A vehicle control method, device and equipment based on automatic driving trajectory verification and vehicle

Abstract

The application discloses a vehicle control method and device based on automatic driving track verification, equipment and vehicle, belongs to vehicle driving control technical field. The method comprises the following steps: firstly, performing safety verification on the to-be-verified track by collision time constraint verification, marking the track which does not pass the collision time constraint verification as unusable track and terminating the subsequent verification process, only for the track which passes the collision time constraint verification, then carrying out compliance verification of lane boundary constraint verification and traffic light constraint verification in turn, if a track meeting the safety and compliance requirements can be screened out, the track is taken as the driving track of the vehicle, otherwise, a retreat signal which can avoid danger is generated to make the vehicle perform emergency safety operation. Through the priority progressive verification mode of the collision time constraint verification, the application realizes the execution logic that the automatic driving life safety priority is higher than the traffic rule compliance and the missing retreat is unacceptable, and can meet the safety guarantee demand of the end-to-end automatic driving vehicle deployment.

Description

Technical Field

[0001] This application relates to the field of vehicle driving control technology, and in particular to a vehicle control method, device, equipment and vehicle based on autonomous driving trajectory verification. Background Technology

[0002] In the field of end-to-end autonomous vehicle control technology, in order to ensure that the vehicle drives safely and stably according to the trajectory predicted by the autonomous driving strategy, it is necessary to conduct multi-dimensional constraint verification on multiple trajectories predicted by the autonomous driving strategy, such as collision risk, lane driving boundary, traffic signal rules, etc., and select driving trajectories that meet safety and compliance requirements to avoid driving risks such as vehicle collision, lane crossing and running red lights.

[0003] Traditional technologies typically employ a randomized, sequential approach when performing multi-constraint verification on autonomous driving trajectories. This involves executing verification rules for different risk levels, such as collision time constraints, lane boundary constraints, and traffic light constraints, simultaneously or in a disordered manner. Because collision time constraint verification, which characterizes the risk of fatal collisions, is not included as a pre-verification step, traditional technologies continue to perform compliance checks on trajectories already exhibiting high-risk collisions, including lane boundary and traffic light checks. This not only generates a large amount of meaningless redundant computation and significantly reduces the efficiency of trajectory selection and driving decisions, but also fails to prioritize and eliminate trajectories with fatal collision risks. This can easily lead to rollbacks that focus on compliance checks while overlooking fatal collision risks, violating the execution logic that prioritizes life safety over traffic rule compliance and that rollbacks are unacceptable in autonomous driving. This results in serious driving safety vulnerabilities and fails to meet the safety assurance requirements of end-to-end autonomous driving vehicle deployments. Summary of the Invention

[0004] The main purpose of this application is to propose a vehicle control method, device, equipment and vehicle based on autonomous driving trajectory verification. By prioritizing the collision time constraint verification, a trajectory screening mechanism that fits the safety priority of autonomous driving is constructed. This realizes the execution logic that prioritizes the life safety of autonomous driving over traffic rule compliance and that omissions and backtracking are unacceptable, which can meet the safety assurance requirements of end-to-end autonomous driving vehicle deployment.

[0005] To achieve the above objectives, one aspect of this application proposes a vehicle control method based on autonomous driving trajectory verification, the method comprising: When the vehicle is in autonomous driving mode, acquire several trajectories to be verified predicted by the autonomous driving strategy within the current time window; For each of the aforementioned trajectories to be verified, a collision time constraint verification is performed on the trajectory. If it is determined that the trajectory fails the collision time constraint verification, the trajectory is marked as an unusable trajectory. If it is determined that the trajectory passes the collision time constraint verification, it is further determined whether the trajectory simultaneously passes the lane boundary constraint verification and the traffic light constraint verification. If so, the trajectory is marked as a verified trajectory; otherwise, the trajectory is marked as an unusable trajectory. If multiple verified trajectories exist, a target trajectory is selected from each of the verified trajectories, and the vehicle is controlled to drive according to the target trajectory; if no verified trajectory exists, a reversal signal is generated to indicate that the vehicle needs to perform a hazard avoidance operation. The collision time constraint check is used to verify whether the vehicle has a collision risk; the lane boundary constraint check is used to verify whether the vehicle has a risk of going out of the lane boundary; and the traffic light constraint check is used to verify whether the vehicle has a risk of crossing the stop line when the traffic light is red.

[0006] Furthermore, in some embodiments, generating a backoff signal to indicate that the vehicle needs to perform a hazard avoidance maneuver if the verified trajectory does not exist includes: If the verified trajectory does not exist, determine whether there is an unusable trajectory that failed the collision time constraint verification; If so, a first retreat signal is generated; wherein the first retreat signal is used to indicate that the vehicle needs to perform a minimum-risk maneuver to avoid danger; If not, a second backoff signal is generated; wherein the second backoff signal is used to indicate that the vehicle needs to perform a trajectory correction avoidance operation.

[0007] Furthermore, in some embodiments, after generating the first backoff signal, the method further includes: When it is determined that the first rollback signal has been generated in multiple consecutive time windows, the vehicle is controlled to perform a minimum-risk maneuver to decelerate to a stop.

[0008] Furthermore, in some embodiments, after generating the second backoff signal, the method further includes: When it is determined that the second rollback signal has been generated in multiple consecutive time windows, the vehicle speed is reduced and the vehicle is kept traveling in the current lane to complete the trajectory correction operation.

[0009] Furthermore, in some embodiments, the collision time constraint verification of the trajectory to be verified includes: Based on the vehicle's internal operating data and external road condition data, the collision time of the vehicle with other vehicles within a preset time window is calculated when the vehicle travels along the trajectory to be verified. When it is determined that the collision time is less than the first preset collision time threshold, the verification result that the trajectory to be verified has failed the collision time constraint verification is output. When it is determined that the collision time is not less than the first preset collision time threshold, the verification result of the trajectory to be verified passing the collision time constraint is output.

[0010] Further, in some embodiments, determining whether the trajectory to be verified has passed both lane boundary constraint verification and traffic light constraint verification simultaneously, and if so, marking the trajectory to be verified as a verified trajectory, and if not, marking the trajectory to be verified as an unusable trajectory, includes: Based on each trajectory point in the trajectory to be verified and the road condition data, lane boundary constraint verification is performed on the trajectory to be verified. Based on each trajectory point in the trajectory to be verified, the running data, and the traffic data, the trajectory to be verified is subjected to traffic light constraint verification. If the trajectory to be verified passes both the lane boundary constraint verification and the traffic light constraint verification, the trajectory to be verified is marked as the verified trajectory. If the trajectory to be verified fails the lane boundary constraint verification or the traffic light constraint verification, the trajectory to be verified will be marked as the unusable trajectory.

[0011] Furthermore, in some embodiments, the step of performing lane boundary constraint verification on the trajectory to be verified based on each trajectory point in the trajectory to be verified and the road condition data includes: Extract the left lane boundary and right lane boundary from the road condition data; Calculate the lateral distance from each trajectory point to the left lane boundary and the right lane boundary, and select the lateral distance with the smallest value as the minimum boundary margin of the trajectory to be verified. When it is determined that the minimum boundary margin is less than the preset boundary margin threshold, the verification result that the trajectory to be verified has failed the lane boundary constraint verification is output. When it is determined that the minimum boundary margin is not less than the preset boundary margin threshold, the verification result of the trajectory to be verified passing the lane boundary constraint verification is output.

[0012] Furthermore, in some embodiments, the step of performing traffic light constraint verification on the trajectory to be verified based on each trajectory point in the trajectory to be verified, the running data, and the traffic data includes: Extract the stop line boundaries and traffic light status information from the road condition data; Based on the running data and each of the trajectory points, the running speed corresponding to each trajectory point is extracted; When it is determined that the red light is in a valid state based on the traffic signal light status information, the valid red light period corresponding to the red light status is obtained; Extract each trajectory point to be verified within the red light's effective time period, and the corresponding running speed to be verified for each trajectory point to be verified; Calculate the longitudinal distance from each of the trajectory points to be verified to the boundary of the stop line; Determine whether there are any trajectory points to be verified whose longitudinal distance is less than a preset parking distance threshold; If so, output the verification result that the trajectory to be verified has failed the traffic light constraint verification; If not, calculate the stop line arrival time corresponding to each of the longitudinal distances and the running speeds to be verified; if it is determined that there is a stop line arrival time less than a preset parking time threshold, output the verification result that the trajectory to be verified has failed the traffic light constraint verification; if it is determined that there is no stop line arrival time less than the preset parking time threshold, output the verification result that the trajectory to be verified has passed the traffic light constraint verification. The stop line arrival time is used to characterize the time when the front reference position of the vehicle reaches the boundary of the stop line.

[0013] Furthermore, in some embodiments, selecting a target trajectory from the various verified trajectories includes: For each of the verified trajectories, the collision violation component corresponding to the collision time, the lane boundary violation component corresponding to the minimum boundary margin, and the traffic light violation component corresponding to the longitudinal distance are combined to obtain the target violation vector corresponding to the verified trajectory. For each verified trajectory, a consistency score is calculated by weighting and summing the target violation vector, the first weight corresponding to the collision time constraint verification, the second weight corresponding to the lane boundary constraint verification, and the third weight corresponding to the traffic light constraint verification; wherein, the first weight is greater than the second weight, and the second weight is greater than the third weight. The verified trajectory with the highest consistency score is taken as the target trajectory; wherein the consistency score is used to indicate the degree of compliance with collision time constraints, lane boundary constraints and traffic light constraints.

[0014] Furthermore, in some embodiments, the generation of the target violation vector includes: The collision violation component is constructed based on the collision time, the first preset collision time threshold, and the collision time confidence level; wherein, the collision time confidence level is used to indicate the reliability of the collision time calculation result; The lane boundary violation component is constructed based on the minimum boundary margin, the preset boundary margin threshold, and the lane boundary confidence; wherein, the lane boundary confidence is used to indicate the reliability of the lateral distance calculation results. The traffic light violation component is constructed based on the stop line arrival time, the preset parking time threshold, the red light valid period, and the parking distance confidence level; wherein, the parking distance confidence level is used to indicate the reliability of the longitudinal distance calculation result; The collision violation component, the lane boundary violation component, and the traffic light violation component are combined to output the target violation vector.

[0015] To achieve the above objectives, another aspect of this application proposes a vehicle control device based on autonomous driving trajectory verification, the device comprising: The trajectory acquisition module is used to acquire several trajectories to be verified predicted by the autonomous driving strategy within the current time window when the vehicle is in autonomous driving mode. The trajectory verification module is used to perform collision time constraint verification on each trajectory to be verified; if the trajectory to be verified fails the collision time constraint verification, the trajectory to be verified is marked as an unusable trajectory; if the trajectory to be verified passes the collision time constraint verification, it is further determined whether the trajectory to be verified passes both lane boundary constraint verification and traffic light constraint verification at the same time. If so, the trajectory to be verified is marked as a verified trajectory; otherwise, the trajectory to be verified is marked as an unusable trajectory. The control module is configured to select a target trajectory from the multiple verified trajectories if such trajectories exist, and control the vehicle to drive according to the target trajectory; if no verified trajectory exists, generate a reversal signal to indicate that the vehicle needs to perform a hazard avoidance operation. The collision time constraint check is used to verify whether the vehicle has a collision risk; the lane boundary constraint check is used to verify whether the vehicle has a risk of going out of the lane boundary; and the traffic light constraint check is used to verify whether the vehicle has a risk of crossing the stop line when the traffic light is red.

[0016] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the vehicle control method based on autonomous driving trajectory verification described above.

[0017] To achieve the above objectives, another aspect of this application provides a vehicle that includes the aforementioned in-vehicle desktop dynamic display device or the aforementioned electronic device.

[0018] The embodiments of this application include at least the following beneficial effects: This application provides a vehicle control method, device, equipment, and vehicle based on autonomous driving trajectory verification. The method first performs collision time constraint verification on the trajectory to be verified, and directly marks the trajectory that fails the collision time constraint verification as an unusable trajectory, thereby terminating the subsequent verification process of the trajectory with fatal collision risk. Only for the trajectory that passes the collision time constraint verification, lane boundary constraint verification and traffic light constraint verification are carried out in sequence. If at least one verified trajectory that meets the safety and compliance requirements can be selected in the end, a target trajectory is selected from each verified trajectory, and the vehicle is controlled to drive according to the target trajectory. If no verified trajectory that meets the safety and compliance requirements can be selected in the end, it means that each trajectory to be verified is an unusable trajectory that does not meet safety and compliance requirements. Then, a risk avoidance backoff signal is generated to enable the vehicle to perform subsequent emergency safety operations, avoiding loss of control when there is no effective safe trajectory to execute. Compared with traditional technologies, this application constructs a trajectory screening mechanism that aligns with the safety priorities of autonomous driving by prioritizing collision time constraint verification. When a trajectory is identified as having a fatal collision risk, subsequent compliance verification is directly stopped. This solves the problem of low efficiency in driving decision execution caused by a large amount of redundant calculation due to disordered verification in traditional technologies. At the same time, by prioritizing the screening and elimination of trajectories with fatal collision risks, this invention avoids the problem of missing fatal collision risks while focusing on compliance verification. The priority-based verification method realizes the execution logic that the life safety priority of autonomous driving is higher than traffic rule compliance and that missing backtracking is unacceptable, thus filling driving safety loopholes and meeting the safety assurance requirements of end-to-end autonomous driving vehicle deployment. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a vehicle control method based on autonomous driving trajectory verification provided in an embodiment of this application; Figure 2 This is a schematic diagram of the overall architecture of the vehicle-mounted deployment system provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of a vehicle control device based on autonomous driving trajectory verification provided in an embodiment of this application; Figure 4 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0021] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0022] As used in this application, the terms "several", "each", etc., "several" include one, two or more, "each" refers to each of the corresponding plurality, and "any" refers to any one of the plurality.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0024] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0025] (1) TTC (Time To Collision) refers to the estimated time when a vehicle collides with a target in the surrounding area that has a risk of intersection while traveling along the planned trajectory. It can be calculated based on the relative distance and relative speed collected by millimeter-wave radar. After defining the closing speed, the value is calculated and used to determine whether the vehicle has a fatal collision risk.

[0026] (2) The constraint dictionary (CD) is the carrier for the same-caliber verification between training and deployment in this invention. The offline supervision extractor compiles the collision time, lane boundary, and traffic light constraints into a structured template, which includes parameters such as constraint type, priority, threshold group, time window, and evidence chain field. After being deployed to the vehicle, it provides a unified judgment benchmark for consistency verification and rollback judgment, avoiding the problem of inconsistent verification rules between the training and deployment stages.

[0027] (3) MRM (Minimum Risk Maneuver), which is the highest level of emergency avoidance operation in this application. When the vehicle triggers MRM, the system will prioritize deceleration to a stop and other avoidance operations. This instruction has a higher priority than the normal driving trajectory and the normal reversal trajectory, and is used to avoid fatal collision risks, in order to meet the safety criterion that missed reversal is unacceptable in autonomous driving.

[0028] In the field of autonomous driving trajectory safety verification, traditional technologies typically employ parallel or random sequential verification without prioritizing the three types of constraints: collision time, lane boundaries, and traffic lights, without distinguishing the order of execution based on risk levels. On the one hand, this results in continued compliance verification of lane boundaries and traffic lights for trajectories already at risk of fatal collisions, generating a large amount of invalid and redundant calculations, leading to poor real-time trajectory selection and slow decision-making response. On the other hand, because collision risks are not prioritized for verification, it is easy to overlook fatal collision hazards while focusing on compliance, resulting in missed rollbacks. This fails to meet the unacceptable rollback requirements of autonomous driving and the safety requirement that life safety takes precedence over traffic rules, posing a serious driving safety hazard.

[0029] In view of this, this application provides a vehicle control method, device, equipment, and vehicle based on autonomous driving trajectory verification. This scheme employs a priority-based verification method and a logic of rapid termination upon failure of collision time constraint verification for each trajectory to be verified. First, collision time constraint verification is performed on the trajectory to be verified. Trajectories that fail the collision time verification and pose a fatal collision risk are directly marked as unusable trajectories, and all subsequent verification processes are terminated. Only trajectories that pass the collision time verification are then sequentially subjected to lane boundary constraint verification and traffic light constraint verification. Therefore, by prioritizing the highest-risk collision time verification, this application eliminates invalid compliance verification for high-risk trajectories, reduces redundant calculations, improves the real-time performance of trajectory selection and decision-making, and can prioritize the identification and exclusion of trajectories with fatal collision risks. This avoids the problem of missed rollbacks caused by subsequent collision verification, adheres to the autonomous driving safety design principle that prioritizes life safety, and thus solves the problems of low verification efficiency, easy missed rollbacks, and insufficient safety guarantees in traditional technologies.

[0030] Figure 1This is an optional flowchart of a vehicle control method based on autonomous driving trajectory verification provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S1 to S3: Step S1: When the vehicle is in autonomous driving mode, obtain several trajectories to be verified predicted by the autonomous driving strategy within the current time window; As an illustration, when a vehicle enters autonomous driving mode, the autonomous driving planning program predicts multiple smooth driving trajectories within the next 3 to 5 seconds at fixed time windows, such as 100ms or 200ms, forming a set of trajectories to be verified. Optionally, the trajectories to be verified consist of a series of continuous trajectory points, including kinematic information such as position, heading, speed, and acceleration, which serve as the basis for subsequent safety verification.

[0031] It is understandable that the number and form of predicted trajectories vary under different autonomous driving strategies and road conditions, but all of them must undergo unified safety constraint verification to avoid directly executing dangerous trajectories.

[0032] Step S2: For each trajectory to be verified, a collision time constraint verification is performed on the trajectory to be verified; if it is determined that the trajectory to be verified fails the collision time constraint verification, the trajectory to be verified is marked as an unusable trajectory; if it is determined that the trajectory to be verified passes the collision time constraint verification, it is further determined whether the trajectory to be verified passes both the lane boundary constraint verification and the traffic light constraint verification at the same time. If so, the trajectory to be verified is marked as a verified trajectory; otherwise, the trajectory to be verified is marked as an unusable trajectory. The collision time constraint check is used to verify whether the vehicle has a collision risk; the lane boundary constraint check is used to verify whether the vehicle has a risk of going out of the lane boundary; and the traffic light constraint check is used to verify whether the vehicle has a risk of crossing the stop line when the traffic light is red. In illustrative terms, this embodiment employs a priority-based and short-circuit elimination verification logic. First, the highest-risk collision time verification is performed. Once a trajectory is determined to have a collision risk, it is immediately marked as unusable and subsequent verifications are terminated to avoid invalid calculations. Then, only safe trajectories that pass the collision time verification are further verified for lane boundary and traffic light compliance. Only trajectories that simultaneously meet all three constraints can become verified trajectories.

[0033] Through the above mechanism, deadly and dangerous trajectories can be quickly eliminated from the source, prioritizing driving and passenger safety, while reducing the waste of computing power and decision-making delays caused by redundant verification.

[0034] Step S3: If multiple verified trajectories exist, select a target trajectory from each of the verified trajectories and control the vehicle to drive according to the target trajectory; if no verified trajectory exists, generate a reversal signal to indicate that the vehicle needs to perform a hazard avoidance operation.

[0035] Indicatively, when multiple qualified trajectories exist, this invention can randomly or preferentially select one trajectory and send it to the vehicle's drive-by-wire chassis for execution; when all trajectories fail to meet safety or compliance requirements, no predicted trajectory is executed, and a safety backoff signal is directly triggered to alert the vehicle to be vigilant and perform evasive maneuvers at any time, preventing the vehicle from continuing to drive without safety assurance.

[0036] Steps S1 to S3 as illustrated in this application embodiment first perform collision time constraint verification on the trajectory to be verified output by the autonomous driving strategy, and then only perform lane boundary and traffic light constraint verification on the trajectory that passes the collision verification. This adopts a risk priority progressive verification and a fast elimination of unqualified trajectories, which eliminates redundant compliance verification of trajectories with fatal collision risks and greatly improves the execution efficiency of trajectory verification and screening. On the other hand, this application embodiment can identify and eliminate high-risk collision trajectories, thereby avoiding the problem of missing backoff due to fatal collision risks in the verification process. It strictly adheres to the autonomous driving safety logic that prioritizes life safety over traffic rule compliance. At the same time, it avoids lane crossing and red light running risks through multi-dimensional constraint verification. When there is no safe and compliant trajectory, it triggers the generation of a risk avoidance backoff signal, which can form a safety net and comprehensively improve the safety and compliance of autonomous driving.

[0037] In some embodiments, the current time window is a time period corresponding to 10Hz; optionally, in embodiments of the present invention, collision time constraint verification can be performed on the trajectory to be verified predicted by the autonomous driving strategy every 20Hz time period.

[0038] Understandably, in addition to the trajectory planning and verification link every 10Hz, this invention also sets up an independently operating 20Hz fast monitoring link to fill the safety gap between the 10Hz planning cycles during autonomous driving, further eliminating the risk of missed backoffs. Because the 20Hz fast monitoring link independently verifies the collision time constraint of the vehicle's current driving trajectory and the trajectory to be verified output by the autonomous driving strategy every 20Hz time period, without relying on the 10Hz verification result, once a collision time less than the emergency threshold is detected, an emergency backoff signal is directly sent to the backoff unit, prioritizing the vehicle's evasive maneuver. This eliminates the need to wait for the next 10Hz planning cycle, ensuring that fatal collision risks can be quickly identified and dealt with promptly, thus further strictly adhering to the autonomous driving safety design principle of prioritizing life safety.

[0039] For step S2, in some embodiments, the collision time constraint verification of the trajectory to be verified includes: Based on the vehicle's internal operating data and external road condition data, the collision time of the vehicle with other vehicles within a preset time window is calculated when the vehicle travels along the trajectory to be verified. When it is determined that the collision time is less than the first preset collision time threshold, the verification result that the trajectory to be verified has failed the collision time constraint verification is output. When it is determined that the collision time is not less than the first preset collision time threshold, the verification result of the trajectory to be verified passing the collision time constraint is output.

[0040] Schematic representation: Vehicle operation data may include vehicle speed, acceleration, and steering angle; road condition data may include the trajectory of surrounding targets, relative distance, and relative speed. This invention calculates the minimum collision time within the next 2-3 seconds using kinematics and compares it with a first preset collision time threshold. If the time is lower than the threshold, a collision risk is determined. The first preset collision time threshold can be set to 2.5 seconds.

[0041] Optionally, this invention can use the relative distance d and relative velocity vrel provided by millimeter-wave radar as the calculation data source. The specific calculation formula is: closing velocity vclose=max(-vrel,0), which can retain only the positive closing velocity of vehicles approaching each other; collision time TTC=d / max(vclose,ε), where ε is a small constant to prevent division by zero. In multi-target scenarios, this invention can prioritize the screening of targets that have a geometric intersection risk with the trajectory to be verified of the vehicle, such as targets moving forward in the same lane or targets that are predicted to intersect, and select the minimum collision time (minTTC) as the hard constraint judgment index.

[0042] When it is determined that the collision time TTC is less than the first preset collision time threshold, the verification result that the trajectory to be verified has failed the collision time constraint verification is output. Indicatively, vehicle operation data may include vehicle speed, acceleration, and steering angle; road condition data can be based on the relative distance and relative speed of surrounding targets collected by millimeter-wave radar, assisted by the fusion of other perception data to ensure the reliability of the computational basis. Through the above calculation process, this invention not only ensures the accuracy of collision time calculation but also improves the stability of verification under complex road conditions by leveraging the advantages of millimeter-wave-dominated data sources.

[0043] In some embodiments, the present invention can perform a unified and streamlined process for dual compliance verification of lane boundaries and traffic lights, resulting in: The step of determining whether the trajectory to be verified has passed both lane boundary constraint verification and traffic light constraint verification simultaneously, and if so, marking the trajectory to be verified as a verified trajectory; otherwise, marking the trajectory to be verified as an unusable trajectory, includes: Based on each trajectory point in the trajectory to be verified and the road condition data, lane boundary constraint verification is performed on the trajectory to be verified. Based on each trajectory point in the trajectory to be verified, the running data, and the traffic data, the trajectory to be verified is subjected to traffic light constraint verification. If the trajectory to be verified passes both the lane boundary constraint verification and the traffic light constraint verification, the trajectory to be verified is marked as the verified trajectory. If the trajectory to be verified fails the lane boundary constraint verification or the traffic light constraint verification, the trajectory to be verified will be marked as the unusable trajectory.

[0044] As an illustration, lane boundaries and traffic light constraints are both driving compliance constraints, and both must be met simultaneously to enter the candidate trajectory pool; failure to meet either one will result in direct elimination.

[0045] This embodiment further refines the calculation method for lane boundary constraint verification and uses the minimum boundary margin as the criterion. Therefore, when performing lane boundary constraint verification on the trajectory to be verified, the following applies: The step of performing lane boundary constraint verification on the trajectory to be verified based on each trajectory point in the trajectory to be verified and the road condition data includes: Extract the left lane boundary and right lane boundary from the road condition data; Calculate the lateral distance from each trajectory point to the left lane boundary and the right lane boundary, and select the lateral distance with the smallest value as the minimum boundary margin of the trajectory to be verified. When the minimum boundary margin is determined to be less than the preset boundary margin threshold, the verification result of the trajectory to be verified failing the lane boundary constraint verification is output; wherein, optionally, the boundary margin threshold can be set to 0.3m to 0.5m, and any point along the entire trajectory should not be too close to the lane line to avoid driving over the line or crossing the boundary.

[0046] When it is determined that the minimum boundary margin is not less than the preset boundary margin threshold, the verification result of the trajectory to be verified passing the lane boundary constraint verification is output.

[0047] Furthermore, this embodiment performs refined timing and distance-based dual judgments on traffic light constraint verification to prevent red light crossings. Therefore, when performing traffic light constraint verification on the trajectory to be verified, the following applies: The step of performing traffic light constraint verification on the trajectory to be verified based on each trajectory point in the trajectory to be verified, the running data, and the traffic data includes: Extract the stop line boundaries and traffic light status information from the road condition data; Based on the running data and each of the trajectory points, the running speed corresponding to each trajectory point is extracted; When it is determined that the red light is in a valid state based on the traffic signal light status information, the valid red light period corresponding to the red light status is obtained; Extract each trajectory point to be verified within the red light's effective time period, and the corresponding running speed to be verified for each trajectory point to be verified; Calculate the longitudinal distance from each of the trajectory points to be verified to the boundary of the stop line; Determine whether there are any trajectory points to be verified whose longitudinal distance is less than a preset parking distance threshold; If so, output the verification result that the trajectory to be verified has failed the traffic light constraint verification; If not, calculate the stop line arrival time corresponding to each of the longitudinal distances and the running speeds to be verified; if it is determined that there is a stop line arrival time less than a preset parking time threshold, output the verification result that the trajectory to be verified has failed the traffic light constraint verification; if it is determined that there is no stop line arrival time less than the preset parking time threshold, output the verification result that the trajectory to be verified has passed the traffic light constraint verification. The stop line arrival time is used to characterize the time when the front reference position of the vehicle reaches the boundary of the stop line.

[0048] In illustrative terms, the traffic light constraint verification mechanism of this invention can determine the passability of a red light from two dimensions: whether the distance is sufficient and whether there is enough time, so as to prevent vehicles from forcibly crossing the stop line during a red light and ensure compliance with traffic rules.

[0049] In this embodiment of the invention, the following advantages or beneficial effects are achieved: Traditional technologies often employ independent and decentralized verification logic when verifying lane boundaries and traffic light constraints. Either they verify compliance constraints first and then collision risks, resulting in redundant compliance verification for trajectories already at risk of fatal collisions, reducing decision-making efficiency; or they use parallel verification of the two types of compliance constraints without priority, with vague verification standards and a single judgment dimension. For example, lane boundary verification simply judges whether the trajectory crosses the line without considering the proximity of the trajectory point to the lane boundary, easily leading to safety hazards such as driving over the line. Traffic light verification only judges whether the stop line is crossed when the light is red, without combining driving speed and arrival time for timing prediction, easily leading to running a red light due to untimely braking, failing to comprehensively avoid compliance risks. Furthermore, the results of the two types of compliance verifications in traditional technologies are not uniformly integrated for judgment, easily leading to misjudgments where passing a single constraint is considered compliant, causing non-compliant trajectories to enter the candidate pool, affecting the safety and compliance of autonomous driving.

[0050] This invention provides a unified and streamlined process for dual compliance verification of lane boundaries and traffic lights. It clearly stipulates that lane boundary and traffic light constraint verification must be performed sequentially after passing the collision time constraint verification. Only trajectories that pass both types of verification can be marked as verified trajectories. If either requirement is not met, the trajectory is directly eliminated. This follows the autonomous driving safety principle that prioritizes life safety over compliance. Furthermore, the unified process avoids problems such as confusing verification logic and misjudgments or omissions.

[0051] Meanwhile, this invention refines and optimizes two types of compliance checks. In the lane boundary constraint check, the minimum boundary margin is used as the criterion. By extracting the left and right lane boundaries and calculating the lateral distance from each trajectory point to the boundary and selecting the minimum value, the closeness between the trajectory and the lane boundary is accurately measured. Combined with a preset threshold of 0.3m to 0.5m, the risk of crossing the line or going out of bounds caused by the trajectory being too close to the lane line is avoided. Compared with the traditional simple line crossing judgment, it is more in line with the actual road driving scenario and improves the accuracy and rigor of lane boundary check.

[0052] Furthermore, the traffic light constraint verification of this invention adopts a dual judgment dimension of distance and time sequence, breaking the limitations of traditional single distance judgment. First, the stop line boundary and traffic light status are extracted. During the valid red light period, it is first determined whether the longitudinal distance from the trajectory point to the stop line meets the preset parking distance threshold, preventing situations where the distance is too close and the vehicle cannot stop in time. If the distance meets the threshold, the arrival time at the stop line is calculated by combining the running speed corresponding to the trajectory point, determining whether the vehicle can stop safely during the red light period, avoiding the risk of running a red light due to insufficient arrival time when the vehicle is traveling too fast. Therefore, the embodiments of this invention can cover compliance risks in red light scenarios from two dimensions: whether the distance is sufficient to stop and whether the time is sufficient to stop, ensuring that vehicles will not forcibly cross the stop line during red lights and strictly abide by traffic rules.

[0053] Meanwhile, the dual compliance verification process of this invention relies on a constraint dictionary compiled offline, employing standardized verification thresholds and calculation methods. This ensures consistent application across the training and deployment phases, resolving the verification deviation issues caused by inconsistent verification standards and methods in traditional technologies. Furthermore, both types of verification are based on various trajectory points, vehicle operation data, and road condition data of the trajectory to be verified, combined with refined calculation logic to ensure the accuracy and reliability of the verification results. Therefore, this invention can adapt to lane boundary and traffic light control requirements in different scenarios such as urban roads and highways.

[0054] For step S3, in some embodiments, the present invention can be based on the optimal selection mechanism under multiple trajectories, and achieve comprehensive selection of safety and compliance based on weighted consistency scores.

[0055] Selecting a target trajectory from each of the verified trajectories includes: For each of the verified trajectories, the collision violation component corresponding to the collision time, the lane boundary violation component corresponding to the minimum boundary margin, and the traffic light violation component corresponding to the longitudinal distance are combined to obtain the target violation vector corresponding to the verified trajectory. For each verified trajectory, a consistency score is calculated by weighting and summing the target violation vector, the first weight corresponding to the collision time constraint verification, the second weight corresponding to the lane boundary constraint verification, and the third weight corresponding to the traffic light constraint verification; wherein, the first weight is greater than the second weight, and the second weight is greater than the third weight. The verified trajectory with the highest consistency score is taken as the target trajectory; wherein the consistency score is used to indicate the degree of compliance with collision time constraints, lane boundary constraints and traffic light constraints.

[0056] In some embodiments, the present invention may also introduce a confidence mechanism for the construction of violating components to improve the reliability of the verification results.

[0057] In this embodiment, the generation of the target violation vector includes: The collision violation component is constructed based on the collision time, the first preset collision time threshold, and the collision time confidence level; wherein, the collision time confidence level is used to indicate the reliability of the collision time calculation result; The lane boundary violation component is constructed based on the minimum boundary margin, the preset boundary margin threshold, and the lane boundary confidence; wherein, the lane boundary confidence is used to indicate the reliability of the lateral distance calculation results. The traffic light violation component is constructed based on the stop line arrival time, the preset parking time threshold, the red light valid period, and the parking distance confidence level; wherein, the parking distance confidence level is used to indicate the reliability of the longitudinal distance calculation result; The collision violation component, the lane boundary violation component, and the traffic light violation component are combined to output the target violation vector.

[0058] Indicatively, the confidence level is dynamically determined by factors such as the type of sensing sensor, occlusion conditions, and ranging accuracy. Lower confidence levels result in a higher weight for the violation component, enabling the onboard system to make conservative and safe trajectory decisions even in environments with uncertain perception. By introducing confidence levels into the calculation of violation components, the verification results no longer rely solely on hard threshold decisions but are dynamically adjusted based on perception reliability. This allows for safe decision-making even in complex environments such as rain, fog, backlight, and occlusion, significantly improving the system's adaptability and safety in complex road conditions.

[0059] In this embodiment of the invention, the following advantages or beneficial effects are achieved: Based on an optimal selection mechanism under multiple trajectories, this invention constructs a target violation vector and a weighted consistency score, realizing a comprehensive selection of safety and compliance while strictly adhering to the risk priority principle. This invention first combines the collision violation component corresponding to the collision time, the lane boundary violation component corresponding to the minimum boundary margin, and the traffic light violation component corresponding to the longitudinal distance into a target violation vector, comprehensively covering the trajectory's performance in both safety and compliance dimensions, avoiding the one-sidedness of traditional single-dimensional selection. Subsequently, by calculating the consistency score through weighted summation, the first weight (collision time constraint) > the second weight (lane boundary constraint) > the third weight (traffic light constraint) are clearly set to ensure that in the process of selecting the best trajectory, the safety performance of the trajectory takes precedence over the compliance performance, and the trajectory with higher safety redundancy is selected first, which fully meets the requirement of life safety first in autonomous driving and avoids collision risks to the greatest extent from the decision-making level.

[0060] Furthermore, this invention innovatively introduces a confidence level mechanism in the generation process of the target violation vector, breaking the limitations of traditional technologies that rely solely on hard threshold decisions. The confidence levels for collision time, lane boundary, and stopping distance correspond to the reliability of various verification data, which can be dynamically determined by factors such as the type of sensing sensor, occlusion conditions, and ranging accuracy. When the accuracy of the sensing data decreases or the confidence level drops, the corresponding violation component weight will increase accordingly. This allows the system to automatically adjust the judgment weights in uncertain perceptual environments, making more conservative and safer trajectory decisions and avoiding selection errors caused by perceptual noise. For example, in rainy or foggy weather, the ranging accuracy of millimeter-wave radar decreases, and the confidence level for collision time decreases. In this case, the collision violation component weight increases, and the system will prioritize trajectories with higher collision safety redundancy, maximizing driving safety even if their compliance varies slightly.

[0061] Meanwhile, the weighted consistency scoring mechanism and confidence calibration logic of this invention both rely on the constraint dictionary compiled offline, and adopt standardized calculation methods and weight parameters, achieving consistent application in both the training and deployment phases. This solves the trajectory selection bias problem caused by inconsistent selection rules and confusing judgment logic in traditional technologies. Therefore, through the synergistic effect of target violation vector construction, weighted consistency scoring selection, and confidence calibration mechanism, this invention not only solves the problems of one-sided trajectory selection and confused risk priority in traditional technologies, but also improves the reliability and conservatism of trajectory decision-making in complex perception environments, further enhancing the safety, stability, and reliability of autonomous driving, and strengthening the engineering adaptability and robustness of the entire trajectory verification scheme.

[0062] In some embodiments, the backoff logic for cases without a qualified trajectory can be further refined to achieve graded risk avoidance based on risk type, thereby improving the rationality and smoothness of the backoff strategy.

[0063] In this embodiment, generating a backoff signal to indicate that the vehicle needs to perform a hazard avoidance operation if the verified trajectory does not exist includes: If the verified trajectory does not exist, determine whether there is an unusable trajectory that failed the collision time constraint verification; If so, a first retreat signal is generated; wherein the first retreat signal is used to indicate that the vehicle needs to perform a minimum-risk maneuver to avoid danger; If not, a second backoff signal is generated; wherein the second backoff signal is used to indicate that the vehicle needs to perform a trajectory correction avoidance operation.

[0064] Indicatively, the first reversal signal corresponds to a high-risk collision scenario, requiring the activation of the highest level of Minimum Risk Maneuver (MRM); the second reversal signal corresponds to a scenario with no collision risk but not compliance requirements, requiring only deceleration and correction within the original lane, without the need for emergency evasive maneuvers. This invention, by differentiating reversals based on risk levels, ensures vehicle safety in extremely dangerous situations while avoiding unnecessary sudden braking that degrades the driving experience.

[0065] In this embodiment of the invention, the following advantages or beneficial effects are achieved: Traditional technologies, when no usable trajectory is available, perform a uniform emergency braking or pullover operation regardless of whether it is a high-risk scenario with a fatal collision risk or a low-risk scenario with only compliance issues such as lane departure or running a red light, without distinguishing the root cause and level of risk. However, in scenarios with only compliance risks and no collision hazards, unnecessary emergency braking can seriously affect driving comfort and may even cause secondary risks such as rear-end collisions. On the other hand, if the highest level of emergency avoidance is not performed in high-risk collision scenarios, it will result in missed retreat, failing to guarantee the safety of drivers and passengers and violating the safety principle that missed retreat is unacceptable.

[0066] This invention employs a refined judgment and signal generation mechanism for the backoff logic when no qualified trajectory is available. First, based on the cause of the unusable trajectory failure, it distinguishes between collision risk and compliance risk, achieving tiered risk avoidance and ensuring the rationality and relevance of the backoff strategy. When no verified trajectory exists, it first checks if there is an unusable trajectory that has failed the collision time constraint check. If so, it indicates a high-risk collision scenario, and a first backoff signal is immediately generated, instructing the vehicle to perform the highest level of risk avoidance operation—Minimum Risk Maneuver (MRM)—to minimize the risk of fatal collisions and strictly adhere to the principle of prioritizing life safety. If no such unusable trajectory exists, it indicates only compliance risk exists, with no collision hazard. A second backoff signal is generated, instructing the vehicle to perform trajectory correction operations only within the original lane, without the need for emergency braking. This ensures compliance with traffic rules while avoiding unnecessary impact and secondary risks from sudden braking. Therefore, this invention adopts a tiered backoff mechanism, precisely matching risk levels with backoff operations, providing a safety net in extremely dangerous scenarios while maintaining driving comfort in low-risk scenarios.

[0067] In some embodiments, to avoid frequent triggering and cancellation of backoff signals due to fluctuations in a single verification, a continuous time window hysteresis determination mechanism is introduced to improve system stability.

[0068] After generating the first backoff signal, the method further includes: When it is determined that the first rollback signal has been generated in multiple consecutive time windows, the vehicle is controlled to perform a minimum-risk maneuver to decelerate to a stop.

[0069] As an example, hysteresis judgment conditions can be set for 3 to 5 consecutive time windows. Only when a high risk of collision is detected in multiple consecutive frames and the first back-off signal is triggered will the minimum risk maneuver of deceleration and braking be officially executed to avoid false triggering due to instantaneous perceived noise.

[0070] After generating the second backoff signal, the method further includes: When it is determined that the second rollback signal has been generated in multiple consecutive time windows, the vehicle speed is reduced and the vehicle is kept traveling in the current lane to complete the trajectory correction operation.

[0071] As an illustration, after multiple consecutive frames confirming compliance anomalies, the vehicle gradually slows down and remains centered in the lane, waiting for the autonomous driving strategy to re-output a compliant trajectory, thus ensuring both traffic rule compliance and vehicle driving stability.

[0072] In this embodiment of the invention, the following advantages or beneficial effects are achieved: The invention introduces a hysteresis judgment mechanism based on continuous time windows, solving the problem of false triggering or frequent switching of backoff signals caused by fluctuations in single verification and sensing noise in traditional technologies. After generating the first backoff signal, the minimum risk maneuver to deceleration to a complete stop is not immediately executed. Instead, the operation is only formally executed when the first backoff signal is generated in multiple consecutive time windows, such as 3 to 5 time windows. Thus, the invention filters out misjudgments caused by instantaneous sensing noise through multiple time windows, avoiding unnecessary emergency braking caused by single data fluctuations. This ensures timely emergency avoidance in high-risk collision scenarios and prevents safety hazards and driving discomfort caused by false triggering.

[0073] After generating the second rollback signal, it is also necessary to confirm that the compliance anomaly has not been alleviated within multiple consecutive time windows before the vehicle can be controlled to gradually reduce speed and maintain driving in the current lane, complete trajectory correction, avoid frequent speed reduction or adjustment of the vehicle due to instantaneous compliance fluctuations, ensure the smoothness of vehicle driving, and wait for the autonomous driving strategy to re-output a new compliance trajectory to achieve a smooth transition from rollback to normal driving.

[0074] Therefore, this invention effectively filters out frequent state switching caused by single-verification jitter through continuous multi-window hysteresis logic, avoiding sudden stops and starts, and jerking behavior of the vehicle, thereby improving the robustness and driving comfort of the autonomous driving system.

[0075] Please see Figure 2 The schematic diagram of the overall architecture of the vehicle-mounted deployment system shown illustrates, in a preferred embodiment, the implementation process of the above-described method of the present invention can be based on... Figure 2 The diagram shows a dual-architecture implementation of offline constraint dictionary compilation and an onboard pluggable safety shell.

[0076] It is understandable that the architecture is divided into two main layers: the offline training supervision layer and the vehicle deployment execution layer. The entire architecture of this invention does not require modification of the main structure of the end-to-end autonomous driving policy network. It can achieve full-process safety control only through the pluggable safety shell layer, and has strong engineering adaptability and practical value.

[0077] Specifically, during the offline operation phase, the device inputs the collected training data into the supervised extractor to extract the verification thresholds, priority logic, and weight parameters of three core safety rules: collision time constraints, lane boundary constraints, and traffic light constraints. After forming a constraint template, it is input into the constraint dictionary CD compiler to compile it into a standardized and configurable constraint dictionary CD. This dictionary is then distributed to the vehicle in a standardized format, providing a unified judgment benchmark for vehicle consistency verification. This addresses the problem of inconsistent training rules and deployment verification standards or unconfigurable verification logic in traditional technologies from the source, ensuring the consistency and traceability of the verification logic throughout the entire process.

[0078] During the vehicle deployment phase, the security shell operates in a dual-frequency collaborative architecture, consisting of a 10Hz main planning link and a 20Hz fast safety monitoring link. The two links operate independently yet collaborate with each other, fully realizing all the technical solutions described in this application.

[0079] Among them, the 10Hz planning main link corresponds to the core process of steps S1 to S3 of this application. First, the sensor perception module collects the vehicle's internal operating data in real time, such as vehicle speed, acceleration and steering angle, and collects the road condition data outside the vehicle, such as the trajectory of surrounding targets, lane lines and traffic light status. The fused data is input into the end-to-end policy network module. This module uses the time period corresponding to 10Hz as the current time window and outputs K candidate trajectories to be verified in the next 3 to 5 seconds, providing basic input for subsequent verification.

[0080] Subsequently, the trajectory to be verified output by the end-to-end policy network is input into the consistency verifier. The consistency verifier calls the constraint dictionary CD issued offline and executes the priority progressive verification and elimination logic described in step S2 of this application. First, collision time constraint verification is performed on each trajectory to be verified. The minimum collision time TTC within the preset time window is calculated by combining vehicle operation data and road condition data. If it is determined that the TTC is less than the first preset collision time threshold, the trajectory is directly marked as an unusable trajectory, and the subsequent lane boundary and traffic light constraint verification is terminated. Trajectories with fatal collision risks are quickly eliminated from the source, avoiding meaningless compliance verification of high-risk trajectories, greatly reducing redundant calculations and improving trajectory screening efficiency. Next, for trajectories that pass the collision time constraint check, the consistency checker continues to perform lane boundary constraint check and traffic light constraint check. When performing lane boundary constraint check, the left and right lane boundaries in the road condition data are extracted, the lateral distance from each trajectory point to the lane boundary is calculated, and the minimum value is selected as the minimum boundary margin. If the minimum boundary margin is less than the preset boundary margin threshold, the check is deemed to have failed. When performing traffic light constraint check, the corresponding trajectory points are extracted during the red light validity period, the longitudinal distance from the trajectory point to the stop line and the arrival time of the stop line are calculated, and the risk of running a red light is determined from both distance and time dimensions. Only when the trajectory passes both lane boundary and traffic light compliance checks simultaneously is it marked as a checked trajectory.

[0081] Furthermore, the output of the consistency verifier is divided into two paths. One path is input into the evidence chain / black box module, which can simultaneously record the verification results, violation components, perception confidence, trigger timestamps, and other full-process data for each trajectory, realizing full-link traceability of trajectory verification and rollback triggering, and meeting the requirements of autonomous driving data security auditing and accident tracing. The other path is used in the trajectory optimization and rollback determination stage of the consistency verifier. At this time, the processing of the consistency verifier is as follows: if there are multiple verified trajectories, the consistency verifier will select the collision violation component corresponding to the collision time, the lane boundary violation component corresponding to the minimum boundary margin, and the longitudinal distance... The corresponding traffic light violation components are combined into a target violation vector. When constructing the violation components, collision time confidence, lane boundary confidence, and stopping distance confidence are incorporated to adapt to scenarios with fluctuating perception data. Then, according to the rule that the first weight of the collision time constraint is greater than the second weight of the lane boundary constraint, and the second weight of the lane boundary constraint is greater than the third weight of the traffic light constraint, the consistency score of each verified trajectory is calculated by weighted summation. The trajectory with the highest score is selected as the target trajectory and sent to the rollback arbitrator. If there is no verified trajectory, the consistency verifier sends a rollback trigger command to the rollback arbitrator, which then performs a hierarchical rollback operation.

[0082] To fill the safety gap between the 10Hz planning cycles of autonomous driving strategies and avoid missed rollbacks due to planning delays, this invention's architecture sets up an independently operating 20Hz fast safety monitoring link, corresponding to... Figure 2 The fast safety monitor (20Hz) operates at a high frequency with a time period corresponding to 20Hz. It only monitors the highest risk collision time constraints quickly and does not rely on the trajectory output of the 10Hz end-to-end policy network. It calculates the minimum collision time (TTC) of the current vehicle's driving trajectory in real time. Once the TTC is detected to be lower than the emergency threshold in multiple consecutive time windows, it directly sends the first backoff signal to the backoff arbitrator to trigger the minimum risk maneuver operation without waiting for the next 10Hz planning cycle, thus eliminating the risk of missed backoff from the mechanism.

[0083] As the scheduling unit of the entire architecture, the rollback arbitrator can receive three input signals: the target trajectory issued by the consistency checker, the MRM trigger signal issued by the fast safety monitor, and the rollback trajectory output by the rule planner and the MRM minimum risk maneuver module. It performs scheduling based on hierarchical rollback and hysteresis control logic. Specifically, when a valid verified trajectory exists in the 10Hz check, the backoff arbitrator prioritizes executing the target trajectory to control the vehicle's normal driving. When no verified trajectory exists, the backoff arbitrator triggers tiered backoff based on the failure type of the unusable trajectory: if an unusable trajectory fails the collision time constraint check, it is determined to be a high-risk collision scenario, triggering the first backoff signal and calling the MRM minimum risk maneuver module to output an emergency backoff trajectory. Only when the first backoff signal is generated in multiple consecutive time windows will the vehicle be controlled to perform a minimum risk maneuver to decelerate to a stop. If only an unusable trajectory fails the lane boundary or traffic light constraint check, it is determined to be a compliance anomaly scenario, triggering the second backoff signal and calling the rule planner to output a regular backoff trajectory. After confirmation in multiple consecutive time windows, the vehicle is controlled to reduce speed and maintain driving in the current lane to complete trajectory correction.

[0084] Finally, the reversal arbitrator sends the selected final execution trajectory to the controller, which sends trajectory point control commands to the vehicle's drive-by-wire chassis at a frequency of 20Hz, controlling the vehicle to travel according to the target trajectory or the reversal trajectory, thus completing the closed-loop control of the entire process.

[0085] To more clearly illustrate the actual operation of this architecture, this embodiment of the invention takes a scenario of a vehicle cutting into the lane from an adjacent lane as an example: The vehicle is driving autonomously at 60 km / h on an urban expressway, with the vehicle in front following normally, when the adjacent vehicle suddenly cuts into the vehicle's lane from the adjacent lane. First, the sensor module collects fused data such as the position, speed, and relative distance of the adjacent vehicle, and inputs it into a 10Hz end-to-end policy network. The network outputs 5 candidate trajectories to be verified. The consistency checker calls the constraint dictionary CD and first performs TTC verification on each trajectory. Three trajectories have a TTC below the preset threshold of 2.5s and are directly marked as unusable trajectories, terminating subsequent verification. Only 2 trajectories pass the TTC verification. Then, lane boundary verification is performed on these 2 trajectories. The minimum boundary margin of the trajectory points meets the threshold requirement of 0.3m. Then, traffic light verification is performed. Since the current intersection is green and there is no risk of crossing the line, both trajectories are marked as verified trajectories. The consistency verifier calculates a consistency score based on the target violation vector and weighted rules, selects the trajectory with the highest score as the target trajectory, and sends it to the backoff arbitrator. Simultaneously, the verification result is written to the evidence chain black box. Meanwhile, the fast safety monitor, with a 20Hz cycle, monitors the Time-to-Cross (TTC) in real time. If, at the moment a vehicle cuts in, the TTC is detected to be below the 1.5s emergency threshold for three consecutive time windows, a backoff signal is sent to the backoff arbitrator. The backoff arbitrator determines that there is a high risk of collision and that the backoff signal is present for multiple consecutive time windows. It prioritizes emergency backoff, calls the MRM module to output a deceleration and braking trajectory, sends it to the controller, and controls the vehicle to brake suddenly, successfully avoiding a collision. Once the collision risk is completely eliminated, the backoff arbitrator resumes executing the target trajectory output by the 10Hz planning cycle, and the vehicle resumes normal autonomous driving.

[0086] Compared to traditional technologies, this invention, through the aforementioned dual-architecture, dual-frequency collaborative design, advances the highest-risk collision time verification and achieves short-circuit elimination, significantly reducing redundant calculations and improving real-time decision-making. At the same time, it fills planning gaps through 20Hz high-frequency monitoring, fundamentally avoiding the risk of missed backoffs. Combined with tiered backoff and hysteresis control logic, it balances driving safety, compliance, and ride comfort. Moreover, the architecture of this invention does not depend on a specific hardware platform and can be adapted to various in-vehicle computing units and autonomous driving systems.

[0087] In one embodiment, road construction causes instability in visual lane boundary detection, leading to the autonomous driving output trajectory gradually moving closer to the right side of the lane. However, the time-to-collision (TTC) check remains within the normal range, indicating no risk of an emergency collision. During the execution of the method of this invention, if the trajectory boundary margin is detected to be insufficient to meet the threshold requirement, a regular rollback is triggered after multiple consecutive frames. The rule planner generates a corrected trajectory that maintains the lane and limits the vehicle speed. Once the trajectory boundary margin meets the threshold and remains stable for multiple frames, the rollback mode is exited according to the hysteresis release condition. This scenario can achieve a safety net against non-collision risks based on the regular rollback of this invention, while relying on hysteresis logic to suppress driving vibrations caused by perception anomalies.

[0088] In one embodiment, when the traffic light and stop line perception data are stable and reliable, the trajectory planned by the autonomous driving system may cross the stop line. This invention triggers a regular backoff based on traffic light constraints, generating a parking trajectory and simultaneously recording and verifying the evidence chain. If the traffic light perception data is unstable or fails, the traffic light constraints automatically degrade, retaining only the data record without triggering a backoff operation, preventing meaningless backoffs due to unreliable input. This invention can achieve dynamic constraint adaptation based on perception reliability, improving the robustness of the system in commercial deployment scenarios while adhering to the low priority of traffic light constraints.

[0089] Optionally, the present invention also sets conditions for exiting the minimum risk maneuver operation: when the vehicle is in the minimum risk maneuver operation state, the minimum collision time TTC must be greater than or equal to the MRM release threshold, and this safe state must remain stable for multiple time windows; at the same time, the duration for which the system maintains the minimum risk maneuver operation state has reached the preset minimum holding time before it can switch from the minimum risk maneuver operation state to the normal driving state of autonomous driving.

[0090] Optionally, this invention also sets conditions for exiting the trajectory correction operation. For trajectory correction triggered by lane boundary violations: the minimum boundary margin of the trajectory is greater than the release threshold, and the detection results for multiple consecutive time windows meet the requirements. For trajectory correction triggered by traffic light violations: the trajectory does not pose a risk of crossing the red light line, meets the dual threshold requirements of stopping distance and stop line arrival time, and remains compliant for multiple consecutive time windows; based on meeting the corresponding constraint recovery requirements, the system can exit the trajectory correction state and resume normal planned driving once the minimum holding time is reached during the trajectory correction operation.

[0091] In summary, this application achieves safe and efficient screening of autonomous driving trajectories through a series of collaborative mechanisms, including priority-based verification, graded backoff, hysteresis determination, weighted selection, and confidence calibration. Compared with traditional technologies, it has the following advantages: By adopting a collision-first verification approach, invalid calculations are significantly reduced, and the real-time performance of decisions is improved. At the same time, the process avoids the problem of missing backoffs due to fatal collision risks. Verification is carried out with the priority of collision time > lane boundary > traffic light, with collision time constraints being verified first, followed by lane boundary and traffic light constraints. This improves the real-time performance of verification and meets the safety principle that missing backoffs are unacceptable.

[0092] Through tiered backoff and continuous frame hysteresis logic, it achieves precise risk avoidance in hazardous scenarios and smooth correction in normal scenarios, balancing safety, comfort, and traffic compliance. Furthermore, it employs a 10Hz planning verification system responsible for complete trajectory verification and backoff decision-making, and a 20Hz rapid monitoring system focused solely on collision time detection, possessing emergency takeover authority for the MRM (Match Response Management) to fill the gap in the 10Hz planning cycle and reduce the risk of missed backoffs. Multi-trajectory selection is achieved based on weighted consistency scores, with safety weights higher than compliance weights, which aligns with the safety design ethics of autonomous driving; a perception confidence calibration mechanism is introduced to improve decision robustness in complex road conditions and scenarios with perception uncertainty; The implementation process of this invention does not depend on a specific hardware platform, and it can also output a structured chain of evidence of violation. It can be adapted to various in-vehicle computing units and autonomous driving systems, and is easy to deploy, expand and implement in the automotive grade.

[0093] Please see Figure 3 This application also provides a vehicle control device based on autonomous driving trajectory verification, which can implement the above-mentioned vehicle control method based on autonomous driving trajectory verification. The device includes: The trajectory acquisition module is used to acquire several trajectories to be verified predicted by the autonomous driving strategy within the current time window when the vehicle is in autonomous driving mode. The trajectory verification module is used to perform collision time constraint verification on each trajectory to be verified; if the trajectory to be verified fails the collision time constraint verification, the trajectory to be verified is marked as an unusable trajectory; if the trajectory to be verified passes the collision time constraint verification, it is further determined whether the trajectory to be verified passes both lane boundary constraint verification and traffic light constraint verification at the same time. If so, the trajectory to be verified is marked as a verified trajectory; otherwise, the trajectory to be verified is marked as an unusable trajectory. The control module is configured to select a target trajectory from the multiple verified trajectories if such trajectories exist, and control the vehicle to drive according to the target trajectory; if no verified trajectory exists, generate a reversal signal to indicate that the vehicle needs to perform a hazard avoidance operation. The collision time constraint check is used to verify whether the vehicle has a collision risk; the lane boundary constraint check is used to verify whether the vehicle has a risk of going out of the lane boundary; and the traffic light constraint check is used to verify whether the vehicle has a risk of crossing the stop line when the traffic light is red.

[0094] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0095] It should be noted that the device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0096] Those skilled in the art will clearly understand that, for convenience and simplicity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0097] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned vehicle control method based on autonomous driving trajectory verification. This electronic device can include any smart terminal such as a tablet computer or an in-vehicle computer.

[0098] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0099] Please see Figure 4 , Figure 4 This illustrates the hardware structure of an electronic device according to another embodiment, the electronic device comprising: The processor can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to achieve the technical solutions provided in the embodiments of this application. The memory can be implemented in the form of read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory can store the operating system and other applications. When the technical solutions provided in the embodiments of this application are implemented through software or firmware, the relevant program code is stored in the memory and called and executed by the processor. Input / output interfaces are used to implement information input and output; The communication interface is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, Wi-Fi, Bluetooth, etc.). A bus is used to transfer information between various components of a device, such as processors, memory, input / output interfaces, and communication interfaces. The processor, memory, input / output interface, and communication interface are interconnected within the device via a bus.

[0100] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0101] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0102] This application also provides a vehicle, which includes the aforementioned in-vehicle desktop dynamic display device or the aforementioned electronic device.

[0103] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described photovoltaic power prediction method based on multi-source heterogeneous data.

[0104] It is understood that the content of the above method embodiments is applicable to the present computer storage medium embodiments. The specific functions implemented by the present computer storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0105] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described photovoltaic power prediction method based on multi-source heterogeneous data.

[0106] It is understood that the content of the above method embodiments is applicable to the embodiments of this computer program product. The specific functions implemented by the embodiments of this computer program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0107] Those skilled in the art will understand that all or some of the steps, apparatuses, or functional modules / units in the methods disclosed above can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0108] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A vehicle control method based on autonomous driving trajectory verification, characterized in that, The method includes: When the vehicle is in autonomous driving mode, acquire several trajectories to be verified predicted by the autonomous driving strategy within the current time window; For each of the aforementioned trajectories to be verified, a collision time constraint verification is performed on the trajectory. If it is determined that the trajectory fails the collision time constraint verification, the trajectory is marked as an unusable trajectory. If it is determined that the trajectory passes the collision time constraint verification, it is further determined whether the trajectory simultaneously passes the lane boundary constraint verification and the traffic light constraint verification. If so, the trajectory is marked as a verified trajectory; otherwise, the trajectory is marked as an unusable trajectory. If multiple verified trajectories exist, a target trajectory is selected from each of the verified trajectories, and the vehicle is controlled to drive according to the target trajectory; if no verified trajectory exists, a reversal signal is generated to indicate that the vehicle needs to perform a hazard avoidance operation. The collision time constraint check is used to verify whether the vehicle has a collision risk; the lane boundary constraint check is used to verify whether the vehicle has a risk of going out of the lane boundary; and the traffic light constraint check is used to verify whether the vehicle has a risk of crossing the stop line when the traffic light is red.

2. The vehicle control method based on autonomous driving trajectory verification according to claim 1, characterized in that, If the verified trajectory does not exist, generating a reversal signal to indicate that the vehicle needs to perform a hazard avoidance maneuver includes: If the verified trajectory does not exist, determine whether there is an unusable trajectory that failed the collision time constraint verification; If so, a first retreat signal is generated; wherein the first retreat signal is used to indicate that the vehicle needs to perform a minimum-risk maneuver to avoid danger; If not, a second backoff signal is generated; wherein the second backoff signal is used to indicate that the vehicle needs to perform a trajectory correction avoidance operation.

3. The vehicle control method based on autonomous driving trajectory verification according to claim 2, characterized in that, After generating the first backoff signal, the method further includes: When it is determined that the first rollback signal has been generated in multiple consecutive time windows, the vehicle is controlled to perform a minimum-risk maneuver to decelerate to a stop.

4. The vehicle control method based on autonomous driving trajectory verification according to claim 2, characterized in that, After generating the second backoff signal, the method further includes: When it is determined that the second rollback signal has been generated in multiple consecutive time windows, the vehicle speed is reduced and the vehicle is kept traveling in the current lane to complete the trajectory correction operation.

5. The vehicle control method based on autonomous driving trajectory verification according to claim 1, characterized in that, The collision time constraint verification of the trajectory to be verified includes: Based on the vehicle's internal operating data and external road condition data, the collision time of the vehicle with other vehicles within a preset time window is calculated when the vehicle travels along the trajectory to be verified. When it is determined that the collision time is less than the first preset collision time threshold, the verification result that the trajectory to be verified has failed the collision time constraint verification is output. When it is determined that the collision time is not less than the first preset collision time threshold, the verification result of the trajectory to be verified passing the collision time constraint is output.

6. The vehicle control method based on autonomous driving trajectory verification according to claim 5, characterized in that, The step of determining whether the trajectory to be verified has passed both lane boundary constraint verification and traffic light constraint verification simultaneously, and if so, marking the trajectory to be verified as a verified trajectory; otherwise, marking the trajectory to be verified as an unusable trajectory, includes: Based on each trajectory point in the trajectory to be verified and the road condition data, lane boundary constraint verification is performed on the trajectory to be verified. Based on each trajectory point in the trajectory to be verified, the running data, and the traffic data, the trajectory to be verified is subjected to traffic light constraint verification. If the trajectory to be verified passes both the lane boundary constraint verification and the traffic light constraint verification, the trajectory to be verified is marked as the verified trajectory. If the trajectory to be verified fails the lane boundary constraint verification or the traffic light constraint verification, the trajectory to be verified will be marked as the unusable trajectory.

7. The vehicle control method based on autonomous driving trajectory verification according to claim 6, characterized in that, The step of performing lane boundary constraint verification on the trajectory to be verified based on each trajectory point in the trajectory to be verified and the road condition data includes: Extract the left lane boundary and right lane boundary from the road condition data; Calculate the lateral distance from each trajectory point to the left lane boundary and the right lane boundary, and select the lateral distance with the smallest value as the minimum boundary margin of the trajectory to be verified. When it is determined that the minimum boundary margin is less than the preset boundary margin threshold, the verification result that the trajectory to be verified has failed the lane boundary constraint verification is output. When it is determined that the minimum boundary margin is not less than the preset boundary margin threshold, the verification result of the trajectory to be verified passing the lane boundary constraint verification is output.

8. The vehicle control method based on autonomous driving trajectory verification according to claim 7, characterized in that, The step of performing traffic light constraint verification on the trajectory to be verified based on each trajectory point in the trajectory to be verified, the running data, and the traffic data includes: Extract the stop line boundaries and traffic light status information from the road condition data; Based on the running data and each of the trajectory points, the running speed corresponding to each trajectory point is extracted; When it is determined that the red light is in a valid state based on the traffic signal light status information, the valid red light period corresponding to the red light status is obtained; Extract each trajectory point to be verified within the red light's effective time period, and the corresponding running speed to be verified for each trajectory point to be verified; Calculate the longitudinal distance from each of the trajectory points to be verified to the boundary of the stop line; Determine whether there are any trajectory points to be verified whose longitudinal distance is less than a preset parking distance threshold; If so, output the verification result that the trajectory to be verified has failed the traffic light constraint verification; If not, calculate the stop line arrival time corresponding to each of the longitudinal distances and the running speeds to be verified; if it is determined that there is a stop line arrival time less than a preset parking time threshold, output the verification result that the trajectory to be verified has failed the traffic light constraint verification; if it is determined that there is no stop line arrival time less than the preset parking time threshold, output the verification result that the trajectory to be verified has passed the traffic light constraint verification. The stop line arrival time is used to characterize the time when the front reference position of the vehicle reaches the boundary of the stop line.

9. The vehicle control method based on autonomous driving trajectory verification according to claim 8, characterized in that, Selecting a target trajectory from each of the verified trajectories includes: For each of the verified trajectories, the collision violation component corresponding to the collision time, the lane boundary violation component corresponding to the minimum boundary margin, and the traffic light violation component corresponding to the longitudinal distance are combined to obtain the target violation vector corresponding to the verified trajectory. For each verified trajectory, a consistency score is calculated by weighting and summing the target violation vector, the first weight corresponding to the collision time constraint verification, the second weight corresponding to the lane boundary constraint verification, and the third weight corresponding to the traffic light constraint verification; wherein, the first weight is greater than the second weight, and the second weight is greater than the third weight. The verified trajectory with the highest consistency score is taken as the target trajectory; wherein the consistency score is used to indicate the degree of compliance with collision time constraints, lane boundary constraints and traffic light constraints.

10. The vehicle control method based on autonomous driving trajectory verification according to claim 9, characterized in that, The generation of the target violation vector includes: The collision violation component is constructed based on the collision time, the first preset collision time threshold, and the collision time confidence level; wherein, the collision time confidence level is used to indicate the reliability of the collision time calculation result; The lane boundary violation component is constructed based on the minimum boundary margin, the preset boundary margin threshold, and the lane boundary confidence; wherein, the lane boundary confidence is used to indicate the reliability of the lateral distance calculation results. The traffic light violation component is constructed based on the stop line arrival time, the preset parking time threshold, the red light valid period, and the parking distance confidence level; wherein, the parking distance confidence level is used to indicate the reliability of the longitudinal distance calculation result; The collision violation component, the lane boundary violation component, and the traffic light violation component are combined to output the target violation vector.

11. A vehicle control device based on autonomous driving trajectory verification, characterized in that, The device includes: The trajectory acquisition module is used to acquire several trajectories to be verified predicted by the autonomous driving strategy within the current time window when the vehicle is in autonomous driving mode. The trajectory verification module is used to perform collision time constraint verification on each trajectory to be verified; if the trajectory to be verified fails the collision time constraint verification, the trajectory to be verified is marked as an unusable trajectory; if the trajectory to be verified passes the collision time constraint verification, it is further determined whether the trajectory to be verified passes both lane boundary constraint verification and traffic light constraint verification at the same time. If so, the trajectory to be verified is marked as a verified trajectory; otherwise, the trajectory to be verified is marked as an unusable trajectory. The control module is configured to select a target trajectory from the multiple verified trajectories if such trajectories exist, and control the vehicle to drive according to the target trajectory; if no verified trajectory exists, generate a reversal signal to indicate that the vehicle needs to perform a hazard avoidance operation. The collision time constraint check is used to verify whether the vehicle has a collision risk; the lane boundary constraint check is used to verify whether the vehicle has a risk of going out of the lane boundary; and the traffic light constraint check is used to verify whether the vehicle has a risk of crossing the stop line when the traffic light is red.

12. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the vehicle control method based on autonomous driving trajectory verification as described in any one of claims 1 to 7.

13. A vehicle, characterized in that, The vehicle includes a dynamic display device for an in-vehicle desktop as described in claim 11 or an electronic device as described in claim 12.

Images