Vehicle control method, vehicle and electronic device for continuous curve scenario

Abstract

This invention belongs to the field of road vehicle driving control, and relates to a vehicle control method, vehicle, and electronic equipment for continuous curve scenarios. When a vehicle is in a continuous curve scenario, the method constructs the road, vehicle, and target states in the continuous curve reference line coordinate system to determine the vehicle's effective forward driving area. Surrounding targets are screened, and for critical hazardous targets, multi-factor adaptive correction collision time, predicted intrusion time, and multi-dimensional expected collision time are calculated and integrated into a multi-dimensional intrusion expected collision time. Combining environmental risk, information completeness, vehicle control capability, multi-dimensional intrusion expected collision time, and driver takeover mismatch risk, a judgment mechanism combining the upper limit of the allowable level and the actual requested level is used to determine the final execution level. The method outputs the optimal action mode and corresponding control commands adapted to the continuous curve scenario, thereby improving the consistency and accuracy of risk identification, risk quantification, and control decisions in continuous curve scenarios.

Description

Technical Field

[0001] This invention belongs to the field of road vehicle driving control systems, specifically relating to a vehicle control method, vehicle, and electronic equipment for continuous curve scenarios. Background Technology

[0002] In recent years, with the rapid development of intelligent driving technology, the application rate of Automatic Emergency Braking (AEB), Forward Collision Warning (FCW), Lane Keeping Assist (LKA), and graded driving intervention systems in heavy vehicles and passenger vehicles has been continuously increasing. However, in complex highway scenarios with continuous curves in multi-lane sections, existing intelligent driving safety control systems face the engineering challenge of both false triggering and missed triggering. Specifically, continuous curve scenarios are characterized by continuously changing curvature, close connection between adjacent curves, shrinking visibility, frequent obstruction, and rapid changes in the relative heading and lateral intrusion relationship of targets. This means that the system not only needs to handle the longitudinal approach risk of targets in front of the vehicle in its own lane, but also must cope with the intrusion threat to the vehicle's future passage space from targets in adjacent lanes that cross, insert, or move abnormally. Traditional control methods based on fixed time-of-collision (TTC) thresholds, minimum distance thresholds, or simple lane associations may, on the one hand, fail to accurately distinguish between normal adjacent lane targets and real intrusive dangerous targets, resulting in excessive braking, disrupting normal traffic flow and increasing the risk of being rear-ended; on the other hand, they may fail to depict the future intrusion trend and multi-dimensional convergence characteristics of targets in continuous curves in time, leading to insufficient system response before dangerous targets actually enter the vehicle's driving space, thus causing serious collision accidents.

[0003] While some existing technologies attempt to improve system performance by optimizing target recognition, collision time calculation, or warning triggering logic in curve scenarios, the following technical bottlenecks remain in complex conflict scenarios involving continuous curves: First, existing methods primarily characterize risks in continuous curve scenarios using single time or distance indicators, making it difficult to simultaneously cover continuous approach risks, sequential intrusion risks, and lateral safety risks. For example, Chinese patent CN110562222B discloses an emergency braking control method for curve scenarios, which mainly triggers braking control through target recognition, driving path determination, and minimum collision distance calculation. Although this solution considers the impact of curve conditions on collision risk assessment, its core still revolves around a single longitudinal collision distance and indicator, insufficiently considering the risks of multi-target intrusion conflicts in continuous curves, and failing to effectively describe the true degree of danger when different conflict mechanisms coexist, such as simultaneous approach, sequential intrusion, and excessively close oblique distance with other vehicles. Second, existing technologies lack a unified fusion mechanism for multi-dimensional collision indicators. Chinese patent CN109710892A uses statistical extreme values ​​of TTC or PET for road traffic safety evaluation. Its focus is on analyzing road traffic safety levels, making it more suitable for safety evaluations at the road infrastructure or intersection levels. However, it is difficult to directly serve real-time graded intervention control for vehicles in continuous curves. In other words, while this type of solution involves safety indicators such as TTC and PET, it remains at the evaluation level and has not formed a complete technical chain that can directly drive real-time vehicle intervention decisions. Furthermore, existing solutions still do not adequately utilize the coupling relationships between multi-source risk variables in curve scenarios. In real continuous curve traffic flow, variables such as curvature, rate of curvature change, sight distance attenuation, degree of obstruction, target relative heading, elevation difference, and relative acceleration are usually significantly correlated. If these variables are simply isolated or used only to correct a single threshold without forming a unified risk quantification core, the system may be insensitive to hazard accumulation and insufficiently responsive when hazard rapidly approaches, thus missing the optimal intervention opportunity.

[0004] In summary, there is an urgent need for an intelligent driving hierarchical intervention control method that can be applied to continuous curve scenarios, taking into account continuous changes in road curvature, sight distance attenuation, future target occupancy conflict relationships, and multi-dimensional trajectory convergence characteristics. This method should not only be able to stably identify truly dangerous intrusive targets, but also integrate multi-factor adaptive correction of collision time, predicted post-intrusion time, and multi-dimensional expected collision time to form a new risk measurement index that can directly serve real-time hierarchical intervention control. This would effectively solve the problems of incomplete risk representation, inconsistent control decisions, and insufficient adaptability and safety of existing technologies in continuous curve scenarios. Summary of the Invention

[0005] Given the shortcomings and deficiencies of existing technologies in continuous curve scenarios, such as lagging identification of dangerous targets, inaccurate scene identification, distorted collision time estimation, inability of a single indicator to fully characterize intrusive risks, and disconnect between intervention level determination and curve constraints, as well as the continued use of straight-line strategies for control actions, the purpose of this invention is to provide a vehicle control method, vehicle, and electronic equipment for continuous curve scenarios. This method establishes a risk quantification basis suitable for complex conflict scenarios in continuous curves by recognizing continuous curve scenarios, constructing a unified road and vehicle state, generating a forward effective driving area, screening dangerous targets, constructing a multi-factor adaptive correction collision time, calculating the predicted intrusion time, calculating multi-dimensional expected collision time, and fusing a unified risk indicator. Based on this, the unified risk indicator, along with environmental risk, information completeness, driver takeover mismatch risk, and vehicle control capability, are incorporated into the graded intervention decision-making process to achieve reasonable output of actions such as deceleration, lane keeping enhancement, enhanced control, and emergency control, thereby improving the consistency and accuracy of risk identification, risk quantification, and control decisions in continuous curve scenarios.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A vehicle control method for continuous curve scenarios, comprising the following steps: Step S1. Collect data and identify whether the vehicle is currently in a continuous curve scenario. If so, construct a unified description of the road state, the vehicle state, and the target state in the continuous curve reference line coordinate system, and determine the effective forward driving area of ​​the vehicle. Step S2. Combining the predicted occupancy relationship between the target and the vehicle's forward effective driving area, the geometric correlation between the target and the vehicle, the target's line-of-sight sensitivity, and the continuity of continuous curves, screen surrounding targets to identify key hazardous targets. Step S3. For the identified key hazardous targets, a multi-factor adaptive correction collision time is constructed by integrating relative angle, elevation difference, relative acceleration, curvature-line-of-sight coupling relationship, and neural network uncertainty on the basic collision time model. At the same time, the predicted intrusion time is calculated based on the entry and exit times of the conflict area occupied by the vehicle and the target in the future envelope, and the multi-dimensional expected collision time is obtained based on the multi-dimensional equivalent separation distance under the reference line of continuous curves. The three are then integrated into a multi-dimensional intrusion expected collision time. Step S4. Combining environmental risk, information completeness, vehicle control capability, multi-dimensional intrusion expected collision time, and driver takeover mismatch risk, a two-layer judgment mechanism combining the upper limit of the allowable level and the actual request level is adopted to determine the final execution level and output the optimal action mode adapted to the continuous curve scenario. Step S5. Based on the final execution level and optimal action mode, output control results adapted to the continuous curve scenario, including longitudinal control, lateral control and lateral envelope tightening.

[0007] As a preferred embodiment of the present invention, in step S1, when determining whether the vehicle is currently in a continuous curve scenario, a continuous curve recognition score is first calculated based on the equivalent curvature, curvature change rate, distance between the current curve and the next curve, and current forward visible distance within the current aiming interval. The continuous curve recognition score is then mapped to a continuous curve scenario confidence level. Then, an entry threshold is used. With exit threshold Separation hysteresis strategy, when When entering a continuous curve scene, when Furthermore, it exits the continuous curve scenario after several control cycles.

[0008] As a preferred embodiment of the present invention, the road state in step S1 includes the equivalent curvature, curvature change rate, road longitudinal slope, forward visibility distance, occlusion intensity factor, and estimated road surface adhesion coefficient within the pre-aiming interval; the vehicle state includes the arc length coordinates and lateral offset of the vehicle in the continuous curve reference line coordinate system, the vehicle speed, the vehicle longitudinal acceleration, the vehicle's heading angle relative to the reference line, and the vehicle's yaw rate; the target state includes the arc length coordinates and lateral offset of the target in the continuous curve reference line coordinate system, the target speed, the target longitudinal acceleration, the target's heading angle relative to the reference line, and the target's elevation difference relative to the vehicle; in the continuous curve scenario, the effective forward driving area of ​​the vehicle is constructed based on the vehicle speed, curvature, cross slope, and the continuity of the continuous curves, and the effective lateral width of this area is: ; in, Indicates the position of arc length The effective forward driving width at the location; Indicates the current lane width; Indicates the safety envelope boundary of this vehicle; Indicates the curvature influence coefficient; Indicates the position of arc length Local curvature at that point; Indicates the cross slope influence coefficient; Indicates the position of arc length Superelevation or transverse slope at the location; The aiming length is: ; in, Indicates the pre-aiming length of the effective forward driving area; Indicates the basic aiming length; Indicates the speed adjustment coefficient; Indicates the speed of this vehicle; Indicates the enhancement coefficient for continuous curves; This represents the continuity factor for consecutive curves, determined based on the rate of curvature change and the distance between the current curve and the next curve.

[0009] As a preferred embodiment of the present invention, in step S2, after obtaining the effective forward driving area of ​​the vehicle, for any target, the predicted occupancy area within the prediction time window is generated based on the target's short-term trajectory prediction result, and the degree of overlap between the predicted occupancy area and the effective forward driving area of ​​the vehicle is calculated to characterize the predicted occupancy relationship between the target and the effective forward driving area of ​​the vehicle; at the same time, the geometric correlation between the target and the vehicle is determined by comprehensively considering the effective forward distance, lateral offset, and relative heading of the target along the continuous curve reference line direction; the sight distance sensitivity of the target is determined by comprehensively considering the occupancy intensity factor, forward visibility distance, and relative elevation difference between the target and the vehicle; finally, the hazard score of the target is calculated, and when the hazard score of the target exceeds the threshold, it is included in the dangerous target set, and the target with the highest hazard score is regarded as the critical dangerous target.

[0010] As a preferred embodiment of the present invention, the basic collision time model in step S3 is: ; in, Indicates the first The base collision time for each target; Indicates the effective forward distance of the target along the reference line of the continuous curves; This represents the equivalent closed velocity of the target relative to the vehicle along the reference line in the tangential direction. ; This represents a positive integer that prevents the denominator from being zero. Indicates the speed of this vehicle; Indicates the first The speed of the target; This indicates the heading angle of the vehicle relative to the tangential direction of the reference line of the continuous curves; The heading angle represents the target's heading relative to the tangential direction of the reference line of the continuous curves; The neural network is a lightweight temporal neural network, which is one of GRU, LSTM or TCN. By inputting the scene-target joint feature vectors of the current time and the historical time into the neural network, the adaptive correction coefficients and uncertainties of the dangerous target are obtained after network processing. The multi-factor adaptive correction collision time is: ; in, Indicates the first Multi-factor adaptive correction of collision time for each target; This represents the adaptive correction coefficients of the neural network output, used to learn and correct the basic collision time. Represents the equivalent curvature normalization quantity; Normalized quantity representing the absolute value of the rate of change of curvature; Indicates the current forward visible distance; This indicates a reference value for the forward visible distance; Indicates the occlusion intensity factor; The enhancement term representing the relative heading difference between the target and the vehicle. Indicates the relative heading angle. This indicates the relative elevation difference between the target and the vehicle. This represents the elevation difference normalization constant. This indicates the relative acceleration between the target and the vehicle along the reference line. This represents the normalized constant of relative acceleration; This indicates the uncertainty in the output of the neural network; to This represents the weighting coefficient of the corresponding influencing factor.

[0011] As a preferred embodiment of the present invention, the multidimensional equivalent separation distance in step S3 is: ; in, Indicates the first The target at the predicted time The multidimensional equivalent separation distance; express The difference in arc length between the target vehicle and the vehicle itself in the direction of the reference line at any given moment; express The relative deviation between the target vehicle and the vehicle in the lateral direction at any given time; express The relative difference in elevation between the target vehicle and the vehicle itself at any given time; Represents the relative heading angle at the predicted time; , and These represent the weighting coefficients for lateral, elevation, and relative heading angles, respectively. This represents the characteristic length constant, with values ​​ranging from 0.8m to 1m. The most dangerous moment for each target in the prediction time domain is determined based on the multidimensional equivalent separation distance, and the multidimensional expected collision time is calculated based on the equivalent multidimensional separation distance and its rate of change at the most dangerous moment. Finally, based on the conflict topology relationship between the target and the vehicle, the weights of multi-factor adaptive correction of collision time, predicted post-intrusion time and multi-dimensional expected collision time are dynamically allocated; among which, the conflict topology relationship includes same-direction approximation, lateral safety and temporal intrusion. After obtaining the weights, the expected collision time for multidimensional intrusion is calculated as follows: ; in, Indicates the first Expected collision time for multi-dimensional intrusion of multiple targets; This indicates a multi-factor adaptive correction of the collision time; Indicates the predicted intrusion time; Indicates the expected collision time in multiple dimensions; This represents a positive integer that prevents the denominator from being zero. , and These represent the weights of the multi-factor adaptive correction collision time, the predicted intrusion time, and the multi-dimensional expected collision time, respectively.

[0012] As a preferred embodiment of the present invention, in step S4, the environmental risk calculation is determined by comprehensively considering curvature risk, longitudinal slope risk, line-of-sight risk, and low adhesion risk; the information completeness calculation is determined by comprehensively considering lane boundary recognition completeness, target tracking stability, and target trajectory prediction stability; and the vehicle control capability calculation is determined by comprehensively considering the currently available longitudinal deceleration. Longitudinal deceleration required to avoid current conflict The available lateral acceleration corresponding to the current available lateral stability margin Lateral acceleration required for current conflict avoidance The available free space and the space required to perform the current evasive action can be used to determine this.

[0013] As a preferred embodiment of the present invention, step S4 involves multidimensional intrusion into the expected collision time. Mapped to the target risk quantity, and the first... The tightening trend of multidimensional intrusion expected collision time for each target and the overall interactive risk of the dangerous target set; whereby the target risk quantity is expressed as: ; in, Indicates the first The risk quantification value of each objective; Indicates the reference time constant; Indicates the first The conservative enhancement terms for each objective are determined based on the expected collision time of multi-dimensional intrusion, the continuity of continuous curves, and the completeness of information. This represents a positive integer that prevents the denominator from being zero. Then, the upper limit of the allowable level is determined based on scenario constraints, vehicle control capabilities, and information completeness. The comprehensive risk is calculated by considering environmental risk, overall interaction risk, information completeness, and driver takeover mismatch risk. The current actual request level is determined based on the comprehensive risk, the minimum expected collision time of multi-dimensional intrusion, and the tightening trend of the maximum expected collision time of multi-dimensional intrusion.

[0014] As a preferred embodiment of the present invention, in step S4, there are multiple action modes within the same execution level, and the action mode with the smallest comprehensive cost function is selected as the optimal action mode; the comprehensive cost function is composed of longitudinal impact, lateral deviation, driving comfort and rule consistency.

[0015] As a preferred embodiment of the present invention, step S5 generates human-computer interaction prompts that match the current execution level and action mode while outputting control commands.

[0016] The present invention also provides an electronic device, comprising: one or more processors and a memory; wherein the memory is used to store one or more programs or instructions, and when the one or more programs or instructions are executed by the one or more processors, the one or more processors implement the vehicle control method for continuous curve scenarios described above.

[0017] Advantages and beneficial effects of the present invention: (1) This invention proposes a continuous curve scene recognition and hysteresis switching mechanism. By using the continuous curve recognition score, scene confidence, and entry / exit separation criteria, it can stably determine whether a vehicle is in a continuous curve scene. This solution can reduce the problem of frequent scene switching caused by local curvature fluctuations, short-term occlusion, or changes in sight distance, thereby improving the consistency of subsequent risk assessment and control decisions.

[0018] (2) This invention establishes a unified state description method for road-vehicle-target under the reference line coordinate system of continuous curves. It no longer relies solely on straight-line distance or simple lane assignment to determine risk, but uses information such as arc length position, lateral offset, and relative heading to depict the real spatial relationship between vehicles. This scheme can more accurately reflect the actual operating state in continuous curves and reduce the distortion of risk estimation caused by curve geometry.

[0019] (3) This invention proposes a method for screening dangerous targets in the future travel space of the vehicle. By constructing a forward effective driving area and comprehensively considering the target prediction occupancy relationship, geometric correlation, sight distance sensitivity, and continuity of continuous curves, surrounding targets are screened. This scheme breaks through the traditional method of only identifying the nearest target or the vehicle in front in the same lane, and can identify potential dangerous targets with an intrusion tendency in adjacent lanes earlier, thus improving the accuracy of target identification in complex scenarios.

[0020] (4) This invention constructs a multi-factor adaptive correction collision time, which combines factors such as curvature, rate of change of curvature, sight distance attenuation, occlusion, relative angle, elevation difference and relative acceleration with the neural network correction results, and introduces an uncertainty tightening mechanism. This scheme takes into account both the interpretability of the physical model and the nonlinear representation capability of the data-driven model, and can improve the accuracy and robustness of collision time estimation in continuous curve scenarios.

[0021] (5) This invention proposes a multi-dimensional intrusion expected collision time fusion index, which integrates collision time, predicted intrusion time and multi-dimensional expected collision time to characterize the same-direction approach risk, sequential intrusion risk and lateral safety risk respectively. This scheme overcomes the limitation that a single time index is difficult to fully describe the complex conflict mechanism of continuous curves and realizes the unified quantification of multiple types of intrusion risks.

[0022] (6) This invention proposes an adaptive weight allocation mechanism based on conflict topology, which can dynamically adjust the proportion of various time indicators in unified risk quantification according to the same-direction approach, sequential intrusion, and lateral safety relationship between the target and the vehicle. This scheme avoids the problem that fixed weights or simple averaging methods are insufficient in characterizing the dominant risk mechanism, and makes the risk indicators more sensitive to critical dangerous states.

[0023] (7) This invention proposes a two-tiered hierarchical intervention control mechanism that combines the upper limit of the permitted level with the actual requested level, and incorporates environmental risk, information completeness, driver takeover mismatch risk and vehicle control capability into the decision-making process. This scheme makes the deceleration, lane keeping enhancement, enhanced control and emergency control actions output by the system more in line with the constraints of continuous curves, which can improve control safety and smoothness, and reduce false triggering and unreasonable strong intervention. Attached Figure Description

[0024] The present invention will be described and illustrated in detail below with reference to the accompanying drawings and through a detailed description of the embodiments.

[0025] Figure 1 The flowchart of the vehicle control method for continuous curve scenarios provided by the present invention is shown. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to examples and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0027] like Figure 1 As shown, this embodiment provides a vehicle control method for continuous curve scenarios, which includes the following steps: Step S1. Collect, time-align, and preprocess road perception data, vehicle motion data, and target detection data during vehicle operation. Identify whether the vehicle is currently in a continuous curve scenario. Construct a unified description of the road state, vehicle state, and target state in the continuous curve reference line coordinate system. Determine the forward effective driving area to provide common input for subsequent dangerous target identification, multi-dimensional intrusion expected collision time calculation, and intervention level determination.

[0028] Furthermore, in this embodiment, step S1 specifically includes the following steps: Step S1.1. Continuous Curve Recognition and Scene Confidence Calculation: Within the current aiming range, when the straight transition section between adjacent curves is short, the road curvature continuously changes, and the forward visibility distance is affected by the terrain and occlusion of the continuous curves, the system needs to determine whether the vehicle has entered a continuous curve scenario. Therefore, a continuous curve recognition score is first constructed:

[0029] in, Indicates time The continuous curve recognition score; It represents the normalized amount of the equivalent curvature within the preview interval; Normalized quantity representing the absolute value of the rate of change of curvature; This indicates the distance between the current curve and the next curve; Indicates the spacing reference value; Indicates the current forward visible distance; This indicates a reference value for the visible distance; , , and Represents the weighting coefficients of each characteristic term; symbol This means that only the non-negative part of the corresponding quantity is retained.

[0030] Furthermore, the continuous curve recognition score is mapped to the confidence level of the continuous curve scene:

[0031] in, Indicates the confidence level for a continuous curve scenario; Indicates the mapping slope parameter; This indicates the threshold for identifying continuous curves.

[0032] To avoid frequent scene state switching near curve exits due to local fluctuations in curvature or view distance, this invention employs an entry threshold. With exit threshold Separation hysteresis strategy: when When entering a continuous curve scene, when Furthermore, it exits the continuous curve scenario after several control cycles.

[0033] Step S1.2. Road State Vector Construction: After confirming the continuous curve scenario, a unified description of the road geometry and perception conditions is needed. The road state vector is defined as follows:

[0034] in, Represents the road state vector; Indicates the equivalent curvature within the current aiming interval; Indicates the rate of change of curvature; Indicates the longitudinal slope of the road; Indicates superelevation or cross slope; Indicates the forward visible distance; This indicates the shading intensity factor caused by guardrails, mountains, slopes, or large vehicles. This represents the current estimated value of the road surface adhesion coefficient; This is the transpose symbol.

[0035] In this embodiment, the road state vector is not simply used for environmental recording. Instead, curvature, rate of change of curvature, longitudinal slope, sight distance, and occlusion factor are continuously incorporated into subsequent hazard target determination, multi-factor adaptive correction of collision time construction, and action mode selection.

[0036] Step S1.3. Unify the expression of the vehicle's state and the target state: To describe the relationship between the vehicle and surrounding targets in the same reference coordinate system, we define the vehicle's state vector and the... The target state vectors are as follows:

[0037]

[0038] in, This represents the vehicle's state vector; and These represent the arc length coordinates and lateral offset of the vehicle in the continuous curve reference line coordinate system, respectively; Indicates the speed of this vehicle; This indicates the longitudinal acceleration of the vehicle; This indicates the heading angle of the vehicle relative to the reference line; This indicates the yaw rate of the vehicle. Indicates the first A target state vector; and These represent the arc length coordinates and lateral offset of the target in the same reference line coordinate system, respectively; and These represent the target velocity and the target longitudinal acceleration, respectively. This indicates the heading angle of the target relative to the reference line; This indicates the elevation difference between the target and the vehicle itself. This represents the prior probability that the target will occupy the vehicle's forward effective driving area within the prediction time window.

[0039] Step S1.4. Calculation of forward effective driving area parameters: To facilitate subsequent assessment of hazardous targets within the vehicle's future travel space, this invention constructs a forward effective driving area in a continuous curve scenario. The effective lateral width of this area is defined as follows:

[0040] in, Indicates the position of arc length The effective forward driving width at the location; Indicates the current lane width; Indicates the safety envelope boundary of this vehicle; Indicates the curvature influence coefficient; Indicates the position of arc length Curvature at that point; Indicates the cross slope influence coefficient; Indicates the position of arc length The location is either extremely high or has a transverse slope.

[0041] Accordingly, the aiming length of the forward effective driving area is defined as:

[0042] in, Indicates the pre-aiming length of the effective forward driving area; Indicates the basic aiming length; Indicates the speed adjustment coefficient; Indicates the speed of this vehicle; Indicates the enhancement coefficient for continuous curves; This represents the continuity factor of continuous curves.

[0043] It should be noted that in this embodiment, the forward effective driving area is not a fixed rectangle, nor is it obtained by directly expanding outward from the current lane centerline. Instead, it varies with vehicle speed, curvature, cross slope, and the continuity of continuous curves.

[0044] Furthermore, to describe the impact of the tightness of the connection between adjacent curves and the change in curvature on the continuity of risk, a continuity factor for consecutive curves is defined as follows:

[0045] in, This represents the continuity factor for continuous curves; This indicates the distance between the current curve and the next curve; This represents the spacing normalization constant; Indicates the coefficient for enhancing curvature variation; It represents the normalized quantity representing the absolute value of the rate of change of curvature. The larger the value, the closer the connection between adjacent curves and the more obvious the curvature change. The danger created by the vehicle at the end of the current curve is more likely to continue into subsequent curves.

[0046] Step S2. Hazardous Target Identification: The purpose of this step is to select, from multiple surrounding targets in a continuous curve scenario, truly dangerous targets that may conflict with the vehicle's future travel area. Unlike simply selecting the nearest target or the vehicle in front in the same lane, this step emphasizes the predicted occupancy relationship between the target and the vehicle's effective forward travel area, and further considers the target's geometric correlation, visual sensitivity, and the characteristics of the continuation of risk in continuous curves.

[0047] Furthermore, in this embodiment, the specific steps of step S2 are as follows: Step S2.1. Calculation of the overlap between the target and the vehicle's forward effective driving area: Let the effective forward driving area of ​​the vehicle obtained from step S1 be denoted as . For any target, the system generates its occupied area within the prediction time window based on the target's short-time trajectory prediction results. And calculate the degree of overlap between it and the vehicle's forward effective driving area:

[0048] in, Indicates the first The overlap between the predicted occupancy of the target and the vehicle's forward effective driving area; Indicates the predicted area occupied by the target; Indicates the vehicle's effective forward driving area; Indicates the area of ​​the region or the number of discrete grid cells; This represents a very small positive number that prevents the denominator from being zero. The larger the value, the more likely the target is to enter the vehicle's actual passage space in the future.

[0049] Furthermore, in this embodiment, the target's occupancy prediction region is obtained by the system using existing methods based on the target's current arc length coordinates, lateral offset, velocity, longitudinal acceleration, and relative heading angle, combined with a short-time trajectory prediction model, to obtain the future position and attitude sequence of the target at each discrete moment within the prediction time window; then, based on the target vehicle's length, width, and attitude information, an instantaneous occupancy envelope corresponding to each prediction moment is constructed, and the union of all instantaneous occupancy envelopes is calculated to form the target's occupancy prediction region within the prediction time window.

[0050] Step S2.2. Calculation of the geometric correlation between the target and the vehicle: Predicting overlap alone is insufficient to fully reflect the level of danger, because while some targets may occupy a portion of the effective forward driving area, their relative forward distance and lateral offset are significant, making the likelihood of a direct collision with the vehicle in the short term still relatively low. Therefore, this invention further considers the target's effective forward distance, lateral offset, and relative heading relationship along the reference line, defining the geometric correlation between the target and the vehicle as follows:

[0051] in, Indicates the first The geometric correlation between each target and the vehicle; Indicates the effective forward distance of the target along the reference line of the continuous curves; This represents the forward distance attenuation scale parameter; This indicates the lateral offset of the target relative to the center of the vehicle's travel. This represents the lateral offset attenuation scale parameter; Indicates the relative heading angle between the target and the vehicle; This represents the relative heading attenuation scale parameter.

[0052] In this embodiment, , and It is used to adjust the decay rate of the corresponding variable with respect to the target geometry. Its value can be determined based on the vehicle geometry, lane width, typical curvature range of continuous curves, and historical sample calibration results.

[0053] As shown in the above formula, when the target is closer to the vehicle along the reference line of continuous curves, has a smaller lateral offset, and a smaller relative heading difference, the geometric correlation of the target is higher. The geometric correlation is larger; however, as the relative heading difference between the target and the vehicle increases, the geometric correlation continuously decreases, but it will not be directly set to zero when the relative heading angle reaches a certain threshold. Compared with the expression form using truncated heading processing, the above smooth attenuation design can highlight the geometric correlation of targets moving in the same direction, while retaining the basic correlation information of lateral intrusion targets, strong oblique targets, and abnormal reverse targets in the danger screening stage, thereby improving the robustness of geometric correlation calculation in complex continuous curve scenarios.

[0054] Step S2.3. Calculation of the target's line-of-sight sensitivity factor: In scenarios involving continuous curves, some targets, while not having the highest geometric correlation, present greater challenges for the driver and system in continuous observation due to insufficient visibility, strong obstruction, or a significant elevation difference between the target and the vehicle. These targets should be classified as more dangerous. Therefore, the visibility sensitivity factor is defined as follows:

[0055] in, Indicates the first The line-of-sight sensitivity factor of each target; , and These represent the occlusion enhancement coefficient, the sight distance attenuation enhancement coefficient, and the elevation sensitivity coefficient, respectively. Indicates the occlusion intensity factor; Indicates the forward visible distance; This indicates a reference value for the visible distance; This indicates the relative elevation difference between the target and the vehicle. This represents the elevation difference normalization constant.

[0056] Step S2.4. Target Hazard Score and Key Hazard Target Identification: The first is defined by considering the predicted overlap of the target occupancy, geometric correlation, line-of-sight sensitivity factor, continuity of continuous curves, and prior target occupancy probability. The danger scores for each target are as follows:

[0057] in, Indicates the first The danger score for each objective; , and These represent the exponential weights of the predicted overlap, geometric correlation, and line-of-sight sensitivity factors, respectively. This represents the continuity enhancement coefficient for continuous curves; This represents the continuity factor for continuous curves; This represents the prior occupancy probability enhancement coefficient; This represents the prior probability that the target will occupy the vehicle's forward effective driving area within the prediction time window.

[0058] When the target danger score exceeds the threshold When this happens, the system will add the target to the set of dangerous targets. Furthermore, the target with the highest danger score is designated as the key danger target:

[0059] in, Indicates time Key hazard target number; This represents a set of dangerous targets.

[0060] In some scenarios with consecutive curves, the hazard scores of two targets may be very close, for example, when a slow vehicle in the same lane ahead and a lateral intrusion target at the curve exit coexist. To address this, this invention retains information on secondary hazard targets. When the ratio of the second hazard score to the maximum hazard score exceeds a threshold, the secondary hazard target and the primary hazard target are both included in the subsequent trend assessment, thus avoiding premature simplification of complex multi-target scenarios.

[0061] Step S3. Risk quantification based on neural network adaptive correction and multi-indicator fusion: The purpose of this step is to address the issues of distorted traditional collision time and the inability of a single time indicator to fully cover intrusive risks in continuous curve scenarios. It introduces a neural network for adaptive correction on top of the basic collision time, further calculates the predicted intrusion time and multi-dimensional expected collision time, and integrates the three into a unified multi-dimensional intrusion expected collision time indicator.

[0062] Furthermore, in this embodiment, step S3 specifically includes the following steps: Step S3.1. Calculation of basic collision time: For the first in the set of dangerous targets For each objective, the basic collision time is first calculated along the reference line of the continuous curves:

[0063] in, Indicates the first The base collision time for each target; Indicates the effective forward distance of the target along the reference line of the continuous curves; This represents the equivalent closing velocity of the target relative to the vehicle along the tangential direction of the reference line; This represents a minimal positive constant to prevent the denominator from being zero. The basic collision time is no longer directly calculated using the Cartesian straight-line distance, but rather based on the effective forward distance and relative closing velocity along the reference line of continuous curves.

[0064] Furthermore, the equivalent closing velocity Defined as:

[0065] in, Indicates the speed of this vehicle; Indicates the first The speed of the target; This indicates the heading angle of the vehicle relative to the tangential direction of the reference line of the continuous curves; This represents the heading angle of the target relative to the tangential direction of the reference line of the continuous curves. Therefore, and These represent the equivalent velocity components of the vehicle and the target in the tangential direction of the reference line on the continuous curves, respectively. The difference between the two is used to characterize the relative approach speed of the vehicle and the target in the reference line direction.

[0066] when When the value is greater than 0, it indicates that the vehicle and the target have a tendency to close the curve along the reference line direction. In this case, the basic collision time can be calculated using the above formula. When the value is ≤0, it indicates that there is no continuous approximation relationship along the reference line direction. In this case, the basic collision time takes a larger value, or it is considered that there is no collision risk caused by approximation along the reference line direction in the short term.

[0067] Step S3.2. Construction of joint feature input and multi-factor adaptive correction collision time: To preserve physical interpretability while absorbing the complex nonlinear relationships in continuous curve scenarios, this invention first constructs a joint scene-target feature vector:

[0068] in, Indicates the first The scene corresponding to each target—the joint feature vector of the targets; , , , , and These represent curvature, rate of change of curvature, longitudinal slope, transverse slope, visibility distance, and occlusion intensity, respectively. Indicates the overlap of target predictions; Indicates geometric correlation degree; Indicates the relative heading angle; Indicates the relative elevation difference; This represents the relative acceleration along the reference line.

[0069] Furthermore, the joint feature vector of the scene target at the current moment and several previous moments is combined. The input is a lightweight temporal neural network (such as GRU, LSTM, or Temporal Convolution Network (TCN)), which processes the data to obtain adaptive correction coefficients for dangerous targets. and its uncertainty The correction coefficients and their uncertainties are obtained.

[0070] in, Represents the adaptive correction coefficients of the neural network output; It represents the uncertainty or confidence inverse of the neural network output; This represents a lightweight temporal neural network. Indicates the control cycle; This indicates the number of steps in the time frame for review.

[0071] Based on this, a multi-factor adaptive collision time correction is constructed by integrating curvature, rate of change of curvature, line-of-sight attenuation, occlusion, relative angle, elevation difference, relative acceleration, and network uncertainty:

[0072] in, Indicates the first Multi-factor adaptive correction of collision time for each target; Used for learning-based correction of basic collision times; Represents the equivalent curvature normalization quantity; The normalized quantity representing the absolute value of the rate of change of curvature. Indicates the occlusion intensity factor; to This represents the weighting coefficient of the corresponding influencing factor; This represents the elevation difference normalization constant. The enhancement term representing the relative heading difference between the target and the vehicle. This indicates the relative elevation difference between the target and the vehicle. This represents the elevation difference normalization constant. This indicates the relative acceleration between the target and the vehicle along the reference line. This represents the normalized constant for relative acceleration. It's not a simple threshold-corrected collision time, but rather it directly incorporates the key distortion sources in continuous curve scenes into the collision time body itself, and through... Automatically tighten when evidence is insufficient.

[0073] Furthermore, to ensure that all risk variables can be incorporated into the collision time correction model under a unified dimension, the following normalization and auxiliary quantities are defined:

[0074] in, In the formula, Indicates the equivalent curvature within the current aiming interval; Indicates the rate of change of curvature; This represents the normalized reference value for curvature; This represents the normalized reference value for the rate of change of curvature. Indicates the current forward visible distance; This indicates a reference value for the forward visible distance; Indicates the occlusion intensity factor; This represents the number of discrete grid cells within the current forward effective driving area that are obscured by guardrails, mountains, slopes, large vehicles, or other obstructions. This represents the total number of discrete grid cells within the current forward effective driving area; This represents the relative elevation difference between the i-th target and the vehicle itself. Indicates the target elevation; Indicates the vehicle's elevation; This indicates the relative acceleration between the target and the vehicle along the tangential direction of the reference line of the continuous curves; and These represent the longitudinal accelerations of the vehicle and the target, respectively. and These represent the heading angles of the vehicle relative to the target in the tangential direction of the continuous curve reference lines; This represents a very small positive number that prevents the denominator from being zero.

[0075] Step S3.3. Calculation of intrusion time after conflict zone construction and prediction: To characterize the time-gap risk when the vehicle and the target successively enter the same conflict zone, the vehicle's time interval in the prediction time domain is recorded. The future occupied envelope within is , No. The future occupied envelope of each target is The conflict zone is then defined as:

[0076] in, Indicates the first The conflict area between the target and the vehicle in the prediction time domain; This indicates the time this vehicle is at The future occupied envelope; Indicates the target at time. The future occupied envelope; This indicates the prediction time window.

[0077] Assume this vehicle enters and leaves the conflict area. The times are respectively and Targets entering and leaving the conflict zone The times are respectively and The predicted intrusion time is defined as:

[0078] in, Indicates the first Predicted intrusion time for each target; and These indicate the times when the vehicle entered and left the conflict zone, respectively. and These represent the times when the target enters and leaves the conflict zone, respectively. When both the vehicle and the target will simultaneously occupy the conflict zone within the prediction time domain... The value is zero; when the vehicle and the target only occupy the same area sequentially but with a time gap, Greater than zero.

[0079] Step S3.4. Calculation of multi-dimensional expected collision time: In continuous curve scenarios, some risks do not manifest as simple same-direction approach or sequential intrusion, but rather as multi-dimensional approaches involving lateral safety, oblique intrusion, and elevation difference coupling. Therefore, the first... The target at the predicted time The multidimensional equivalent separation distance is:

[0080] in, Indicates the predicted time The multidimensional equivalent separation distance; express The difference in arc length between the target vehicle and the vehicle itself in the direction of the reference line at any given moment; express The relative deviation between the target vehicle and the vehicle in the lateral direction at any given time; express The relative difference in elevation between the target vehicle and the vehicle itself at any given time; Represents the relative heading angle at the predicted time; , and These represent the weighting coefficients for lateral, elevation, and relative heading angles, respectively. This represents the characteristic length constant, with a value ranging from 0.8m to 1m.

[0081] Furthermore, the most dangerous moment in the prediction time domain is defined as:

[0082] in, Indicates the first The moment when an objective minimizes the equivalent multidimensional separation distance within the prediction time domain.

[0083] Subsequently, the first is defined. The multi-dimensional expected collision time for each target is:

[0084] in, Indicates the first The expected collision time of each target in multiple dimensions; The rate of change of the equivalent multidimensional separation distance at the most dangerous moment; This represents a very small positive constant to prevent the denominator from being zero. The smaller this value, the faster the target and the vehicle converge to the multidimensional conflict state in the future time domain. Compared to the traditional two-dimensional expected collision time, It also considers longitudinal approach, lateral intrusion, heading convergence and elevation difference coupling in continuous curve scenarios, making it more suitable for describing complex intrusive risks in continuous curves.

[0085] Step S3.5. Fusion of conflict topology adaptive weights and multidimensional intrusion expected collision time: To avoid simply linearly adding the three time indicators and obscuring the dominant risk mechanism, this invention dynamically allocates weights for multi-factor adaptive correction of collision time, predicted post-intrusion time, and multi-dimensional expected collision time based on the conflict topology between the target and the vehicle. First, the simultaneous occupancy ratio is defined:

[0086] in, Indicates the first The proportion of the conflict area occupied by each target and the vehicle in the prediction time domain; This represents the characteristic function, which takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Three conflict topology features are then defined:

[0087] in, Indicates the intensity of the feature that approximates in the same direction; Indicates the strength of the lateral safety feature; This represents the intensity of temporal intrusion features. Further, the adaptive weights for the three types of time indicators are defined as follows:

[0088] in, , and These represent the weights of multi-factor adaptive correction collision time, predicted intrusion time, and multi-dimensional expected collision time, respectively. to Indicates the weight mapping coefficients; This represents the normalized mapping function, which makes the sum of the three weights equal to 1; This represents the continuity factor for continuous curves. Therefore, when the target is closer to a continuous approximation relationship in the same direction, It will increase; when targets increasingly intrude into the same conflict zone one after another, It will increase; when the target exhibits more lateral security or oblique intrusion, It will increase.

[0089] After obtaining the weights, further define the first... The expected collision time for the multidimensional intrusion of the target is:

[0090] in, Indicates the first Expected collision time for multi-dimensional intrusion of multiple targets; This indicates a multi-factor adaptive correction of the collision time; Indicates the predicted intrusion time; Indicates the expected collision time in multiple dimensions; This represents extremely small normal numbers. The harmonic fusion method, employing the reciprocal weighting and then taking the reciprocal of the entire value, is used instead of a simple linear weighted average. This is to prevent a highly precarious dimension from being masked by two larger time values. Therefore, a significant decrease in any sub-indicator will drive... Rapid reduction, thereby increasing the sensitivity of the indicator to critical dangerous states.

[0091] Step S4. Intervention Timing and Intervention Level Determination: The purpose of this step is to determine whether the system should intervene and at what level, based on constraints of the continuous curve scenario, expected collision time of multi-dimensional intrusion, comprehensive risk, information completeness, and vehicle control capabilities. This step first determines the maximum level of execution allowed by the current scenario and vehicle conditions, then determines the level that the system needs to request from a risk perspective, and finally takes the limited result of the two as the final execution level.

[0092] Furthermore, in this embodiment, step S4 specifically includes the following steps: Step S4.1. Calculation of environmental risk, information completeness, and control capability: To support the determination of intervention level, the environmental risk of continuous curves is first calculated, as expressed by:

[0093] in, This indicates the environmental risk associated with continuous curves; Indicates the first Normalized risk terms for adverse environmental factors; This indicates the corresponding weight. In this embodiment, the four types of adverse environmental factors can be respectively identified as curvature risk, longitudinal slope risk, line-of-sight risk, and low adhesion risk. Curvature risk, i.e., the current curvature of the curve. Absolute value normalization, This represents the maximum curvature of the curve. For longitudinal slope risk, i.e., the current longitudinal slope normalization, This represents the maximum longitudinal slope. Risk based on line of sight, i.e., forward visible distance. The greater the deficiency, the higher the risk. This is a reference value for forward visibility distance. Low adhesion risk, meaning a high risk when the road surface friction coefficient is low. This is a reference value for the road surface friction coefficient.

[0094] Furthermore, to describe the completeness of perceived and predicted information, information completeness is defined as:

[0095] in, Indicates the overall completeness of information; Indicates the completeness of lane boundary recognition; Indicates target tracking stability; Indicates the stability of the target trajectory prediction; , and This indicates the corresponding weight.

[0096] have:

[0097] in, The number of lane boundary points identified. This represents the total expected number of points at the lane boundary. To successfully track the target frame number, Total target frames The number of frames in which the trajectory prediction error is less than a threshold. Total number of predicted frames.

[0098] Meanwhile, to describe whether the vehicle currently possesses the ability to perform higher-level interventions, the control capability margin is defined as:

[0099] in, Indicates the vehicle's control capability margin; This indicates the currently available longitudinal deceleration, which is the maximum longitudinal deceleration that the vehicle can currently achieve. It can be calculated from vehicle dynamics and inverse dynamics parameters. This represents the longitudinal deceleration required to avoid the current conflict, i.e., the deceleration required for a longitudinal conflict. ; To represent the relative closing velocity between the vehicle and a critical hazardous target along the tangential direction of the continuous curve reference line, it can also be understood as the longitudinal approach velocity between the two in the direction of the curve reference line. To demonstrate the safe distance between the vehicle and a critical hazardous target that can be used for braking and hazard avoidance, the effective longitudinal distance between the front end of the vehicle and the rear end of the target, after deducting necessary safety margins, is shown along the reference line direction of a continuous curve. This represents the available lateral acceleration corresponding to the current available lateral stability margin, used to determine whether the vehicle can safely complete a lateral maneuver; ;in, The coefficient of friction of the road surface. For the current longitudinal acceleration, It is the acceleration due to gravity; This represents the lateral acceleration required to avoid the current conflict. It is the lateral acceleration the vehicle needs to complete the avoidance maneuver under the current conflict conditions. ;in, For the expected lateral displacement, This refers to the driver's reaction time. Indicates available free space. ;in, Distance to the obstacle ahead. This refers to the length of the vehicle. This indicates the space required to execute the current evasive action. ; Represents extremely small positive numbers. This is the speed of the vehicle.

[0100] thus, , and The assessment will provide a basis for determining the upper limit of the rating based on three aspects: the severity of the environment, the reliability of perception and prediction, and the feasibility of vehicle actions.

[0101] Step S4.2. Risk Mapping and Tightening Trend Calculation: After obtaining the expected collision time of multi-dimensional intrusion, further considering the continuity of consecutive curves, information completeness, and the rapid approach trend of danger, a conservative enhancement term is defined:

[0102] in, Indicates the first Conservative enhancements for each objective; , and Weighting coefficients representing continuity of consecutive curves, lack of information completeness, and tightening trend; Indicates that for the first The completeness of perception and prediction information for each target; This represents the derivative of the expected collision time with respect to time in a multidimensional intrusion.

[0103] Based on this, Mapped to target risk quantity:

[0104] in, Indicates the first The risk quantification value of each objective; This represents the reference time constant. The closer a value is to 1, the higher the target risk. To reflect the overall interactive risk and its tightening trend, it is further defined as follows:

[0105]

[0106] in, Indicates the first The trend of tightening expected collision time for multi-dimensional intrusion of individual targets; This represents the overall interactive risk of a set of dangerous targets; This represents the prior risk weights obtained by mapping the target risk score; This represents the weighting coefficient for the tightening trend.

[0107] Step S4.3. Determine the maximum allowed level: In scenarios with continuous curves, even with high risk, the system may not be allowed to execute actions of arbitrary strength and form. Therefore, the upper limit of the permissible level is first determined based on scenario constraints, control capabilities, and information completeness. The scenario constraint index is defined as:

[0108] in, Indicates the scenario constraint index; , and These represent the weighting coefficients for curvature, continuity of continuous curves, and sight distance attenuation, respectively.

[0109] set up , and These represent mapping the scenario constraint index, control margin, and information completeness to a set of levels, respectively. The quantization function is then used to determine the highest level of intervention allowed by the system at the current moment:

[0110] in, This indicates the upper limit of the permissible level. This applies when there is high curvature, strong continuity of consecutive curves, and insufficient visibility. It will raise or lower the upper limit of the permissible level; when vehicle control capabilities are insufficient or information completeness is low, or The upper limit of allowed levels will also be tightened accordingly.

[0111] Step S4.4. Determine the overall risk level and actual request level: After obtaining the maximum allowed level, it is also necessary to assess the current request level from a risk perspective. Therefore, the comprehensive risk is defined as follows:

[0112] in, Indicates comprehensive risk; Indicates overall interaction risk; This indicates the environmental risk associated with continuous curves; Indicates the completeness of information; This indicates a risk of driver mismatch during takeover; , , and These represent the weighting coefficients.

[0113] Furthermore, in this embodiment, the risk of driver takeover mismatch can be further described as:

[0114] in, Represents a saturation function; This indicates the time required to complete the takeover under the current risk conditions; This indicates the effective takeover time currently available to the driver; This indicates a driver mismatch indicator.

[0115] Let the minimum expected collision time of multidimensional intrusion in the set of dangerous targets be . The corresponding maximum tightening trend is Then we have:

[0116] Further construct an intervention request index:

[0117] in, Indicates the intervention request index; Indicates the reference time scale corresponding to the requested index; , and These represent the weight coefficients of the minimum multidimensional intrusion expected collision time term, the comprehensive risk term, and the tightening trend term, respectively.

[0118] Reset Let the quantization function that maps the intervention request index to the actual request level be:

[0119] in, This indicates the actual request level. At this point, the system has obtained the "maximum allowed level" and the "actual request level" respectively.

[0120] Step S4.5. Final Execution Level and Action Mode Selection: By combining the maximum allowed level and the actual requested level, the final execution level is obtained:

[0121] in, Indicates the final execution level. If... Higher than In this case, the system will not blindly output actions that exceed the scene constraints and vehicle capabilities, but will limit the execution level to the upper limit of the allowed range.

[0122] Furthermore, within the same execution level, the system does not rely on a single fixed action, but rather selects a mode more suitable for the current continuous curve scenario from the set of action modes corresponding to that level. Let... Indicates level Given the set of allowed action patterns, the optimal action pattern can be represented as:

[0123] in, Indicates the optimal action pattern; Indicates action mode The comprehensive cost function. To balance security and smoothness, it can be further defined as follows:

[0124] in, Indicates action mode The resulting longitudinal impact force is expressed as follows: ;in, Let m be the longitudinal acceleration of action pattern m at time t. As the reference longitudinal acceleration, To control the cycle; and ;in, The ideal acceleration is calculated based on the road curvature and the vehicle's desired speed: , To determine the desired speed, it can be calculated from the speed limit, the speed of the vehicle in front, or the safe speed. The current road curvature radius, For longitudinal velocity weighting parameters, This represents the vehicle's current speed.

[0125] Indicates action mode The degree of amplification of lateral deviation is expressed as ; Let m be the horizontal position of action pattern m at time t. For reference, the lateral position (lane center line) is used. Indicates action mode The cost of driving comfort and rule consistency is expressed as:

[0126] in, , For longitudinal and lateral acceleration, For the front wheel steering angle, , , Corresponding weights are used to balance comfort and rule consistency.

[0127] , and These represent the corresponding weights. Through the above processing, in situations with high curvature and low line-of-sight, the system of this invention can preferentially select the deceleration and maintenance mode rather than the aggressive lateral avoidance mode.

[0128] Furthermore, in this embodiment, the action modes include (1) deceleration and holding mode: longitudinal deceleration, maintaining lateral position, and not triggering lateral avoidance; (2) enhanced lane keeping mode: slight lateral correction and longitudinal speed maintenance; (3) enhanced control mode: longitudinal deceleration and enhanced lateral avoidance gain; (4) emergency control mode: maximum longitudinal deceleration and lateral emergency avoidance.

[0129] Step S5. Tiered intervention control output: The purpose of this step is to output control results adapted to the continuous curve scenario based on the final execution level and optimal action mode. Different levels of action not only differ in intensity, but are also subject to geometric constraints, control capability constraints, and information completeness constraints in terms of action form.

[0130] Furthermore, in this embodiment, step S5 includes the following steps: Step S5.1. Longitudinal control, lateral control, and lateral envelope tightening: For the final execution level and optimal action mode The system first generates longitudinal control commands and lateral control commands:

[0131]

[0132] in, This indicates the longitudinal acceleration command output by the system. The longitudinal intervention gain associated with execution level and action pattern is expressed explicitly as: ;in, Based on the longitudinal gain constant, , These are the weighting coefficients for level and action pattern; This represents the basic longitudinal control reference value, which has ;in, To maintain a safe following distance; This represents the longitudinal control adjustment amount obtained by mapping the comprehensive risk and the minimum multidimensional intrusion expected collision time, which is calculated by using a linear or nonlinear mapping function: ;in, These are the vertical mapping coefficients.

[0133] This indicates the system outputting a lateral control or front wheel steering angle correction command. The basic lateral control reference value is explicitly expressed as... ;in, This refers to the vehicle's wheelbase. For road curvature, This is the horizontal correction factor; The lateral correction gain related to execution level and action pattern is expressed explicitly as: ;in, Based on the fundamental transverse gain constant, , These are the weighting coefficients for level and action pattern; The lateral correction amount, derived from the mapping of comprehensive risk and minimum multidimensional intrusion expected collision time, is: ;in, This represents the horizontal mapping coefficient.

[0134] To avoid overly aggressive lateral control in high-risk scenarios involving continuous curves, this invention also applies a level-related tightening to the available lateral envelope. Let the tightened available lateral width be... Then we have:

[0135] in, Indicates the available horizontal envelope width at the current level; This represents the average effective width of the forward effective driving area; Indicates the tightening factor of the grade; This indicates the final execution level. The above processing means that as the execution level increases, the system's lateral movement range in consecutive curves will be moderately tightened, thus prioritizing deceleration and stable tracking rather than encouraging large lateral evasive maneuvers.

[0136] Step S5.2. Output of human-computer interaction prompts and explanations: While outputting control commands, the system also needs to generate human-computer interaction prompts that match the current execution level and action mode. The prompt output will be denoted as:

[0137] in, This indicates the output of human-computer interaction prompts; Indicates the mapping function; Indicates the final execution level; Indicates the optimal action pattern; This indicates the number of the critical hazard target. For example, in a Level 1 intervention, Visual cues or mild audible alerts can be provided. Level 2 and 3 interventions can be further enhanced with steering wheel vibration, brake pre-charge, or stronger warnings. Level 4 interventions can be overlaid with explicit emergency control status warnings. By synchronizing the warning information with the level and mode, drivers can more clearly understand why the system is taking the current action in a series of curves.

[0138] Furthermore, to facilitate system recording, calibration, and fault tracing, this invention also constructs an explanation output vector:

[0139] in, This represents the interpretation of the output vector; Indicates the critical hazard target number; Represents the minimum multidimensional intrusion expected collision time in the set of dangerous targets; Indicates overall interaction risk; Indicates comprehensive risk; Indicates the final execution level; This indicates the optimal action mode. The output vector can be sent to the event logger, diagnostic module, or upper-level policy management module.

[0140] Step S5.3. Rolling update and control closed loop implementation: After each control cycle, the system re-enters step S1 to update the continuous curve scenario state, road state, vehicle state, target state, and dangerous target set, and repeats the calculation process from steps S2 to S5 in sequence. Because this invention introduces a continuous curve scenario confidence level in S1, a forward effective driving area and a continuous curve continuity factor in S2, multi-factor adaptive correction of collision time, predicted intrusion time, multi-dimensional expected collision time, and multi-dimensional intrusion expected collision time in S3, and a dual-layer judgment mechanism of the allowable level upper limit and the actual requested level in S4, the entire system can balance scenario continuity, risk continuity, and action continuity during rolling execution. When the continuous curves gradually end, visibility recovers, the hazard score of key dangerous targets decreases, the expected collision time of minimum multidimensional intrusion increases, and the overall risk decreases, the system will automatically reduce the request level and gradually release the tightening of the lateral envelope. Conversely, when the continuous curves are more closely connected, visibility continues to decrease, key dangerous targets continue to approach, and the expected collision time of minimum multidimensional intrusion tightens rapidly, the system will gradually increase the execution level within the upper limit of the allowable level and prioritize outputting deceleration and stabilization control actions that match the current curve constraints.

[0141] The present invention also provides an electronic device, comprising: one or more processors and a memory; wherein the memory is used to store one or more programs or instructions, and when the one or more programs or instructions are executed by the one or more processors, the one or more processors implement the vehicle control method for continuous curve scenarios described above.

[0142] The present invention also provides a vehicle including the aforementioned electronic device, wherein the memory of the electronic device stores a program or instructions that can run on a processor, and when the program or instructions are executed by the processor, the aforementioned vehicle control method for continuous curve scenarios is implemented.

[0143] The present invention also provides a computer-readable medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the vehicle control method for continuous curve scenarios described above.

[0144] Those skilled in the art will understand that all or part of the functions of the various methods / modules in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved.

[0145] In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the programs can also be stored in storage media such as servers, other computers, disks, optical discs, flash drives, or portable hard drives. They can be downloaded or copied to the memory of the local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be implemented.

[0146] The above describes specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A vehicle control method for continuous curve scenarios, characterized in that, The method includes the following steps: Step S1. Collect data and identify whether the vehicle is currently in a continuous curve scenario. If so, construct the road state, vehicle state, and target state in the continuous curve reference line coordinate system, and determine the effective forward driving area of ​​the vehicle. Step S2. Combining the predicted occupancy relationship between the target and the vehicle's forward effective driving area, the geometric correlation between the target and the vehicle, the target's line-of-sight sensitivity, and the continuity of continuous curves, screen surrounding targets to identify key hazardous targets. Step S3. For the identified key hazardous targets, a multi-factor adaptive correction collision time is constructed by integrating relative angle, elevation difference, relative acceleration, curvature-line-of-sight coupling relationship, and neural network uncertainty on the basic collision time model. At the same time, the predicted intrusion time is calculated based on the entry and exit times of the conflict area occupied by the vehicle and the target in the future envelope, and the multi-dimensional expected collision time is obtained based on the multi-dimensional equivalent separation distance under the reference line of continuous curves. The three are then integrated into a multi-dimensional intrusion expected collision time. Step S4. Combining environmental risk, information completeness, vehicle control capability, multi-dimensional intrusion expected collision time, and driver takeover mismatch risk, a two-layer judgment mechanism combining the upper limit of the allowable level and the actual request level is adopted to determine the final execution level and output the optimal action mode adapted to the continuous curve scenario. Step S5. Based on the final execution level and optimal action mode, output control results adapted to the continuous curve scenario, including longitudinal control, lateral control and lateral envelope tightening.

2. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, In step S1, when determining whether the vehicle is currently in a continuous curve scenario, the continuous curve recognition score is first calculated based on the equivalent curvature, curvature change rate, distance between the current curve and the next curve, and current forward visible distance within the current aiming interval. The continuous curve recognition score is then mapped to the continuous curve scenario confidence level. ; Then, an entry threshold is used. With exit threshold Separation hysteresis strategy, when When entering a continuous curve scene, when Furthermore, it exits the continuous curve scenario after several control cycles.

3. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, In step S1, the road status includes the equivalent curvature, rate of change of curvature, road longitudinal slope, forward visibility distance, occlusion intensity factor, and estimated road surface adhesion coefficient within the pre-aiming interval; the vehicle status includes the arc length coordinates and lateral offset of the vehicle in the continuous curve reference line coordinate system, the vehicle speed, the vehicle longitudinal acceleration, the vehicle's heading angle relative to the reference line, and the vehicle's yaw rate; the target status includes the arc length coordinates and lateral offset of the target in the continuous curve reference line coordinate system, the target speed, the target longitudinal acceleration, the target's heading angle relative to the reference line, and the target's elevation difference relative to the vehicle. In a continuous curve scenario, the effective forward driving area of ​​the vehicle is constructed based on the vehicle speed, curvature, cross slope, and the continuity of the curves. The effective lateral width of this area is: ； in, Indicates the position of arc length The effective forward driving width at the location; Indicates the current lane width; Indicates the safety envelope boundary of this vehicle; Indicates the curvature influence coefficient; Indicates the position of arc length Local curvature at that point; Indicates the cross slope influence coefficient; Indicates the position of arc length Superelevation or transverse slope at the location; The aiming length is: ； in, Indicates the pre-aiming length of the effective forward driving area; Indicates the basic aiming length; Indicates the speed adjustment coefficient; Indicates the speed of this vehicle; Indicates the enhancement coefficient for continuous curves; This represents the continuity factor for consecutive curves, determined based on the rate of curvature change and the distance between the current curve and the next curve.

4. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, In step S2, after obtaining the vehicle's forward effective driving area, for any target, the predicted occupancy area within the prediction time window is generated based on the target's short-time trajectory prediction results, and the degree of overlap between the predicted occupancy area and the vehicle's forward effective driving area is calculated to characterize the predicted occupancy relationship between the target and the vehicle's forward effective driving area; at the same time, the effective forward distance, lateral offset, and relative heading of the target along the continuous curve reference line direction are comprehensively considered to determine the geometric correlation between the target and the vehicle. The target's line-of-sight sensitivity is determined by comprehensively considering the occlusion intensity factor, forward visibility distance, and the relative elevation difference between the target and the vehicle. Finally, the target's hazard score is calculated. When the target's hazard score exceeds the threshold, it is included in the set of hazardous targets, and the target with the highest hazard score is designated as the critical hazardous target.

5. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, The basic collision time model in step S3 is as follows: ； in, Indicates the first The base collision time for each target; Indicates the effective forward distance of the target along the reference line of the continuous curves; This represents the equivalent closed velocity of the target relative to the vehicle along the reference line in the tangential direction. ; This represents a positive integer that prevents the denominator from being zero. Indicates the speed of this vehicle; Indicates the first The speed of the target; This indicates the heading angle of the vehicle relative to the tangential direction of the reference line of the continuous curves; The heading angle represents the target's heading relative to the tangential direction of the reference line of the continuous curves; The neural network is a lightweight temporal neural network, which is one of GRU, LSTM or TCN. By inputting the scene-target joint feature vectors of the current time and the historical time into the neural network, the adaptive correction coefficients and uncertainties of the dangerous target are obtained after network processing. The multi-factor adaptive correction collision time is: ； in, Indicates the first Multi-factor adaptive correction of collision time for each target; This represents the adaptive correction coefficients of the neural network output, used to learn and correct the basic collision time. Represents the equivalent curvature normalization quantity; Normalized quantity representing the absolute value of the rate of change of curvature; Indicates the current forward visible distance; This indicates a reference value for the forward visible distance; Indicates the occlusion intensity factor; The enhancement term representing the relative heading difference between the target and the vehicle. Indicates the relative heading angle. This indicates the relative elevation difference between the target and the vehicle. This represents the elevation difference normalization constant. This indicates the relative acceleration between the target and the vehicle along the reference line. This represents the normalized constant of relative acceleration; This indicates the uncertainty in the output of the neural network; to This represents the weighting coefficient of the corresponding influencing factor.

6. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, The multidimensional equivalent separation distance in step S3 is: ; in, Indicates the first The target at the predicted time The multidimensional equivalent separation distance; express The difference in arc length between the target vehicle and the vehicle itself in the direction of the reference line at any given moment; express The relative deviation between the target vehicle and the vehicle in the lateral direction at any given time; express The relative difference in elevation between the target vehicle and the vehicle itself at any given time; Represents the relative heading angle at the predicted time; , and These represent the weighting coefficients for lateral, elevation, and relative heading angles, respectively. This represents the characteristic length constant, with values ​​ranging from 0.8m to 1m. The most dangerous moment for each target in the prediction time domain is determined based on the multidimensional equivalent separation distance, and the multidimensional expected collision time is calculated based on the equivalent multidimensional separation distance and its rate of change at the most dangerous moment. Finally, based on the conflict topology relationship between the target and the vehicle, the weights of multi-factor adaptive correction of collision time, predicted post-intrusion time and multi-dimensional expected collision time are dynamically allocated; among which, the conflict topology relationship includes same-direction approximation, lateral safety and temporal intrusion. After obtaining the weights, the expected collision time for multidimensional intrusion is calculated as follows: ； in, Indicates the first Expected collision time for multi-dimensional intrusion of multiple targets; This indicates a multi-factor adaptive correction of the collision time; Indicates the predicted intrusion time; Indicates the expected collision time in multiple dimensions; This represents a positive integer that prevents the denominator from being zero. , and These represent the weights of the multi-factor adaptive correction collision time, the predicted intrusion time, and the multi-dimensional expected collision time, respectively.

7. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, In step S4, environmental risk calculation is performed by comprehensively considering curvature risk, longitudinal slope risk, line-of-sight risk, and low adhesion risk; information completeness calculation is performed by comprehensively considering lane boundary recognition completeness, target tracking stability, and target trajectory prediction stability; and vehicle control capability calculation is performed by comprehensively considering the currently available longitudinal deceleration. Longitudinal deceleration required to avoid current conflict The available lateral acceleration corresponding to the current available lateral stability margin Lateral acceleration required for current conflict avoidance The available free space and the space required to execute the current evasive maneuver can be used to determine this; Step S4 will determine the multidimensional intrusion expected collision time. Mapped to the target risk quantity, and the first... The tightening trend of multidimensional intrusion expected collision time for each target and the overall interactive risk of the dangerous target set; whereby the target risk quantity is expressed as: ； In the formula, Indicates the first The risk quantification value of each objective; Indicates the reference time constant; Indicates the first The conservative enhancement terms for each objective are determined based on the expected collision time of multi-dimensional intrusion, the continuity of continuous curves, and the completeness of information. This represents a positive integer that prevents the denominator from being zero. Then, the upper limit of the allowable level is determined based on scenario constraints, vehicle control capabilities, and information completeness. The comprehensive risk is calculated by considering environmental risk, overall interaction risk, information completeness, and driver takeover mismatch risk. The current actual request level is determined based on the comprehensive risk, the minimum expected collision time of multi-dimensional intrusion, and the tightening trend of the maximum expected collision time of multi-dimensional intrusion.

8. The vehicle control method for continuous curve scenarios according to claim 1, characterized in that, In step S4, multiple action modes are set within the same execution level, and the action mode with the smallest comprehensive cost function is selected as the optimal action mode; the comprehensive cost function is composed of longitudinal impact, lateral deviation, driving comfort and rule consistency; in step S5, while outputting control commands, a human-machine interaction prompt matching the current execution level and action mode is generated.

9. An electronic device, comprising: One or more processors and a memory; wherein the memory is used to store one or more programs or instructions, characterized in that, when the one or more programs or instructions are executed by the one or more processors, the one or more processors implement the vehicle control method for continuous curve scenarios as described in any one of claims 1 to 8.

10. A vehicle comprising electronic equipment, wherein the memory of the electronic equipment stores programs or instructions capable of running on a processor, characterized in that, When the program or instructions are executed by the processor, they implement the vehicle control method for continuous curve scenarios as described in any one of claims 1 to 8.

Images