundary until the safe distance between the vehicle and the road boundary is greater than or equal to the target safe distance.

[0102] Adjusting the target driving trajectory refers to correcting the generated target driving trajectory online by shifting the trajectory away from the road boundary to increase the actual distance between the vehicle and the road boundary.

[0103] In some examples, the specific implementation methods for adjusting the target driving trajectory may include two types: local trajectory correction and global trajectory replanning. The former corrects the position of local path points of the original trajectory, while the latter regenerates the complete driving trajectory based on new safety constraints.

[0104] In this embodiment, "until the safe distance between the vehicle and the road boundary is greater than or equal to the target safe distance" means that the adjustment process uses the attainment of the target safe distance as a convergence condition. The system continuously iterates and corrects the trajectory until this condition is met. This convergence condition ensures that the trajectory adjustment has a clear target orientation and verifiable results. The safe distance between the vehicle and the road boundary is obtained using the same calculation method as the initial safe distance in step S302.

[0105] In a possible implementation, when the initial safe distance determined in step S302 is less than the target safe distance determined in step S303, it indicates that the safety redundancy between the current target driving trajectory and the road boundary is insufficient, and the target driving trajectory needs to be adjusted. The system performs online correction on the generated target driving trajectory, shifting the trajectory away from the road boundary by a preset amount, and then recalculates the safe distance between the vehicle and the road boundary. If the recalculated safe distance is still less than the target safe distance, the shift amount is increased until the safe distance is greater than or equal to the target safe distance. Through the above adjustments, the system finally generates a target driving trajectory that maintains a sufficient safe distance from the road boundary, ensuring that the vehicle remains within a safe range from the road boundary during driving.

[0106] Optionally, after completing the above trajectory adjustment, the vehicle control system will use the adjusted trajectory as the final target driving trajectory for subsequent driving control and collision risk prediction.

[0107] The processes described in S301-S304 above determine the position and direction of the road boundary based on the dynamic tracking results, and dynamically determine the target safety distance by combining the target scene type and the current state of the vehicle. When the initial safety distance is insufficient, the target driving trajectory is actively adjusted, which realizes the precise coordination between perception signals and trajectory planning, and effectively avoids the collision risk caused by insufficient safety distance between the trajectory and the road boundary.

[0108] like Figure 4 The diagram shown is a fourth flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0109] S401: Determine the vehicle's current driving progress based on the current spatial pose relationship.

[0110] The current scene driving progress refers to the degree to which the vehicle has completed its journey from the starting position to the ending position in the current target scene. This current scene driving progress can be characterized from two dimensions: spatial and attitude. Spatially, the current scene driving progress is reflected in the proportion of the length of the trajectory already traveled by the vehicle to the total planned trajectory length of the entire scene. In terms of attitude, the current scene driving progress is reflected in the proportion of the cumulative change in the vehicle's heading angle relative to the initial heading angle of the scene to the total planned heading change of the scene.

[0111] In this embodiment, the current spatial pose relationship is used to verify whether the determined driving progress matches the actual spatial relationship between the vehicle and the road boundary.

[0112] In a possible implementation, the system calculates the spatial dimension progress and the attitude dimension progress based on the vehicle's current position and attitude in the state data.

[0113] For spatial dimension progress, the system determines the trajectory length between the projection position of the vehicle's current position on the preset trajectory and the starting position of the scene, and calculates the ratio of this trajectory length to the total planned trajectory length of the scene.

[0114] For the attitude dimension progress, the system calculates the angle difference between the vehicle's current heading angle and the scene's initial heading angle, and then calculates the ratio of this angle difference to the planned total heading change in the scene. After obtaining the progress in the above two dimensions, the system determines the current scene driving progress based on either one or the weighted average of the two, according to a preset strategy.

[0115] The system verifies the current driving progress using relative distances and angles in the current spatial pose relationship: it infers the relative position between the vehicle and the road boundary based on the relative distances and angles, and determines whether this relationship matches the expected position corresponding to the current driving progress. If the verification passes, the driving progress is confirmed as valid. If there is a deviation, it is corrected by incorporating known geometric information of the road boundary to ensure that the final determined driving progress matches the actual spatial relationship between the vehicle and the road boundary.

[0116] For example, in a U-turn scenario, the system calculates the spatial dimension progress based on the projected position of the vehicle's current location on the U-turn trajectory. It then calculates the attitude dimension progress based on the angle difference between the vehicle's current heading angle and the starting heading angle of the U-turn. Based on this, the system uses the relative distance and relative angle in the current spatial pose relationship to verify whether the above progress is consistent with the actual spatial relationship between the vehicle and the road boundary.

[0117] S402: Based on the current driving progress, adjust the tracking strategy for the road boundary to adapt to the scenario and obtain an updated tracking strategy.

[0118] The tracking strategy refers to the set of control parameters and logic used by the vehicle control system to track the road boundary. Specifically, it may include tracking sensitivity (i.e., the response speed to changes in the position of the road boundary), tracking window size (i.e., the range of tracking search), and filtering coefficient (i.e., the strength of suppression of perceived noise).

[0119] Scene adaptation adjustment refers to dynamically adjusting various parameters in the tracking strategy based on the vehicle's current driving progress in the scene, so that the tracking strategy matches the current scene stage.

[0120] In a possible implementation, the system adapts and adjusts the tracking strategy for the road boundary based on the current driving progress determined in step S401. For example, in the initial stage of a U-turn scenario (when the current driving progress is low), the vehicle may be far from the road boundary, and the tracking strategy can be relatively lenient; in the middle stage of a U-turn scenario (when the current driving progress is moderate), the vehicle approaches the core area of ​​the road boundary, and the tracking strategy should be more sensitive and accurate; in the final stage of a U-turn scenario (when the current driving progress is high), the vehicle gradually moves away from the road boundary, and the tracking strategy can gradually return to the normal state. Through the above adaptation and adjustment, the system obtains an updated tracking strategy.

[0121] S403: Based on the updated tracking strategy, track the road boundaries and generate dynamic tracking results.

[0122] The dynamic tracking result refers to the tracking information output by continuously tracking the road boundary using the updated tracking strategy, including the real-time position, direction, and changing trend of the road boundary over time. Compared with the dynamic tracking result generated in step S103, the dynamic tracking result generated in this step is obtained based on the updated tracking strategy after scene adaptation. Both are used for subsequent trajectory planning, but the difference is that the tracking strategy used in this step has been adapted and adjusted according to the current driving progress.

[0123] In a possible implementation, the system uses the updated tracking strategy obtained in step S402 to track the road boundary. The system continuously tracks the road boundary based on the updated tracking strategy, generating dynamic tracking results, which are used in subsequent trajectory planning steps.

[0124] S404: When an abnormality occurs in the sensing signal, execute the exception handling logic.

[0125] Among them, abnormal sensing signals include a sharp drop in road boundary sensing confidence and large-scale area fractures.

[0126] A sudden drop in perception confidence refers to a significant decrease in the boundary confidence of the perception model output within a short period of time, which may be caused by factors such as sensor obstruction, strong light interference, or severe weather.

[0127] Discontinuity in a region refers to the interruption or gap in a previously continuous effective region unit, which may be caused by missing sensing data or local false detection.

[0128] In one possible implementation, as the vehicle travels along the target trajectory, the system continuously monitors road boundary perception signals. When an anomaly is detected in the perception signal, the system immediately retrieves historical road boundary data and panoramic image data to assist in determining the location of the road boundary. Simultaneously, the system controls the vehicle to reduce its speed (e.g., to below 3 km / h) to decrease the probability of a collision due to the perception anomaly. Once the road boundary perception signal returns to normal, the system resumes normal driving speed.

[0129] The processes described in S401-S404 above achieve adaptive optimization of road boundary tracking by dynamically adjusting the tracking strategy based on the vehicle's progress in the current scenario. Furthermore, based on the anomaly handling logic shown in S404, the system's fault tolerance and safety are further enhanced by retrieving historical data and implementing speed reduction control when anomalies are detected.

[0130] like Figure 5 The diagram shown is a fifth flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0131] S501: Determine the safety distance margin based on the real-time distance and the preset safety distance.

[0132] Real-time distance refers to the current actual distance between the vehicle and the road boundary, which can be obtained through ranging data from onboard sensors or through spatial calculations of vehicle positioning data and road boundary area information.

[0133] The preset safety distance is the minimum allowable distance set to ensure that the vehicle does not collide with the road boundary. It can be dynamically adjusted according to the target scenario type and the current vehicle speed. For example, the preset safety distance can be appropriately increased when the vehicle speed is high or the road boundary is a rigid structure.

[0134] Safety margin refers to the difference between the real-time distance and the preset safety distance, used to characterize the degree of safety redundancy between the vehicle and the road boundary. When the safety margin is positive, it means that the distance between the vehicle and the road boundary is greater than the preset safety distance, and the vehicle is in a safe state. When the safety margin is negative, it means that the distance between the vehicle and the road boundary is less than the preset safety distance, and there is a risk of collision.

[0135] In a possible implementation, the system acquires the real-time distance between the vehicle and the road boundary, compares the real-time distance with a preset safety distance, and calculates the difference between the two as a safety distance margin.

[0136] S502: Determine the collision risk trend based on the safety distance margin and attitude deviation.

[0137] Attitude deviation refers to the difference between the vehicle's current heading angle and its desired heading angle. This deviation reflects the degree of attitude deviation of the vehicle relative to the road boundary. The desired heading angle can be the tangent direction of the target trajectory at the current vehicle position, or it can be the direction parallel to the road boundary.

[0138] When assessing collision risk trends, taking into account both safety distance margin (which provides lateral safety information at the current location) and attitude deviation (which provides information on future motion trends) can upgrade from static assessment based on current distance to dynamic assessment based on both current distance and future trends.

[0139] In a possible implementation, the system comprehensively assesses the collision risk trend based on the safety distance margin and attitude deviation determined in step S501. Specifically, the system compares the safety distance margin with a preset margin threshold and the attitude deviation with a preset attitude deviation threshold, and determines the level of the collision risk trend based on the comparison results.

[0140] In some examples, the system calls the collision risk prediction model at a preset period (e.g., every 50ms) to obtain the safety distance margin and attitude deviation, and predicts the trend of the relative position change between the vehicle and the road boundary within a preset time period (e.g., 0.5s) based on the safety distance margin and attitude deviation at the current moment.

[0141] Optionally, the collision risk trend includes a first risk trend, a second risk trend, and a third risk trend, where the first risk trend has a higher risk level than the second risk trend, and the second risk trend has a higher risk level than the first risk trend. The process for determining the collision risk trend based on the safety distance margin and attitude deviation can be found in [reference needed]. Figure 6 The steps are shown.

[0142] The processes shown in S501-S502 above, by combining two dimensions—safety distance margin and attitude deviation—to conduct a collision risk assessment, achieve a comprehensive and accurate determination of collision risk trends.

[0143] like Figure 6 The diagram shown is a sixth flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0144] S601: When the safety distance margin is less than the first margin threshold, or the attitude deviation is greater than or equal to the attitude deviation threshold, the collision risk trend is determined as the first risk trend.

[0145] The first margin threshold is a critical value for the safety distance margin used to distinguish between the first risk trend and the second risk trend. The attitude deviation threshold is a critical value used to determine whether the attitude deviation has reached a high-risk level.

[0146] It should be noted that when the safe distance margin is less than the first margin threshold, even if the vehicle's posture is normal, the actual distance between the vehicle and the road boundary is already close to or touches the danger boundary. When the posture deviation is greater than or equal to the posture deviation threshold, even if the current safe distance margin is still acceptable, the vehicle's posture indicates that it will rapidly approach the road boundary in a short period of time.

[0147] Generally speaking, the first risk trend indicates that the vehicle currently faces an immediate risk of collision.

[0148] In a possible implementation, the system compares the safety distance margin determined in step S501 with a first margin threshold, and compares the attitude deviation with an attitude deviation threshold. When the safety distance margin is less than the first margin threshold, or the attitude deviation is greater than or equal to the attitude deviation threshold, the system determines the collision risk trend as the first risk trend.

[0149] S602: When the safety distance margin is greater than or equal to the first margin threshold and less than the second margin threshold, the collision risk trend is determined to be the second risk trend.

[0150] The second margin threshold is a critical safety distance margin used to distinguish between the second and third risk trends, and the first margin threshold is smaller than the second margin threshold. The interval between the first and second margin thresholds is the medium-risk interval. In this medium-risk interval, although the distance between the vehicle and the road boundary has not yet reached the danger threshold, there is no sufficient safety redundancy.

[0151] Generally speaking, the second risk trend indicates that the vehicle currently faces a potential collision risk.

[0152] In a possible implementation, when the safety distance margin is greater than or equal to a first margin threshold and less than a second margin threshold, the system determines the collision risk trend as the second risk trend.

[0153] S603: When the safety distance margin is greater than or equal to the second margin threshold and the attitude deviation is less than the attitude deviation threshold, the collision risk trend is determined to be the third risk trend.

[0154] The third risk trend indicates that the vehicle currently faces no collision risk and can maintain normal driving status. When the safety distance margin is greater than or equal to the second margin threshold, the vehicle maintains a sufficient safety distance from the road boundary, and the attitude deviation is less than the attitude deviation threshold, indicating that the vehicle attitude is within the normal range.

[0155] In a possible implementation, when the safety distance margin is greater than or equal to the second margin threshold and the attitude deviation is less than the attitude deviation threshold, the system determines the collision risk trend as the third risk trend.

[0156] The processes shown in S601-S603 above achieve a refined classification assessment of collision risk by dividing the collision risk trend into three levels: high, medium, and low, and setting clear quantitative judgment thresholds.

[0157] like Figure 7 The diagram shown is a seventh flowchart of a vehicle control method provided in an embodiment of this application, which includes the following steps.

[0158] S701: When the collision risk trend is the first risk trend, control the vehicle to perform the first deceleration action and adjust the target driving trajectory to increase the distance between the vehicle and the road boundary.

[0159] The first deceleration action refers to the emergency deceleration operation corresponding to the first risk trend. Compared with the second deceleration action under the second risk trend, the first deceleration action has a higher deceleration intensity and a more urgent response priority.

[0160] Adjusting the target driving trajectory refers to making online corrections to the target driving trajectory generated in step S104 in order to increase the safe distance between the vehicle and the road boundary.

[0161] In a possible implementation, when the collision risk trend is the first risk trend, the system immediately triggers a first deceleration action, controlling the vehicle to perform emergency deceleration (e.g., reducing the vehicle speed to below 3 km / h), while simultaneously fine-tuning the target driving trajectory (e.g., shifting the trajectory away from the road boundary by a preset amount, such as within 5°) to increase the distance between the vehicle and the road boundary. After the first deceleration action is completed, the system continuously monitors the real-time relative status, and gradually restores normal driving speed and trajectory after the risk is eliminated.

[0162] Optionally, when the collision risk trend is the first risk trend, the embodiments of this application can also be implemented through... Figure 8 The steps shown trigger a user takeover prompt.

[0163] S702: When the collision risk trend is the second risk trend, control the vehicle to perform a second deceleration action and adjust the target driving trajectory so that the safety distance margin is greater than or equal to the second margin threshold.

[0164] The second deceleration action refers to a moderate deceleration operation corresponding to the second risk trend. Compared with the first deceleration action under the first risk trend, the second deceleration action has a lower deceleration intensity and a slower response priority.

[0165] In a possible implementation, when the collision risk trend is the second risk trend, the system controls the vehicle to perform a second deceleration action (e.g., reducing the vehicle speed to below 6 km / h) and dynamically adjusts the target driving trajectory to increase the safety distance margin to greater than or equal to the second margin threshold. The system continuously monitors changes in real-time relative status and collision risk trend until the risk level drops to the third risk trend.

[0166] S703: When the collision risk trend is the third risk trend, control the vehicle to maintain the current driving state and continuously monitor the real-time relative state between the vehicle and the road boundary.

[0167] Specifically, when the collision risk trend is at the third risk level, the system controls the vehicle to maintain its current speed and trajectory, without performing any additional deceleration or trajectory adjustment. Simultaneously, the system continuously monitors the real-time relative status of the vehicle and the road boundary, as well as the collision risk trend, to promptly trigger corresponding preventative actions when the risk level changes.

[0168] The processes shown in S701-S703 above achieve refined hierarchical prevention and control by implementing differentiated prevention and control actions according to the three levels of collision risk trends (i.e., the first risk trend, the second risk trend, and the third risk trend).

[0169] like Figure 8The diagram shown is an eighth flowchart of a vehicle control method provided in this application embodiment, including the following steps.

[0170] S801: After performing the first deceleration action, obtain the updated real-time relative state.

[0171] The updated real-time relative state refers to the new real-time spatial relationship between the vehicle and the road boundary after the first deceleration action and trajectory adjustment. This state is the data basis for evaluating the effectiveness of the prevention and control actions.

[0172] In a possible implementation, after the system performs the first deceleration action and trajectory adjustment, the system reacquires the real-time distance and attitude deviation between the vehicle and the road boundary as the updated real-time relative state.

[0173] S802: Determine the updated collision risk trend based on the updated real-time relative status.

[0174] The updated collision risk trend refers to the risk level and direction of change of the vehicle-road boundary collision, reassessed based on the updated real-time relative state after the initial deceleration and trajectory adjustment. Compared to the initial collision risk trend, the updated collision risk trend reflects the risk change after the implementation of preventative measures and is used to verify the effectiveness of the implemented preventative measures.

[0175] In a possible implementation, the system re-determines the collision risk trend based on the updated real-time relative state obtained in step S801. Specifically, the system recalculates the safety distance margin based on the updated real-time distance and the preset safety distance, and combines this with the updated attitude deviation, according to... Figure 6 The classification rules shown determine the updated collision risk trend.

[0176] S803: When the updated collision risk trend is still the first risk trend, trigger a user takeover prompt.

[0177] The user takeover prompt refers to a notification message used to remind the driver to take over control of the vehicle. This prompt can be output through acoustic, optical, or tactile means (such as alarm sounds, display screen prompts, steering wheel vibrations, etc.). When the initial deceleration action and trajectory adjustment have failed to effectively eliminate the collision risk, it indicates that the current risk situation has exceeded the system's automatic prevention and control capabilities, requiring driver intervention.

[0178] In a possible implementation, when the updated collision risk trend is still the first risk trend, the system triggers a user takeover prompt, reminding the driver to take over vehicle control in a timely manner.

[0179] The processes shown in S801-S803 above construct a complete security redundancy mechanism by conducting a secondary risk assessment after performing high-risk prevention and control actions and triggering user takeover when the risk is not eliminated.

[0180] like Figure 9 The diagram shown is a ninth flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0181] S901: Obtain the actual driving trajectory of the vehicle.

[0182] The actual driving trajectory refers to the path the vehicle actually travels after performing an avoidance maneuver. This actual driving trajectory can be obtained in real time through onboard positioning units (such as global navigation satellite systems, inertial measurement units, etc.). The actual driving trajectory is usually stored in the form of trajectory points, each containing information such as timestamp, latitude and longitude coordinates, heading angle, and vehicle speed.

[0183] In a possible implementation, after the vehicle performs an avoidance maneuver, the system records the vehicle's actual driving trajectory through the onboard positioning unit to obtain actual driving trajectory data.

[0184] S902: Determine the trajectory deviation based on the actual driving trajectory and the target driving trajectory.

[0185] Trajectory deviation refers to the degree of deviation between the actual driving trajectory and the target driving trajectory. This trajectory deviation can be quantified by calculating indicators such as lateral offset, longitudinal offset, or combined distance between the two trajectories. Lateral offset refers to the vertical distance between the actual trajectory point and the target trajectory at the same path mileage.

[0186] In a possible implementation, the system compares the actual driving trajectory obtained in step S901 with the target driving trajectory generated in step S104, and calculates the deviation between the two as the trajectory deviation.

[0187] S903: When the trajectory deviation is greater than or equal to the preset deviation threshold, the tracking parameters in the updated tracking strategy are corrected according to the trajectory deviation to obtain the corrected tracking strategy, and the corrected tracking strategy is used to track the road boundary.

[0188] The preset deviation threshold is used to determine whether the trajectory deviation has reached a critical value that requires correction of the tracking strategy.

[0189] Tracking parameters refer to the adjustable variables in a tracking strategy (such as tracking sensitivity, filter coefficients, tracking window size, etc.). A tracking strategy is a high-level control framework, and tracking parameters are the adjustable variables within that strategy.

[0190] In a possible implementation, when the trajectory deviation is greater than or equal to a preset deviation threshold, it indicates that the tracking accuracy of the current tracking strategy is insufficient, and the tracking strategy needs to be corrected. The system adjusts the tracking parameters in the updated tracking strategy (e.g., adjusting tracking sensitivity, filtering coefficients, etc.) based on the specific value of the trajectory deviation, resulting in a corrected tracking strategy. Subsequently, the system uses this corrected tracking strategy to track the road boundary. If the trajectory deviation is less than the preset deviation threshold, no correction is needed, and the system continues to operate with the current tracking strategy.

[0191] S904: Collect prevention and control data and build an optimized sample library.

[0192] Among them, prevention and control data refers to various data related to the prevention and control effect recorded during the driving process in the target scenario, including missed detection cases (situations where road boundaries are not detected and their scenario information), false detection cases (situations where non-road boundaries are mistakenly identified as road boundaries and their scenario information), stuck cases (situations where vehicles cannot drive smoothly in the target scenario and their status information), risk prevention and control data (time, location, level and corresponding prevention and control action records of risk events at all levels), and parameter adjustment data (time, reason, and parameter values ​​before and after each parameter adjustment).

[0193] The optimized sample library refers to the structured dataset formed by integrating the above-mentioned prevention and control data with the reinjection data and the actual vehicle test data.

[0194] In a possible implementation, the system collects prevention and control data for each target scenario and combines the prevention and control data with the reinjection data and real vehicle test data to construct an optimized sample library.

[0195] S905: Update the tracking parameters in the tracking strategy based on the optimized sample library.

[0196] The closed-loop optimization system refers to a complete mechanism that achieves adaptive improvement of tracking strategies and related parameters through a continuous cycle of four stages: "collection-verification-optimization-iteration". The collection stage acquires prevention and control data under the target scenario; the verification stage evaluates whether the current prevention and control effect is qualified; the optimization stage adjusts relevant parameters based on the verification results; and the iteration stage updates the optimized solution to the system and applies it to subsequent scenarios. The four stages are connected sequentially and cyclically.

[0197] In a possible implementation, the system or offline training platform uses machine learning algorithms to analyze and optimize high-frequency problem scenarios and their causes in the sample library, dynamically optimizing the parameters of the perception model, the thresholds of the post-processing logic, and the parameters of the risk prediction model. The optimized solution is synchronously updated to the vehicle control system's database for subsequent prevention and control of target scenarios.

[0198] It should be noted that S904-S905 are responsible for collecting data on a longer time scale, building a sample library, and performing iterative parameter optimization. Optionally, embodiments of this application may also utilize... Figure 11 The steps shown are responsible for detecting and correcting problems in real time during operation. Together, they form a complete closed-loop system from "real-time immediate correction" to "offline long-term evolution", which enables the prevention and control capabilities to continuously improve with the accumulation of operating time.

[0199] The processes shown in S901-S905 above achieve adaptive iterative optimization of the tracking strategy by performing closed-loop correction of the tracking strategy based on the deviation between the actual trajectory and the target trajectory after the avoidance action is executed, and by combining the collection of prevention and control data and the construction of a sample library.

[0200] like Figure 10 The diagram shown is a tenth flowchart of a vehicle control method provided in an embodiment of this application, including the following steps.

[0201] S1001: Obtain sample data of road boundaries in multiple scenarios.

[0202] Among them, multi-scenario road boundary sample data refers to road boundary sample data collected under different target scenario types (including urban intersections, underpasses, U-shaped areas, wide road boundaries, and narrow road boundaries), covering different lighting conditions (sunny days, cloudy days, and nighttime) and different road surface environments (asphalt roads, cement roads, and stained roads). The diversity of sample data ensures that the trained perception model has good generalization ability.

[0203] In a possible implementation, the system acquires road boundary sample data for multiple scenarios. A preset number of valid sample data are collected for each scenario. Each set of sample data includes the location, height, width, orientation, surrounding environmental imagery, and sensor signal strength information of the road boundary, and labels the road boundary type (e.g., concrete boundary, stone boundary, plastic boundary). After denoising, filtering, and labeling, the collected sample data is standardized in data format and coordinate system to form a standardized training dataset.

[0204] S1002: Based on multi-scenario road boundary sample data, the initial perception model is trained to generate an optimized perception model.

[0205] The initial perception model is a road boundary detection model built based on deep learning algorithms (such as convolutional neural network models). Convolutional neural network models extract spatial features from input data through multi-layer convolution and pooling operations, and can effectively identify typical patterns of road boundaries in images or point clouds.

[0206] In some examples, the scene adaptation layer is a specific network layer or sub-network structure in the road boundary detection model, whose parameters can be dynamically adjusted according to the scene features of the input data, thereby achieving scene adaptive perception.

[0207] In a possible implementation, the offline training platform inputs the multi-scene road boundary sample data processed in step S1001 into the initial perception model, uses deep learning algorithms to train the model, focuses on optimizing the scene adaptation layer of the model, so that the model can automatically adjust the perception parameters according to the characteristics of different target scene types.

[0208] S1003: Set an occlusion recognition branch in the optimized perception model.

[0209] The occlusion recognition branch is an independent sub-network structure in the perception model, specifically designed to identify occluded road boundary regions. This branch operates independently of the main detection branch and employs attention mechanisms or multi-scale feature fusion to enhance the model's ability to perceive weak features of road boundaries in occluded regions.

[0210] In a possible implementation, for scenarios prone to obstruction, such as U-turns under elevated roads and U-turns, the system adds an occlusion recognition branch to the perception model. This branch identifies obstructed road boundaries by fusing LiDAR point cloud data and surround-view camera image data, filling the perception gaps in obstructed areas.

[0211] S1004: Set a feature verification branch in the optimized perception model.

[0212] The feature verification branch is used to distinguish road boundaries from non-boundary objects (such as road stains and manhole covers). Road boundaries have typical geometric features such as abrupt height changes, limited width, and continuous extension, while road stains and manhole covers do not possess these features. The feature verification branch extracts these discriminative geometric features to perform secondary verification on the preliminary detection results of the main detection branch.

[0213] In a possible implementation, the system adds a feature verification branch to the perception model. This branch extracts typical geometric features of the road boundary (such as height, width, and direction continuity) and distinguishes them from features of non-boundary objects such as road stains and manhole covers, thus filtering out false perception signals.

[0214] S1005: Validate the optimized perception model.

[0215] One of the validation methods for the optimized perception model is refeedback validation. Refeedback validation refers to an offline validation method that involves playing back pre-recorded sensor data to the perception model to verify the model's detection accuracy on known labeled data. Unlike real-vehicle testing, refeedback validation uses a fixed test dataset, and the validation results are repeatable, facilitating performance comparisons between different model versions.

[0216] In a possible implementation, the system will perform refeeding verification and real-vehicle testing on the optimized perception model. Refeeding verification uses refeeding data from various target scenario types, while real-vehicle testing covers all target scenario types to ensure that the model's perception confidence level meets preset requirements (e.g., ≥0.85), and that the false negative rate and false positive rate meet preset requirements (e.g., false negative rate ≤1%, false positive rate ≤0.5%) under different scenarios. Perception confidence level refers to the model's degree of certainty regarding its detection results, ranging from 0 to 1; a higher value indicates greater confidence in the model's detection results. False negative rate refers to the proportion of all actual road boundaries that are not detected by the perception model. False positive rate refers to the proportion of all road boundaries output by the perception model that are not actually road boundaries. After successful verification, the optimized perception model is applied to the vehicle control method.

[0217] The processes shown in S1001-S1005 above effectively improve the recognition accuracy and robustness of the perception model in different scenarios by collecting road boundary sample data in multiple scenarios and training and optimizing the perception model, and by setting occlusion recognition branch and feature verification branch in the perception model.

[0218] like Figure 11 The diagram shown is an eleventh step in a vehicle control method provided in an embodiment of this application, including the steps described below.

[0219] S1101: Real-time verification of prevention and control effectiveness.

[0220] The effectiveness of the prevention and control measures is measured using multiple validation indicators, including the road boundary perception false detection rate, road boundary perception false detection rate, driving delay rate, and collision risk incidence rate. These indicators together constitute a quantitative evaluation system for the effectiveness of the prevention and control measures, comprehensively evaluating the overall performance of the system from three dimensions: perception quality, driving smoothness, and safety.

[0221] In a possible implementation, while the vehicle is traveling in the target scenario, the system verifies the prevention and control effect at a preset interval (e.g., every 50ms). The system compares various verification indicators with preset qualification conditions: when the road boundary perception missed detection rate is less than or equal to 1%, the false detection rate is less than or equal to 0.5%, the driving lag rate is less than or equal to 0.3%, and the collision risk occurrence rate is less than or equal to 0.1%, the prevention and control effect is deemed qualified.

[0222] S1102: When the prevention and control effect is not up to standard, implement abnormal handling.

[0223] If the verification reveals that the prevention and control effect is unsatisfactory (e.g., missed detection of road boundaries or vehicle stagnation), the system immediately pauses the current driving trajectory and analyzes the cause of the problem using real-time data. If the problem is determined to be due to insufficient road boundary data, the system supplements the road boundary data for the corresponding scenario and optimizes the perception model. If the problem is determined to be due to inappropriate post-processing logic adaptation, the system adjusts the detection threshold for large areas or the tracking logic. If the problem is determined to be due to an unreasonable risk classification threshold, the system fine-tunes the risk classification threshold. Once the cause of the problem is eliminated, the system resumes normal operation.

[0224] The processes shown in S1101-S1102 above ensure the continuous effectiveness and reliability of the prevention and control system by verifying the prevention and control effect in real time during the driving process in the target scenario and performing targeted anomaly handling when the prevention and control effect is unqualified.

[0225] like Figure 12 The diagram shown is a schematic flowchart of the twelfth step of a vehicle control method provided in an embodiment of this application, which includes the following steps.

[0226] S1201: Obtain vehicle travel route information.

[0227] The driving path information refers to the navigation path information used by the vehicle for its current journey. This information is provided by the vehicle's navigation path planning module and includes navigation elements such as road segments, turning points, and U-turn points along the path. Driving path information is typically stored as a sequence of path points, with each path point containing attributes such as latitude and longitude coordinates, heading angle, and curvature. Driving path information differs from the perception data in step S101; the former is prior path information provided by maps and navigation algorithms, while the latter is environmental perception information collected in real-time by onboard sensors.

[0228] In a possible implementation, the system obtains the vehicle's driving path information from the navigation path planning module. This driving path information contains the complete navigation path from the starting point to the destination, marking key driving nodes along the path (such as intersections, U-turn points, and turning points). The system can determine the locations of special operation points in the path by parsing the instruction sequence in the navigation path.

[0229] Optionally, obtaining vehicle travel path information also includes two sub-steps, S1201a and S1201b. S1201a can identify the target scene type, and S1201b is used to initialize the prevention and control mechanism to determine the prevention and control parameters. Together, the two sub-steps give the data obtained by S1201 substantial application value.

[0230] S1201a: Identify the target scene type based on the driving route information.

[0231] The target scenario types include U-turn scenarios, U-turn under elevated roads scenarios, U-turn scenarios at urban intersections, U-turn scenarios with wide curbs, and U-turn scenarios with narrow curbs. Each target scenario type has a corresponding set of feature parameters. For example, U-turn under elevated roads is characterized by obstructions and low lighting, while U-turn scenarios are characterized by a large U-turn radius and a U-shaped road boundary. Different types of U-turn scenarios impose different requirements on road boundary perception and collision prevention.

[0232] In a possible implementation, the vehicle control system determines the target scene type based on navigation instructions (such as U-turn instructions) in the driving path information, combined with the relative relationship between the vehicle's current position and key nodes in the path, using a scene recognition algorithm. The scene recognition algorithm compares and matches road geometric features (such as lane width, U-turn radius, and whether there is an overpass obstruction) in multi-source perception data with feature parameters of various target scene types to quickly complete scene classification.

[0233] S1201b: Initialize the prevention and control mechanism and determine the prevention and control parameters.

[0234] The control parameters include a boundary confidence threshold, a patch detection threshold, a risk grading threshold, and a trajectory fine-tuning trigger threshold. The boundary confidence threshold is used in step S203 to determine whether the sensing unit is effective; the patch detection threshold is used to determine the minimum number of consecutive sensing units required for an effective area; the risk grading threshold is used in steps S601-S603 to classify high, medium, and low risk trends; and the trajectory fine-tuning trigger threshold is used to determine whether a trajectory adjustment operation needs to be performed.

[0235] In a possible implementation, when a vehicle is identified as being in a target scenario, the system immediately activates the prevention and control mechanism and initializes relevant prevention and control parameters based on the target scenario type. For example, the system dynamically adjusts the boundary confidence threshold (default value set to 0.8) according to the scenario type, and sets the threshold for the number of consecutive units detected in a patch (default value set to 3), the safety distance margin threshold for risk classification, and the attitude deviation threshold. Simultaneously, the system synchronously calls the perception module, decision-making module, and control module to ensure that each module works collaboratively.

[0236] S1202: When the vehicle is identified as being in a target scene type based on the driving path information, the step of acquiring multi-source perception data is executed.

[0237] The system identifies whether the vehicle is currently in the target scenario type based on the driving path information obtained in step S1201.

[0238] In some examples, the system analyzes navigation instructions (such as U-turn instructions) in the driving path information and combines this with the relative relationship between the vehicle's current position and key nodes in the path to determine whether the vehicle is about to enter or is currently in a target scenario type. When the system identifies that the vehicle is in a target scenario type, it executes the operation of acquiring multi-source perception data in step S101 and initiates subsequent prevention and control procedures. If no target scenario type is identified, the system can continue to monitor the driving path information without initiating prevention and control procedures.

[0239] The process shown in S1201-S2102 above achieves scenario-based triggering of the prevention and control mechanism by identifying the scenario type based on the driving path information before acquiring multi-source perception data, thus avoiding unnecessary consumption of computing resources in irrelevant scenarios.

[0240] like Figure 13 The diagram shown is a schematic representation of a vehicle architecture according to an embodiment of this application. The vehicle includes a processor 1301, a memory 1302, and a bus 1303. The processor 1301 and the memory 1302 are connected via the bus 1303. The memory 1302 is used to store programs, and the processor 1301 is used to run the programs. When the programs run, they execute the vehicle control method provided in this application, which includes the following steps:

[0241] Acquire road boundary perception data and vehicle status data during vehicle operation;

[0242] Based on the road boundary perception data and the state data, the current spatial pose of the vehicle relative to the road boundary is determined.

[0243] Based on the current spatial pose relationship, the road boundary is dynamically tracked to generate dynamic tracking results;

[0244] Based on the dynamic tracking results and the status data, the target driving trajectory of the vehicle is generated;

[0245] During the process of the vehicle traveling along the target driving trajectory, the collision risk trend is determined based on the real-time relative state between the vehicle and the road boundary.

[0246] Based on the collision risk trend, the vehicle is controlled to perform corresponding avoidance maneuvers.

[0247] Optionally, based on the road boundary sensing data, road boundary area information is generated, including:

[0248] The sensing area corresponding to the road boundary sensing data is divided into multiple sensing units;

[0249] Obtain the boundary confidence level corresponding to each of the sensing units;

[0250] Multiple sensing units whose boundary confidence levels meet the preset confidence conditions and whose positions are adjacent and uninterrupted are identified as effective region units;

[0251] Based on each of the effective area units, road boundary area information is generated.

[0252] Optionally, based on the dynamic tracking results and the state data, a target driving trajectory of the vehicle is generated, including:

[0253] Based on the dynamic tracking results, the location and orientation of the road boundary are determined;

[0254] The initial safe distance between the vehicle and the road boundary is determined based on the location and direction of the road boundary and the current location in the status data.

[0255] Determine the target safety distance based on the target scenario type and the current state of the vehicle;

[0256] When the initial safe distance is less than the target safe distance, the target driving trajectory is adjusted to increase the distance between the vehicle and the road boundary until the safe distance between the vehicle and the road boundary is greater than or equal to the target safe distance.

[0257] Optionally, based on the current spatial pose relationship, the road boundary is dynamically tracked to generate dynamic tracking results, including:

[0258] Based on the current spatial pose relationship, determine the current driving progress of the vehicle in the scene;

[0259] Based on the current driving progress, the tracking strategy for the road boundary is adjusted to adapt to the scene, resulting in an updated tracking strategy.

[0260] Based on the updated tracking strategy, the road boundary is tracked to generate the dynamic tracking result.

[0261] Optionally, the real-time relative state includes the real-time distance and attitude deviation between the vehicle and the road boundary;

[0262] Based on the real-time relative state between the vehicle and the road boundary, a collision risk trend is determined, including:

[0263] The safety distance margin is determined based on the real-time distance and the preset safety distance;

[0264] The collision risk trend is determined based on the safety distance margin and the attitude deviation.

[0265] Optionally, the collision risk trend includes a first risk trend, a second risk trend, and a third risk trend; the risk level of the first risk trend is higher than that of the second risk trend, and the risk level of the second risk trend is higher than that of the first risk trend.

[0266] Determining the collision risk trend based on the safety distance margin and the attitude deviation includes:

[0267] When the safety distance margin is less than the first margin threshold, or the attitude deviation is greater than or equal to the attitude deviation threshold, the collision risk trend is determined to be the first risk trend.

[0268] When the safety distance margin is greater than or equal to the first margin threshold and less than the second margin threshold, the collision risk trend is determined to be the second risk trend.

[0269] When the safety distance margin is greater than or equal to the second margin threshold and the attitude deviation is less than the attitude deviation threshold, the collision risk trend is determined to be the third risk trend.

[0270] Wherein, the first margin threshold is less than the second margin threshold.

[0271] Optionally, based on the collision risk trend, the vehicle is controlled to perform a corresponding avoidance maneuver, including:

[0272] When the collision risk trend is the first risk trend, the vehicle is controlled to perform a first deceleration action, and the target driving trajectory is adjusted to increase the distance between the vehicle and the road boundary;

[0273] When the collision risk trend is the second risk trend, the vehicle is controlled to perform a second deceleration action, and the target driving trajectory is adjusted so that the safety distance margin is greater than or equal to the second margin threshold.

[0274] When the collision risk trend is the third risk trend, the vehicle is controlled to maintain its current driving state, and the real-time relative state between the vehicle and the road boundary is continuously monitored.

[0275] Optionally, when the collision risk trend is the first risk trend, the method further includes:

[0276] After performing the first deceleration action, obtain the updated real-time relative state;

[0277] Based on the updated real-time relative state, determine the updated collision risk trend;

[0278] If the updated collision risk trend is still the same as the first risk trend, a user takeover prompt is triggered.

[0279] Optionally, after performing the avoidance action, the method further includes:

[0280] Obtain the actual driving trajectory of the vehicle;

[0281] Based on the actual driving trajectory and the target driving trajectory, determine the trajectory deviation;

[0282] When the trajectory deviation is greater than or equal to a preset deviation threshold, the tracking parameters in the updated tracking strategy are corrected according to the trajectory deviation to obtain a corrected tracking strategy, and the corrected tracking strategy is used to track the road boundary.

[0283] In summary, by establishing a dynamic collaborative mechanism of perception, decision-making, and control, along with a forward-looking risk prediction system, the vehicle achieves adaptive tracking and proactive prevention and control of road boundaries in target scenarios, effectively improving driving safety and traffic flow.

[0284] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0285] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A vehicle control method, characterized in that, include: Acquire road boundary perception data and vehicle status data during vehicle operation; Based on the road boundary perception data and the state data, the current spatial pose of the vehicle relative to the road boundary is determined. Based on the current spatial pose relationship, the road boundary is dynamically tracked to generate dynamic tracking results; Based on the dynamic tracking results and the status data, the target driving trajectory of the vehicle is generated; During the process of the vehicle traveling along the target driving trajectory, the collision risk trend is determined based on the real-time relative state between the vehicle and the road boundary. Based on the collision risk trend, the vehicle is controlled to perform corresponding avoidance maneuvers.

2. The method according to claim 1, characterized in that, Based on the road boundary perception data and the state data, the current spatial pose of the vehicle relative to the road boundary is determined, including: The sensing area corresponding to the road boundary sensing data is divided into multiple sensing units; Obtain the boundary confidence level corresponding to each of the sensing units; Multiple sensing units whose boundary confidence levels meet the preset confidence conditions and whose positions are adjacent and uninterrupted are identified as effective region units; Based on each of the effective area units, road boundary area information is generated; Based on the road boundary area information and the state data, the current spatial pose of the vehicle relative to the road boundary is determined.

3. The method according to claim 1, characterized in that, Based on the dynamic tracking results and the status data, the target driving trajectory of the vehicle is generated, including: Based on the dynamic tracking results, the location and orientation of the road boundary are determined; The initial safe distance between the vehicle and the road boundary is determined based on the location and direction of the road boundary and the current location in the status data. Determine the target safety distance based on the target scenario type and the current state of the vehicle; When the initial safe distance is less than the target safe distance, the target driving trajectory is adjusted to increase the distance between the vehicle and the road boundary until the safe distance between the vehicle and the road boundary is greater than or equal to the target safe distance.

4. The method according to claim 1, characterized in that, Based on the current spatial pose relationship, the road boundary is dynamically tracked to generate dynamic tracking results, including: Based on the current spatial pose relationship, determine the current driving progress of the vehicle in the scene; Based on the current driving progress, the tracking strategy for the road boundary is adjusted to adapt to the scene, resulting in an updated tracking strategy. Based on the updated tracking strategy, the road boundary is tracked to generate the dynamic tracking result.

5. The method according to claim 1, characterized in that, The real-time relative state includes the real-time distance and attitude deviation between the vehicle and the road boundary; Based on the real-time relative state between the vehicle and the road boundary, a collision risk trend is determined, including: The safety distance margin is determined based on the real-time distance and the preset safety distance; The collision risk trend is determined based on the safety distance margin and the attitude deviation.

6. The method according to claim 5, characterized in that, The collision risk trends include a first risk trend, a second risk trend, and a third risk trend; The risk level of the first risk trend is higher than that of the second risk trend, and the risk level of the second risk trend is higher than that of the first risk trend. Determining the collision risk trend based on the safety distance margin and the attitude deviation includes: When the safety distance margin is less than the first margin threshold, or the attitude deviation is greater than or equal to the attitude deviation threshold, the collision risk trend is determined to be the first risk trend. When the safety distance margin is greater than or equal to the first margin threshold and less than the second margin threshold, the collision risk trend is determined to be the second risk trend. When the safety distance margin is greater than or equal to the second margin threshold and the attitude deviation is less than the attitude deviation threshold, the collision risk trend is determined to be the third risk trend. Wherein, the first margin threshold is less than the second margin threshold.

7. The method according to claim 6, characterized in that, Based on the collision risk trend, control the vehicle to perform corresponding avoidance actions, including: When the collision risk trend is the first risk trend, the vehicle is controlled to perform a first deceleration action, and the target driving trajectory is adjusted to increase the distance between the vehicle and the road boundary; When the collision risk trend is the second risk trend, the vehicle is controlled to perform a second deceleration action, and the target driving trajectory is adjusted so that the safety distance margin is greater than or equal to the second margin threshold. When the collision risk trend is the third risk trend, the vehicle is controlled to maintain its current driving state, and the real-time relative state between the vehicle and the road boundary is continuously monitored.

8. The method according to claim 7, characterized in that, When the collision risk trend is the first risk trend, the method further includes: After performing the first deceleration action, obtain the updated real-time relative state; Based on the updated real-time relative state, determine the updated collision risk trend; If the updated collision risk trend is still the same as the first risk trend, a user takeover prompt is triggered.

9. The method according to claim 4, characterized in that, After performing the avoidance action, the method further includes: Obtain the actual driving trajectory of the vehicle; Based on the actual driving trajectory and the target driving trajectory, determine the trajectory deviation; When the trajectory deviation is greater than or equal to a preset deviation threshold, the tracking parameters in the updated tracking strategy are corrected according to the trajectory deviation to obtain a corrected tracking strategy, and the corrected tracking strategy is used to track the road boundary.

10. A vehicle, characterized in that, include: Processor, memory, and bus; The processor and the memory are connected via the bus; The memory is used to store a program, and the processor is used to run the program, wherein the program is executed by the processor to perform the vehicle control method according to any one of claims 1 to 9.

Images