A vehicle control method and vehicle

Abstract

This application relates to the field of vehicle control technology, and in particular to a vehicle control method and a vehicle. The control method includes: acquiring vehicle status information and vehicle obstacle information in real time; using an intent recognition model to identify the vehicle status information and vehicle obstacle information to determine whether the vehicle is in a misoperation scenario, wherein the intent recognition model performs parallel judgment based on a rule engine and a gradient boosting algorithm, and combines the judgment results of both to determine whether the vehicle is in a misoperation scenario; if so, multi-level intervention operations are executed. By combining a rule engine and a gradient boosting algorithm for misoperation identification in this application, the duration of false alarms is reduced, unnecessary repeated prompts are decreased, and the intelligence level of the warning strategy is improved.

Description

Technical Field

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

[0002] In actual vehicle use, especially in scenarios such as low-speed starts, reversing, parking maneuvers, and hill starts, drivers may still accidentally shift gears due to factors such as distraction, fatigue, unfamiliarity with gear shifting logic, or nervousness. For example, if a driver accidentally shifts to reverse while the vehicle is stationary and then presses the accelerator, the vehicle may move in an unexpected direction. Similarly, if a driver accidentally shifts to neutral or park while starting on a hill, the vehicle may roll backward or fail to start properly, thus posing a safety hazard.

[0003] Existing gear position indication solutions mostly rely on gear position sensor signals or make judgments based on preset fixed logic. They can provide basic indications of the current gear status, but they usually lack the ability to comprehensively analyze whether the driver's operating intention matches the vehicle's operating state. For example, when the vehicle is stationary and the accelerator pedal opening is rapidly increased while the current gear is inconsistent with the driver's normal starting intention, existing systems often can only output general prompts, making it difficult to identify potentially dangerous operations in a timely manner or to implement more targeted warnings or interventions.

[0004] Furthermore, existing technologies typically do not fully integrate multi-dimensional data such as accelerator pedal opening, vehicle speed, engine speed, braking status, and information about the vehicle's surrounding environment, lacking a joint judgment mechanism for driving behavior, vehicle status, and environmental conditions. Especially in complex conditions such as the presence of obstacles in front and behind, slopes, and narrow passages, single-signal-source alerts are prone to high false alarm rates, delayed response, and insufficient targeting. For drivers whose attention has decreased or who fail to receive timely audible cues, relying solely on audible alerts is also insufficient to effectively prevent collisions or rollover risks caused by accidentally shifting gears. Summary of the Invention

[0005] This application addresses, to at least some extent, one of the technical problems in the related art.

[0006] In a first aspect, embodiments of this application provide a vehicle control method, the method comprising the following steps: Real-time acquisition of vehicle status information and vehicle obstacle information, wherein the vehicle status information includes vehicle gear, accelerator pedal opening, vehicle speed, and environmental slope, and the vehicle obstacle information includes obstacle distance; The vehicle status information and vehicle obstacle information are identified using an intent recognition model to determine whether the vehicle is in a misoperation scenario. The intent recognition model performs parallel judgment based on a rule engine and a gradient boosting algorithm, and combines the judgment results of the two to determine whether the vehicle is in a misoperation scenario. If so, multi-level intervention operations are performed, and the multi-level intervention operations provide early warning through information prompts at least.

[0007] In the above embodiments, the control method is applied to the vehicle controller or driver assistance controller to identify unexpected operating behaviors that may occur when the driver starts, moves the vehicle at low speed, travels on a slope, parks in front or behind, or is near an obstacle, and to intervene in a graded manner before or in the early stage of risk formation.

[0008] Specifically, vehicle status information and vehicle obstacle information are collected in real time. After time synchronization, outlier removal and unit standardization, the above data is input into the intent recognition model to ensure that data from different sources are comparable within the same judgment period.

[0009] The intent recognition model employs a parallel approach of a rule engine and a gradient boosting algorithm to identify erroneous operation scenarios. The rule engine handles high-risk patterns with strong determinism, such as when the vehicle is in drive or reverse, the accelerator pedal opening is abnormally large, the vehicle speed is low, the distance to obstacles in front of or behind the vehicle is less than a preset threshold, and the environmental slope poses a risk of the vehicle rolling or lurching forward. The rule engine can set different trigger conditions based on different gear positions, slope ranges, speed ranges, and obstacle distance ranges to achieve fast and interpretable prior judgments.

[0010] Meanwhile, the XGBoost algorithm (eXtreme Gradient Boosting, or XGBoost for short) jointly models vehicle gear position, accelerator pedal opening, vehicle speed, environmental slope, obstacle distance and their changing trends to uncover the nonlinear coupling relationship between multiple variables, thereby identifying complex misoperation scenarios that are difficult to cover by relying solely on fixed thresholds, such as situations where the driver mistakenly applies acceleration intent when near an obstacle, or accidentally presses the accelerator when starting on a slope, causing the vehicle's longitudinal output to exceed the safe range.

[0011] After combining the results of both assessments to determine whether the vehicle is in a misoperation scenario, multi-level intervention operations are executed. The first level of multi-level intervention includes at least an information warning, which can be achieved through one or more of the following methods: instrument panel text prompts, buzzer alarms, voice reminders, head-up display, or central control interface prompts, to inform the driver that the current operation poses a collision risk or a risk of misoperation. As the risk level increases, the multi-level intervention operations can be further expanded to limit power output, suppress torque request, mitigate changes in vehicle acceleration, or trigger braking assistance, in order to reduce the probability of a collision while ensuring driving smoothness.

[0012] In some embodiments, the rule engine is configured with a mapping relationship between vehicle status information and vehicle obstacle information and misoperation scenarios; The vehicle state information and vehicle obstacle information are identified using an intent recognition model, including: Based on the mapping relationship, it is determined whether the current vehicle status information and vehicle obstacle information match the misoperation scenario, thereby determining whether the vehicle is in a misoperation scenario and obtaining the rule judgment result. The mapping relationship can be understood as a set of rules used to represent the correspondence between "input status - risk scenario - judgment result".

[0013] In the above implementation, the process of determining whether the current vehicle is in a misoperation scenario based on the mapping relationship may include: acquiring the vehicle status information and vehicle obstacle information at the current moment; inputting the current information into the rule engine; having the rule engine call the corresponding mapping relationship item to match the current input parameters; and outputting the rule judgment result when the match is successful. The rule judgment result can be a binary result or a multi-level risk result. For example, the rule judgment result may include "non-misoperation scenario," "suspected misoperation scenario," and "confirmed misoperation scenario"; or it may include multiple risk levels such as low risk, medium risk, and high risk.

[0014] To improve stability, the rule engine can also introduce duration constraints, rate of change constraints, and multi-cycle consistency judgments. For example, the rule judgment result confirming the erroneous operation scenario will only be output when a certain state combination satisfies the mapping condition for multiple consecutive control cycles, thereby reducing the probability of misjudgment caused by instantaneous sensor fluctuations or short-term normal driving correction actions.

[0015] In some embodiments, the identification rules can be stored in the ECU (Electronic Control Unit) as parameter tables, logic tables, state machines, or configuration files. The ECU calls the rule engine to determine the current input state according to a predetermined refresh cycle. Different rules can be executed independently or have priorities set.

[0016] In some embodiments, the gradient boosting algorithm is obtained by pre-training based on vehicle state information and vehicle obstacle information under a preset misoperation scenario; The method of using an intent recognition model to identify the vehicle state information and vehicle obstacle information also includes: Based on the gradient boosting algorithm, the current vehicle status information and vehicle obstacle information are identified to determine whether the vehicle is in a misoperation scenario and output the enhanced judgment confidence.

[0017] In the above embodiments, the gradient boosting algorithm is based on pre-collected sample data related to low-speed vehicle operation. This sample data includes at least vehicle state information and obstacle information under preset misoperation scenarios. The gradient boosting algorithm establishes a nonlinear mapping relationship between the vehicle input state and the misoperation scenario through offline training on the training sample set. During online recognition, the vehicle controller acquires the vehicle state information and obstacle information within the current period in real time and preprocesses the input data. The model outputs the probability value of the current vehicle being in a misoperation scenario based on the current input features. This output result can be used as an enhanced judgment confidence score. The enhanced judgment confidence score characterizes the credibility of the current input state being identified as a misoperation scenario; the higher the value, the greater the probability that the current vehicle is in a misoperation scenario.

[0018] In the above implementation, by pre-training on vehicle state information and obstacle information under preset misoperation scenarios, the gradient boosting algorithm can learn the multi-dimensional input features and their inter-coupling relationships in the misoperation scenarios. Compared with rule-based judgment methods that rely solely on manually set thresholds, this approach can better identify misoperation behaviors with ambiguous boundaries, complex feature coupling, or continuous scene evolution, thereby improving the accuracy of misoperation scenario identification. The gradient boosting algorithm has a strong modeling ability for nonlinear relationships and multi-feature combination relationships, and can simultaneously consider the combined influence of multiple factors such as gear position, pedal opening, speed, slope, and obstacle distance. This method helps identify scenarios where significant risks only manifest under specific feature combinations.

[0019] In some embodiments, determining whether the vehicle is in a misoperation scenario includes: If the rule judgment result is yes, and the enhanced judgment confidence exceeds the confidence threshold, then the vehicle is judged to be in a misoperation scenario, and the triggering event is recorded. When the enhanced judgment confidence is higher than the confidence threshold, it indicates that the current vehicle status information and obstacle information have a high similarity to known misoperation scenarios in the multi-dimensional feature space.

[0020] The above steps, by requiring a "yes" rule judgment result and an enhanced judgment confidence level exceeding a confidence threshold before confirming the erroneous operation scenario, effectively reduce the risk of misjudgment arising from relying solely on a single rule or model judgment. Compared to methods that use rule triggering as the final judgment criterion, this approach utilizes model confidence to reconfirm the rule result, thereby reducing false alarms caused by instantaneous sensor fluctuations, overly lenient rule boundary settings, or short-term anomalies in individual parameters. By combining the rule judgment result with the confidence threshold, the vehicle can improve the stability of the final judgment by leveraging the model's ability to supplement complex boundary scenarios while ensuring rule interpretability. This helps to avoid unnecessary frequent interventions while maintaining safety, increasing driver acceptance of the vehicle.

[0021] In some embodiments, the rule engine determines whether the vehicle status information and vehicle obstacle information correspond to an erroneous operation scenario by looking up a table. If so, it determines that the vehicle is in an erroneous operation scenario.

[0022] In the above implementation, by using a lookup table to determine whether vehicle status information and vehicle obstacle information correspond to a misoperation scenario, the online judgment process of the rule engine can be significantly simplified. Compared with the method of performing complex condition judgments one by one, the lookup table method directly solidifies the pre-calibrated rule results into table item mapping relationships. In the real-time operation phase, only index retrieval is required, thus having a high response speed and being suitable for deployment in latency-sensitive vehicle control systems.

[0023] The lookup table method offers strong engineering feasibility and maintainability. The identification conditions for different erroneous operation scenarios can be pre-organized into structured table entries, facilitating parameter calibration, version management, and scenario expansion by developers. For different vehicle models, powertrain configurations, or regional driving habits, only the index ranges and scenario correspondences in the lookup table need to be adjusted to achieve identification strategy adaptation, without modifying the underlying decision-making framework.

[0024] In some embodiments, the execution of multi-level intervention operations includes: At least a Level 1 warning should be issued via voice prompts and / or icon prompts, and After setting the adjustment time, the intent recognition model is invoked to identify the current vehicle status information and vehicle obstacle information, and to determine again whether the vehicle is in a misoperation scenario. If the vehicle is again determined to be in a misoperation scenario, a secondary warning will be issued at least through voice prompts and / or a buzzer alarm, and the triggering event will be recorded again; If it is determined again that the vehicle is not in a misoperation scenario, then it is determined that the driver has performed corrective action, and no further prompting is required.

[0025] Based on the above steps, when the intent recognition model initially determines that the vehicle is in a misoperation scenario, the controller does not immediately initiate high-intensity intervention, but instead executes a Level 1 warning. The purpose of the Level 1 warning is to promptly alert the driver after initial risk confirmation, giving them an opportunity to proactively correct their current operation before the system escalates its intervention. After executing the Level 1 warning, the intent recognition model is repeatedly used to re-identify the latest vehicle status and obstacle information. This re-identification can be performed within the continuous control cycle following the Level 1 warning trigger to dynamically confirm whether the driver has performed corrective action. If the re-assessment indicates that the risk has been eliminated, it means the driver has performed corrective action, and no further prompts are needed, thus avoiding unnecessary repeated warnings. If, after a set adjustment period, the vehicle is still determined to be in a misoperation scenario, it indicates that the Level 1 warning failed to prompt the driver to complete effective correction, or the current risk scenario continues or even worsens. In this case, a Level 2 warning is executed.

[0026] In the above embodiments, by setting up a multi-level intervention process of primary and secondary warnings, the system can adopt different intensities of alerts at different risk stages. Compared with triggering a high-intensity alarm immediately after the first identification, this application first provides a primary warning through voice prompts and / or icon prompts, which can reduce abrupt interference to drivers and improve the gentleness and acceptability of the system prompts.

[0027] By repeatedly using the intent recognition model for reassessment after a Level 1 warning, the driver's actual corrective actions can be incorporated into the intervention logic. This application does not simply rely on the initial recognition result for continuous alarms, but rather dynamically updates the alert based on whether the driver has released the accelerator, applied the brakes, adjusted the gear, or eliminated the obstacle risk. This approach helps reduce the duration of false alarms, minimizes unnecessary repetitive prompts, and improves the intelligence level of the warning strategy.

[0028] In some embodiments, if the vehicle is determined to be in a misoperation scenario again after the set adjustment time, and the vehicle status information or vehicle obstacle information meets the braking conditions, the vehicle automatically applies braking control and performs secondary intervention.

[0029] Specifically, the braking conditions are used to characterize that the current risk has reached a level that requires suppression through active longitudinal control of the vehicle. The braking conditions include: the obstacle information of the vehicle body is less than the emergency braking distance, or the accelerator pedal opening or vehicle speed in the vehicle status information shows an upward trend. Among them, the obstacle information of the vehicle body is less than the emergency braking distance, which means that the remaining safety space between the vehicle and the obstacle is insufficient, and there is a risk of collision if the driver makes subsequent corrections alone. When the accelerator pedal opening in the vehicle status information shows an upward trend, it indicates that the driver may be continuously increasing the drive request, and the risk of misoperation is likely to further increase. When the vehicle speed shows an upward trend, it indicates that the vehicle has shown a continuous acceleration or increasing slippage trend. If intervention is not timely, it may evolve into a risk event such as rear-end collision, forward collision, or rollback.

[0030] Compared to solutions that only alert drivers via voice or beeping, this method intervenes directly in the vehicle's longitudinal movement when the driver fails to correct their actions in time, thereby reducing the probability of collisions or skidding accidents. By setting the braking conditions to be either an obstacle distance less than the emergency braking distance, an increasing trend in accelerator pedal opening, or an increasing trend in vehicle speed, both spatial and kinematic risks can be considered as trigger criteria. The former reflects the remaining safety margin between the vehicle and the obstacle, while the latter reflects whether the risk is still escalating. This method makes the triggering of secondary intervention more consistent with the risk evolution patterns in actual driving, helping to improve the accuracy of active braking triggering.

[0031] By comprehensively judging the obstacle distance conditions and vehicle dynamic trend conditions, the vehicle automatically applies braking control. The applied braking control is configured to be a slow braking of 0.2 to 0.5 seconds with a force of less than 0.3g, in order to prevent the vehicle from rushing backward or rolling backward. This parameter configuration makes the braking intervention have a certain suppression effect, while avoiding excessively fast braking force to prevent obvious discomfort to the driver or sudden changes in vehicle posture.

[0032] In some embodiments, the method further includes: Obtain the vehicle status information and vehicle obstacle information corresponding to the triggering event, and add a label for the triggering event indicating whether it is a valid alarm or a false alarm; Triggering events marked as valid alarms are used as positive samples, and triggering events marked as false alarms are used as negative samples to build a historical operation database; The confidence threshold is adjusted based on the historical operation database.

[0033] In the above implementation, the triggered events are labeled as either valid alarms or false alarms. A valid alarm refers to an alarm that, after triggering a warning or intervention, is confirmed through manual review, event playback, vehicle log analysis, or user feedback to correspond to a genuine misoperation risk scenario. A false alarm refers to an alarm or intervention that, although triggered by the system, is subsequently confirmed by analysis to not correspond to a genuine misoperation risk, or to be a case of normal driving behavior being misidentified. The labeling process can be completed offline by developers during the testing phase, or it can be completed by an after-sales diagnostic system, remote maintenance platform, or cloud data platform after analyzing event records, video data, perception data, and driver operation trajectories. Through the above labeling, the system can establish a clear correspondence between each triggered event and its true nature.

[0034] In the above implementation, by establishing a historical operation database, a data accumulation mechanism oriented towards real vehicle operation scenarios can be formed. This approach can not only cover experimental scenarios but also gradually incorporate actual event data of mass-produced vehicles under different road environments, driving habits in different regions, and vehicle configurations, thereby making system optimization closer to real application scenarios. Adjusting the confidence threshold based on the historical operation database enables the confidence threshold to have data-driven adaptive optimization capabilities. Compared to a threshold scheme that remains fixed after initial calibration, this approach can dynamically adjust the threshold setting according to false alarms and missed alarms during operation, which is beneficial for balancing recognition sensitivity and false alarm control requirements at different stages and improving the overall performance of misoperation scenario judgment.

[0035] In some embodiments, adjusting the confidence threshold based on the historical operation library includes: The learning module is used to extract features from the vehicle state information and vehicle obstacle information of the negative and positive samples to determine the offset. The offset is then summed with the confidence threshold to obtain the adjusted confidence threshold.

[0036] In the above embodiments, the learning module can be deployed in the vehicle's local controller, domain controller, or cloud training platform to adaptively adjust the current confidence threshold based on positive and negative samples in the historical operation library. By summing the offset with the original confidence threshold to obtain the adjusted confidence threshold, adaptive optimization of the existing recognition framework can be achieved with low engineering complexity. This approach does not require a complete replacement of the original intent recognition model; it only requires differentiated adjustment of the decision threshold to improve recognition performance in different scenarios, facilitating rapid deployment in existing vehicle control systems.

[0037] Specifically, the learning module acquires labeled trigger event data from the historical operation database, extracts features from vehicle state information and vehicle obstacle information in positive and negative samples to obtain feature combinations that describe scene features and driving behavior features. Based on the distribution differences between positive and negative samples under the same or similar feature combinations, the corresponding offset is determined.

[0038] The offset is used to characterize whether the current default confidence threshold should be adjusted towards a more sensitive or more robust direction in a specific scenario. Specifically, if the learning module analysis finds that, under a certain feature combination, the proportion of negative samples is high, meaning that false alarms are frequently triggered in this type of scenario, it indicates that the current default confidence threshold is too low, causing the warning triggering condition to be overly sensitive. In this case, the learning module can generate a positive offset, so that the offset, when summed with the original confidence threshold, yields a higher adjusted confidence threshold, thereby increasing the subsequent triggering threshold. Conversely, if the learning module analysis finds that, under a certain feature combination, the proportion of positive samples is high, and there are multiple real erroneous operation risk events, it indicates that the current default confidence threshold is too high, causing the system's sensitivity to identify this type of risk scenario to be insufficient. In this case, the learning module can generate a negative offset, so that the offset, when summed with the original confidence threshold, yields a lower adjusted confidence threshold, thereby improving the warning sensitivity for this type of scenario.

[0039] Secondly, embodiments of this application provide a vehicle including a controller, the controller being configured to perform the control method described in the first aspect above.

[0040] As can be seen from the above technical solutions, additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the control method according to an embodiment of this application; Figure 2 This is a step-by-step flowchart of the control method according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the step-by-step principle of the control method according to an embodiment of this application; Figure 4 This is a schematic flowchart of another step of the control method according to an embodiment of this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0043] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0044] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0045] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0046] This application provides a vehicle control method. Figure 1 This is a flowchart of a control method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps: S100: Real-time acquisition of vehicle status information and vehicle obstacle information, wherein the vehicle status information includes vehicle gear position, accelerator pedal opening, vehicle speed, and environmental slope, and the vehicle obstacle information includes obstacle distance; the vehicle obstacle information can be obtained by fusing any one or more sensors among ultrasonic radar, millimeter-wave radar, camera, and lidar, and the obstacle distance is preferably the nearest obstacle distance in the current driving direction of the vehicle.

[0047] S200: The intent recognition model is used to identify the vehicle status information and vehicle obstacle information to determine whether the vehicle is in a misoperation scenario. The intent recognition model is based on a rule engine and a gradient boosting algorithm to make parallel judgments and combines the judgment results of the two to determine whether the vehicle is in a misoperation scenario. The gradient boosting algorithm can be the XGBoost algorithm. S300: If so, then perform multi-level intervention operations, which shall at least provide early warning through information prompts.

[0048] In the above embodiments, the control method is applied to the vehicle controller or driver assistance controller to identify unexpected operating behaviors that may occur when the driver starts, moves the vehicle at low speed, travels on a slope, parks in front or behind, or is near an obstacle, and to intervene in a graded manner before or in the early stage of risk formation.

[0049] Regarding obstacle distance, for example, but not limited to: when the vehicle is in drive, the obstacle distance can be defined as the distance to the nearest obstacle in front of the vehicle; when the vehicle is in reverse, the obstacle distance can be defined as the distance to the nearest obstacle in the rearward direction or on the predicted rearward trajectory. This allows obstacle distance to more accurately characterize the collision risk associated with the vehicle's current or potential direction of motion.

[0050] Specifically, vehicle status information and vehicle obstacle information are collected in real time. After time synchronization, outlier removal and unit standardization, the above data is input into the intent recognition model to ensure that data from different sources are comparable within the same judgment period.

[0051] The intent recognition model employs a parallel approach of a rule engine and a gradient boosting algorithm to identify erroneous operation scenarios. The rule engine handles high-risk patterns with strong determinism, such as when the vehicle is in drive or reverse, the accelerator pedal opening is abnormally large, the vehicle speed is low, the distance to obstacles in front of or behind the vehicle is less than a preset threshold, and the environmental slope poses a risk of the vehicle rolling or lurching forward. The rule engine can set different trigger conditions based on different gear positions, slope ranges, speed ranges, and obstacle distance ranges to achieve fast and interpretable prior judgments.

[0052] Meanwhile, the XGBoost algorithm jointly models vehicle gear position, accelerator pedal opening, vehicle speed, environmental slope, obstacle distance and their changing trends to uncover the nonlinear coupling relationship between multiple variables, thereby identifying complex misoperation scenarios that are difficult to cover by relying solely on fixed thresholds, such as situations where the driver mistakenly applies acceleration intent when near an obstacle, or accidentally presses the accelerator when starting on a slope, causing the vehicle's longitudinal output to exceed the safe range.

[0053] After combining the results of both assessments to determine whether the vehicle is in a misoperation scenario, multi-level intervention operations are executed. The first level of multi-level intervention includes at least an information warning, which can be achieved through one or more of the following methods: instrument panel text prompts, buzzer alarms, voice reminders, head-up display, or central control interface prompts, to inform the driver that the current operation poses a collision risk or a risk of misoperation. As the risk level increases, the multi-level intervention operations can be further expanded to limit power output, suppress torque request, mitigate changes in vehicle acceleration, or trigger braking assistance, in order to reduce the probability of a collision while ensuring driving smoothness.

[0054] In the above implementation, by simultaneously introducing a rule engine and the XGBoost algorithm for parallel identification of erroneous operation scenarios, the determinism and real-time nature of rule-based judgments, as well as the non-linear feature learning capabilities of model-based judgments, can be balanced. Compared to schemes that only use fixed threshold judgments, this scheme can more accurately distinguish between normal low-speed driving behavior and potential erroneous operation behavior, reducing the false alarm rate caused by a single rule trigger. Compared to schemes that only use data-driven models, this scheme can also utilize the rule engine to quickly respond to typical high-risk scenarios, improving the interpretability and stability of the vehicle in safety-related scenarios.

[0055] Optionally, the vehicle is pre-deployed with a multi-source sensor fusion module. Among them, the vehicle gear position can be obtained based on the gear position sensor or the gear shift actuator, the accelerator pedal opening can be obtained based on the high-precision accelerator pedal opening sensor, the vehicle speed can be calculated based on the wheel speed sensor, the anti-lock braking system or the prediction module, the environmental slope can be obtained based on the inertial measurement unit, the vehicle body attitude sensor or based on longitudinal acceleration and attitude angle estimation, and the obstacle distance can be obtained based on millimeter-wave radar or wide-angle camera, etc.

[0056] In another embodiment, the vehicle status information collected is not limited to vehicle gear position, accelerator pedal opening, vehicle speed, and environmental slope. It can also include, depending on the actual application needs, information such as brake pedal status, steering wheel angle, wheel speed difference, drive motor torque request, brake master cylinder pressure, parking status, door status, or seatbelt status to further improve the completeness of misoperation scenario recognition. In addition to obstacle distance, vehicle obstacle information can also include obstacle orientation, relative speed, obstacle size information, obstacle category information, or obstacle distance change rate to enhance adaptability to dynamic obstacles and complex environments.

[0057] In the above implementation, the gradient boosting algorithm is not limited to the XGBoost algorithm; other implementations related to gradient boosting decision trees can also be used, as long as they can classify the input features or output risk probabilities. When using the XGBoost algorithm, in addition to the original collected data, the input features can also include statistical features extracted based on time windows, rate of change features, combined features, or normalized features to improve the model's ability to represent erroneous behavior patterns. The model can be trained using historical driving data, test scenario data, and labeled erroneous event samples, and its parameters can be updated according to vehicle software iterations.

[0058] In some embodiments, the rule engine is configured with a mapping relationship between vehicle status information and vehicle obstacle information and misoperation scenarios; The vehicle state information and vehicle obstacle information are identified using an intent recognition model, including: S210: Based on the mapping relationship, determine whether the current vehicle status information and vehicle obstacle information match the misoperation scenario, thereby determining whether the vehicle is in a misoperation scenario and obtaining the rule judgment result. The mapping relationship can be understood as a set of rules used to represent the correspondence between "input status - risk scenario - judgment result".

[0059] In the above implementation, by pre-configuring the mapping relationship between vehicle status information, vehicle obstacle information, and misoperation scenarios in the rule engine, the identification logic of typical misoperation scenarios can be expressed in a structured way, thereby achieving rapid judgment of high-risk conditions. Compared with solutions that rely entirely on data-driven models, this approach can shorten the judgment chain and improve response timeliness in critical risk scenarios, and is especially suitable for safety control with high real-time requirements in low-speed near-obstacle scenarios.

[0060] In the above implementation, the process of determining whether the current vehicle is in a misoperation scenario based on the mapping relationship may include: acquiring the vehicle status information and vehicle obstacle information at the current moment; inputting the current information into the rule engine; having the rule engine call the corresponding mapping relationship item to match the current input parameters; and outputting the rule judgment result when the match is successful. The rule judgment result can be a binary result or a multi-level risk result. For example, the rule judgment result may include "non-misoperation scenario," "suspected misoperation scenario," and "confirmed misoperation scenario"; or it may include multiple risk levels such as low risk, medium risk, and high risk.

[0061] To improve stability, the rule engine can also introduce duration constraints, rate of change constraints, and multi-cycle consistency judgments. For example, the rule judgment result confirming the erroneous operation scenario will only be output when a certain state combination satisfies the mapping condition for multiple consecutive control cycles, thereby reducing the probability of misjudgment caused by instantaneous sensor fluctuations or short-term normal driving correction actions.

[0062] Specifically, the misoperation scenarios include at least: dangerous backward rushing scenario, slope slippage scenario, and identification of incorrect gear shift.

[0063] Based on the above-mentioned erroneous operation scenarios, the mapping relationship may include: The first identification rule is as follows: The vehicle's gear is reverse (R), the accelerator pedal opening is less than the first accelerator hazard threshold, the vehicle speed is less than the first speed threshold, and the distance to the rear obstacle is less than the approach threshold. These conditions are used to identify dangerous rear-end collision scenarios. In this state, even with a small accelerator pedal opening, the vehicle may still be at risk of rear-end collision due to driver error, slow and continuous acceleration, creep function combined with driving force output, or misjudgment of rear space. The first accelerator hazard threshold can be configured to 5%, 10%, or 20%, the first speed threshold can be configured to 1 km / h, and the approach threshold can be configured to 1.5 meters, focusing the identification efforts on typical near-obstacle scenarios such as low-speed maneuvering, parking, or reversing into a parking space. The second identification rule is as follows: The vehicle's gear is in neutral (N), the accelerator pedal opening is less than the second throttle danger threshold, the environmental slope is greater than the slope threshold, and the vehicle speed is less than the second speed threshold. These conditions are used to identify slope rollback scenarios. In this scenario, because the vehicle is in neutral, the drive system cannot provide active holding force. If the slope parking conditions are insufficient, the driver does not brake in time, or the slope is too steep, the vehicle is prone to unexpected slippage under gravity, thus creating a slope rollback risk. The second throttle danger threshold can be configured to 5%, 10%, or 20% to characterize the state where the driver has not formed a clear driving intention or the throttle input is insufficient. The slope threshold can be configured to 5°, and the second speed threshold can be configured to 2 km / h. The third identification rule: The vehicle status information shows the vehicle is in drive (D), the accelerator pedal opening is less than the first accelerator danger threshold, and the brake pedal opening is 0 for more than 5 minutes. This status can be used to identify situations where the driver is not sufficiently restraining the vehicle in congested traffic, waiting to start, moving slowly in narrow passages, or stopping near obstacles. Especially when the vehicle has a creep function, prolonged lack of braking may cause the vehicle to continue moving slowly forward, thereby increasing the risk of approaching obstacles or pedestrians.

[0064] In the above implementation, the mapping relationship in the rule engine is further refined into multiple executable recognition rules. Each recognition rule corresponds to a different type of misoperation scenario, used to quickly map the real-time operating status of the vehicle to a specific risk category. During each control cycle, the vehicle receives input information such as gear position, accelerator pedal opening, vehicle speed, environmental slope, brake pedal opening, and obstacle distance, and compares this input information with the preset recognition rules item by item. When a recognition rule is satisfied, the rule engine outputs the rule judgment result for the corresponding scenario, which is then called by the subsequent fusion judgment module, thereby improving the targeting of misoperation scenario classification.

[0065] In the above implementation, the first identification rule, by simultaneously considering reverse gear, low vehicle speed, low throttle opening, and distance to obstacles behind, can identify the risk of dangerous rear-end collisions in low-speed reversing scenarios near obstacles. This method not only covers situations where the driver accidentally presses the accelerator, but also situations where collisions may still occur due to small throttle input but caused by creep, slope effects, or spatial misjudgment, thereby improving safety assurance capabilities in reversing situations.

[0066] In the above implementation, the second identification rule, by introducing neutral gear status and slope conditions, can more accurately distinguish between normal low-speed operation and slope slip risk status. Compared with the method of judging slip risk solely based on vehicle speed changes, combining gear and slope information makes the rule more consistent with actual vehicle dynamics characteristics, which is conducive to timely identification of risks when the vehicle is still in the early stage of low-speed slippage, reducing the safety hazards caused by the vehicle rolling backward or forward.

[0067] In the above embodiments, the third identification rule can serve as a supplement to the short-term, sudden-type erroneous operation identification logic, used to identify low-speed risk states that are persistent and highly concealed. Especially in vehicles with creep functionality or low-speed automatic following features, prolonged periods without braking in forward gear may lead to insufficient driver awareness of the risk of the vehicle continuing to move forward. By setting a duration condition, this type of risk can be assisted in identification while ensuring robustness, reducing the probability of false alarms.

[0068] In some embodiments, the identification rules can be stored in the ECU as parameter tables, logic tables, state machines, or configuration files. The ECU calls the rule engine to determine the current input state according to a predetermined refresh cycle. Different rules can be executed independently or have priorities set.

[0069] The specific rules for identifying misoperation scenarios can be configured based on the definition of "low-speed misoperation risk scenarios" in the national automotive standard GB / T 40429-2021 "Classification of Driving Automation for Automobiles". All rules are stored in the ECU in a configurable manner and support OTA (Over-the-Air Technology) remote updates, which facilitates subsequent adaptation to different vehicle models or regional driving habits.

[0070] In the above embodiments, the first throttle danger threshold and the second throttle danger threshold are not limited to 5%, 10%, or 20%, but can be calibrated according to the vehicle power type, drive motor torque response, engine idle torque characteristics, creep function strength, and vehicle weight. For example, for pure electric vehicles with faster low-speed response, the threshold can be appropriately lowered; for models with weaker creep force, the threshold can be appropriately increased. The first speed threshold and the second speed threshold are also not limited to 1 km / h and 2 km / h, but can be selected from a suitable value within the range of 0.5 km / h to 5 km / h to match different application scenarios.

[0071] In the above embodiments, the approach threshold is not limited to 1.5 meters, and can be adjusted according to the vehicle length, the field of view of the reversing camera, the detection capability of the ultrasonic radar, and the target category. For example, in scenarios involving large vehicles, vehicles with limited visibility, or narrow parking spaces, the approach threshold can be set to any adaptive value between 1.0 meter and 2.5 meters; for dynamic obstacle scenarios, the approach threshold can also be adaptively corrected by combining relative speed or distance change rate.

[0072] In some embodiments, the gradient boosting algorithm is obtained by pre-training based on vehicle state information and vehicle obstacle information under a preset misoperation scenario; The method of using an intent recognition model to identify the vehicle state information and vehicle obstacle information also includes: S220: Based on the gradient boosting algorithm, identify the current vehicle status information and obstacles on the vehicle body, identify whether the vehicle is in a misoperation scenario, and output the enhanced judgment confidence level.

[0073] In the above embodiments, the gradient boosting algorithm is based on pre-collected sample data related to low-speed vehicle operation. This sample data includes at least vehicle state information and obstacle information under preset misoperation scenarios. The gradient boosting algorithm establishes a nonlinear mapping relationship between the vehicle input state and the misoperation scenario through offline training on the training sample set. During online recognition, the vehicle controller acquires the vehicle state information and obstacle information within the current period in real time and preprocesses the input data. The model outputs the probability value of the current vehicle being in a misoperation scenario based on the current input features. This output result can be used as an enhanced judgment confidence score. The enhanced judgment confidence score characterizes the credibility of the current input state being identified as a misoperation scenario; the higher the value, the greater the probability that the current vehicle is in a misoperation scenario.

[0074] In the above implementation, by pre-training on vehicle state information and obstacle information under preset misoperation scenarios, the gradient boosting algorithm can learn the multi-dimensional input features and their inter-coupling relationships in the misoperation scenarios. Compared with rule-based judgment methods that rely solely on manually set thresholds, this approach can better identify misoperation behaviors with ambiguous boundaries, complex feature coupling, or continuous scene evolution, thereby improving the accuracy of misoperation scenario identification. The gradient boosting algorithm has a strong modeling ability for nonlinear relationships and multi-feature combination relationships, and can simultaneously consider the combined influence of multiple factors such as gear position, pedal opening, speed, slope, and obstacle distance. This method helps identify scenarios where significant risks only manifest under specific feature combinations.

[0075] Taking the XGBoost algorithm as an example, vehicle gear position, accelerator pedal opening, vehicle speed, environmental slope, obstacle distance, and the changes of these parameters within a continuous time window can be used as feature inputs. Whether the current sample belongs to a misoperation scenario is used as the supervision label. During training, multiple weak classifiers are built through multiple iterations, and the classification error of each round is optimized using a loss function. This allows subsequent classifiers to focus on learning samples where errors occurred in the previous round, thereby improving the model's ability to distinguish complex boundary scenes. After pre-training, a gradient boosting recognition model is obtained and deployed in the vehicle controller. This model is used to infer and judge the real-time collected current vehicle state information and vehicle obstacle information during online operation.

[0076] The aforementioned preprocessing steps may include missing value handling, outlier removal, normalization, time alignment, and feature combination. For example, the rate of change of accelerator pedal opening, vehicle speed, obstacle distance, and slope over several consecutive sampling periods can be extracted and used as additional features input to the gradient boosting algorithm.

[0077] In the above implementation, the gradient boosting algorithm is pre-trained and deployed in the vehicle controller. During the online inference phase, only real-time vehicle state information and obstacle information need to be input to output the judgment result, which has good real-time performance and engineering feasibility. Compared with the computationally complex deep neural network solution, this method is easier to implement under the condition of limited vehicle controller resources, and it is also easier to manage model parameters, calibrate and optimize, and iterate versions.

[0078] In some embodiments, determining whether the vehicle is in a misoperation scenario includes: S230: If the rule judgment result is yes, and the enhanced judgment confidence exceeds the confidence threshold, then the vehicle is judged to be in a misoperation scenario, and the triggering event is recorded. Optionally, the confidence threshold is configured with an initial value, which can be 0.85. When the enhanced judgment confidence is higher than the confidence threshold, it indicates that the current vehicle state information and obstacle information have a high similarity to the known misoperation scenario in the multi-dimensional feature space.

[0079] The above steps, by requiring a "yes" rule judgment result and an enhanced judgment confidence level exceeding a confidence threshold before confirming the erroneous operation scenario, effectively reduce the risk of misjudgment arising from relying solely on a single rule or model judgment. Compared to methods that use rule triggering as the final judgment criterion, this approach utilizes model confidence to reconfirm the rule result, thereby reducing false alarms caused by instantaneous sensor fluctuations, overly lenient rule boundary settings, or short-term anomalies in individual parameters. By combining the rule judgment result with the confidence threshold, the vehicle can improve the stability of the final judgment by leveraging the model's ability to supplement complex boundary scenarios while ensuring rule interpretability. This helps to avoid unnecessary frequent interventions while maintaining safety, increasing driver acceptance of the vehicle.

[0080] In the above steps, steps S210 and S220 can be executed sequentially or in parallel.

[0081] In another embodiment, the confidence threshold can be a fixed threshold or an adaptive threshold. The fixed threshold can be set to 0.6, 0.7, 0.8, or other suitable values ​​based on test data and calibration results; the adaptive threshold can be dynamically adjusted based on vehicle gear position, obstacle distance, environmental slope, risk scenario category, or vehicle operating mode. For example, when the obstacle is very close or the slope is steep, the confidence threshold can be appropriately lowered to improve response sensitivity in high-risk scenarios; in normal low-speed traffic scenarios, the confidence threshold can be appropriately increased to reduce the probability of false alarms.

[0082] In some embodiments, the rule engine determines whether the vehicle status information and vehicle obstacle information correspond to an erroneous operation scenario by looking up a table. If so, it determines that the vehicle is in an erroneous operation scenario.

[0083] In the above implementation, by using a lookup table to determine whether vehicle status information and vehicle obstacle information correspond to a misoperation scenario, the online judgment process of the rule engine can be significantly simplified. Compared with the method of performing complex condition judgments one by one, the lookup table method directly solidifies the pre-calibrated rule results into table item mapping relationships. In the real-time operation phase, only index retrieval is required, thus having a high response speed and being suitable for deployment in latency-sensitive vehicle control systems.

[0084] The lookup table method offers strong engineering feasibility and maintainability. The identification conditions for different erroneous operation scenarios can be pre-organized into structured table entries, facilitating parameter calibration, version management, and scenario expansion by developers. For different vehicle models, powertrain configurations, or regional driving habits, only the index ranges and scenario correspondences in the lookup table need to be adjusted to achieve identification strategy adaptation, without modifying the underlying decision-making framework.

[0085] In the above implementation, the structure of the lookup table is not limited to a single static array; it can also be implemented as a two-dimensional matrix table, a multi-dimensional nested table, a hash map table, a state encoding table, or a decision index table. For scenarios with a large number of parameters, a hierarchical lookup table approach can be adopted. First, the primary scenario category is determined based on the vehicle gear and obstacle location, and then a secondary lookup table is performed based on speed, accelerator pedal opening, and slope information to reduce the size of table entries and improve storage efficiency.

[0086] In the above embodiments, the contents of the lookup table can be updated via a local calibration tool or remotely via software upgrade or OTA. Updates can include threshold ranges, status coding rules, and table entry content, as well as adding new error scenario categories or deleting inapplicable scenario categories. For different vehicle platforms, the lookup table can also be configured as a dedicated set of entries to adapt to different vehicle sizes, sensor layouts, and power response characteristics.

[0087] In some embodiments, reference Figure 3 As shown, step S300 involves performing multi-level intervention operations, including: At least a Level 1 warning should be issued via voice prompts and / or icon prompts. After a set adjustment period, the intent recognition model should be invoked to identify the current vehicle status information and vehicle obstacle information to determine again whether the vehicle is in a misoperation scenario. If the vehicle is again determined to be in a misoperation scenario, a secondary warning will be issued at least through voice prompts and / or a buzzer alarm, and the triggering event will be recorded again; If it is determined again that the vehicle is not in a misoperation scenario, then it is determined that the driver has performed corrective action, and no further prompting is required.

[0088] In the above embodiments, the buzzer alarm can be output by a vehicle speaker, a dedicated buzzer, or an alarm module. The sound frequency can be configured to 800Hz, and the sound intensity can be configured to be greater than 75dB, so that the alarm sound has sufficient recognition and penetration, thereby effectively attracting the driver's attention even when there is ambient noise in the vehicle.

[0089] In the above embodiments, the adjustment time is set to reserve a reasonable reaction window for the driver, so that the driver can complete actions such as releasing the accelerator, pressing the brake, shifting gears, correcting steering, or terminating incorrect operations after hearing the voice prompt and seeing the icon prompt.

[0090] Information prompts via voice and / or icon prompts can include displaying a flashing red gear icon in the center of the instrument panel, accompanied by a voice prompt such as "Danger! Gear incorrect, please confirm!" The content of the voice output can also be adapted to the actual application scenario, such as reminding the driver to pay attention to the distance between the vehicle and obstacles, check the gear status, check pedal operation, or be aware of the risk of skidding on a slope.

[0091] Based on the above steps, when the intent recognition model initially determines that the vehicle is in a misoperation scenario, the controller does not immediately initiate high-intensity intervention, but instead executes a Level 1 warning. The purpose of the Level 1 warning is to promptly alert the driver after initial risk confirmation, giving them an opportunity to proactively correct their current operation before the system escalates its intervention. After executing the Level 1 warning, the intent recognition model is repeatedly used to re-identify the latest vehicle status and obstacle information. This re-identification can be performed within the continuous control cycle following the Level 1 warning trigger to dynamically confirm whether the driver has performed corrective action. If the re-assessment indicates that the risk has been eliminated, it means the driver has performed corrective action, and no further prompts are needed, thus avoiding unnecessary repeated warnings. If, after a set adjustment period, the vehicle is still determined to be in a misoperation scenario, it indicates that the Level 1 warning failed to prompt the driver to complete effective correction, or the current risk scenario continues or even worsens. In this case, a Level 2 warning is executed.

[0092] In the above embodiments, by setting up a multi-level intervention process of primary and secondary warnings, the system can adopt different intensities of alerts at different risk stages. Compared with triggering a high-intensity alarm immediately after the first identification, this application first provides a primary warning through voice prompts and / or icon prompts, which can reduce abrupt interference to drivers and improve the gentleness and acceptability of the system prompts.

[0093] By repeatedly using the intent recognition model for reassessment after a Level 1 warning, the driver's actual corrective actions can be incorporated into the intervention logic. This application does not simply rely on the initial recognition result for continuous alarms, but rather dynamically updates the alert based on whether the driver has released the accelerator, applied the brakes, adjusted the gear, or eliminated the obstacle risk. This approach helps reduce the duration of false alarms, minimizes unnecessary repetitive prompts, and improves the intelligence level of the warning strategy.

[0094] In the above embodiments, the form of the Level 1 warning is not limited to voice prompts and icon prompts, but may also include one or more of the following: text prompts, steering wheel vibration prompts, seat vibration prompts, head-up display highlight prompts, or instrument panel flashing prompts. For different vehicle platforms, an appropriate Level 1 warning form can be selected based on the in-vehicle display system and human-machine interaction configuration. For vehicles in silent mode or night mode, a combination of icon prompts and low-volume voice prompts may also be preferred.

[0095] In some embodiments, if the vehicle is determined to be in a misoperation scenario again after the set adjustment time, and the vehicle status information or vehicle obstacle information meets the braking conditions, a control command is output to the braking actuator to enable the vehicle to automatically apply braking control and perform secondary intervention.

[0096] Specifically, the braking conditions are used to characterize that the current risk has reached a level that requires suppression through active longitudinal control of the vehicle. The braking conditions include: the obstacle information of the vehicle body is less than the emergency braking distance, or the accelerator pedal opening or vehicle speed in the vehicle status information shows an upward trend. Among them, the obstacle information of the vehicle body is less than the emergency braking distance, which means that the remaining safety space between the vehicle and the obstacle is insufficient, and there is a risk of collision if the driver makes subsequent corrections alone. When the accelerator pedal opening in the vehicle status information shows an upward trend, it indicates that the driver may be continuously increasing the drive request, and the risk of misoperation is likely to further increase. When the vehicle speed shows an upward trend, it indicates that the vehicle has shown a continuous acceleration or increasing slippage trend. If intervention is not timely, it may evolve into a risk event such as rear-end collision, forward collision, or rollback.

[0097] Compared to solutions that only alert drivers via voice or beeping, this method intervenes directly in the vehicle's longitudinal movement when the driver fails to correct their actions in time, thereby reducing the probability of collisions or skidding accidents. By setting the braking conditions to be either an obstacle distance less than the emergency braking distance, an increasing trend in accelerator pedal opening, or an increasing trend in vehicle speed, both spatial and kinematic risks can be considered as trigger criteria. The former reflects the remaining safety margin between the vehicle and the obstacle, while the latter reflects whether the risk is still escalating. This method makes the triggering of secondary intervention more consistent with the risk evolution patterns in actual driving, helping to improve the accuracy of active braking triggering.

[0098] By comprehensively assessing the obstacle distance and vehicle dynamics, the vehicle automatically applies braking control. This braking control is configured as a gentle braking force of less than 0.3g for 0.2 to 0.5 seconds to prevent backward acceleration or rollback. This parameter configuration ensures that the braking intervention has a certain suppressive effect while avoiding excessively rapid braking force build-up that could cause significant discomfort to the driver or sudden changes in vehicle attitude. Specifically, a longitudinal deceleration of less than 0.3g keeps the active braking within a relatively gentle range, suitable for low-speed misoperation protection scenarios. A duration of 0.2 to 0.5 seconds allows for the rapid application of a short braking pulse when the risk initially escalates, suppressing further movement towards the obstacle or inhibiting slope slippage. In dangerous backward acceleration scenarios, this gentle braking limits the vehicle's further approach to the obstacle; in slope slippage scenarios, it establishes longitudinal restraint force in the early stages of vehicle slippage, thereby improving vehicle stability.

[0099] In the above embodiments, the emergency braking distance can be determined based on the vehicle's current speed, environmental slope, braking system response time, road surface adhesion conditions, and the relative position of the obstacle. For low-speed, near-obstacle scenarios, the emergency braking distance can be preset to a fixed threshold or dynamically calculated based on real-time conditions. For example, the stopping distance can be estimated based on the vehicle's current speed and a preset minimum safe deceleration, and then the braking setup time and system response redundancy can be added to obtain the emergency braking distance under the current operating conditions. When the obstacle distance is less than this emergency braking distance, it indicates that if the current motion trend continues, the vehicle will have difficulty stopping safely in front of the obstacle, thus triggering automatic braking control.

[0100] In the above embodiments, the braking conditions are not limited to obstacles being less than the emergency braking distance, accelerator pedal opening trending upwards, or vehicle speed trending upwards. They may further include one or more of the following conditions: environmental slope exceeding a preset slope threshold, brake pedal opening remaining at 0, drive torque request continuously increasing, obstacle distance change rate continuously decreasing, and vehicle being in reverse or neutral gear. By adding additional conditions, the targeting of the secondary intervention trigger logic under different misoperation scenarios can be improved.

[0101] In the above embodiments, the automatic braking control is not limited to one-time easing braking; pulse-type easing braking, segmented easing braking, or gradually increasing easing braking can also be used. For example, a smaller braking pulse can be applied first, and the braking duration can be increased if the risk has not been eliminated; or the easing braking intensity can be adaptively adjusted according to the degree of reduction in obstacle distance and the speed increase trend. All of the above methods can be implemented without departing from the core technical concept of this application.

[0102] In some embodiments, reference Figure 4 As shown, the method further includes the following steps: S410: Obtain the vehicle status information and vehicle obstacle information corresponding to the triggering event, and add a label for the triggering event indicating whether it is a valid alarm or a false alarm; S420: Use trigger events marked as valid alarms as positive samples and trigger events marked as false alarms as negative samples to build a historical operation database; S430: Adjust the confidence threshold based on the historical operation library.

[0103] In the above implementation, the triggered events are labeled as either valid alarms or false alarms. A valid alarm refers to an alarm that, after triggering a warning or intervention, is confirmed through manual review, event playback, vehicle log analysis, or user feedback to correspond to a genuine misoperation risk scenario. A false alarm refers to an alarm or intervention that, although triggered by the system, is subsequently confirmed by analysis to not correspond to a genuine misoperation risk, or to be a case of normal driving behavior being misidentified. The labeling process can be completed offline by developers during the testing phase, or it can be completed by an after-sales diagnostic system, remote maintenance platform, or cloud data platform after analyzing event records, video data, perception data, and driver operation trajectories. Through the above labeling, the system can establish a clear correspondence between each triggered event and its true nature.

[0104] In the above implementation, by establishing a historical operation database, a data accumulation mechanism oriented towards real vehicle operation scenarios can be formed. This approach can not only cover experimental scenarios but also gradually incorporate actual event data of mass-produced vehicles under different road environments, driving habits in different regions, and vehicle configurations, thereby making system optimization closer to real application scenarios. Adjusting the confidence threshold based on the historical operation database enables the confidence threshold to have data-driven adaptive optimization capabilities. Compared to a threshold scheme that remains fixed after initial calibration, this approach can dynamically adjust the threshold setting according to false alarms and missed alarms during operation, which is beneficial for balancing recognition sensitivity and false alarm control requirements at different stages and improving the overall performance of misoperation scenario judgment.

[0105] In the above implementation, the confidence threshold adjustment can be performed according to a preset cycle, such as daily, weekly, monthly, or after the cumulative sample size reaches a preset scale. It can also be performed during software version updates, vehicle model changes, or regional parameter changes. The adjusted confidence threshold can be redistributed to the controller and used for subsequent joint determination of erroneous operation scenarios. This forms a closed-loop update mechanism of "trigger event recording—sample labeling—establishment of historical operation database—confidence threshold adjustment—subsequent recognition optimization," thereby achieving continuous improvement of the erroneous operation recognition strategy.

[0106] In the above embodiments, the annotation method for triggered events is not limited to manual offline annotation; semi-automatic annotation or automatic assisted annotation methods can also be used. For example, triggered events can be initially classified based on rule review, video playback results, subsequent driver operation results, or obstacle distance change trends, and then final confirmation can be completed manually. For systems with remote data platforms, event data uploaded by different vehicles can also be uniformly processed by a cloud-based annotation platform to improve annotation efficiency and consistency.

[0107] In some embodiments, adjusting the confidence threshold based on the historical operation library includes: The learning module is used to extract features from the vehicle state information and vehicle obstacle information of the negative and positive samples to determine the offset. The offset is then summed with the confidence threshold to obtain the adjusted confidence threshold.

[0108] In the above embodiments, the learning module can be deployed in the vehicle's local controller, domain controller, or cloud training platform to adaptively adjust the current confidence threshold based on positive and negative samples in the historical operation library. By summing the offset with the original confidence threshold to obtain the adjusted confidence threshold, adaptive optimization of the existing recognition framework can be achieved with low engineering complexity. This approach does not require a complete replacement of the original intent recognition model; it only requires differentiated adjustment of the decision threshold to improve recognition performance in different scenarios, facilitating rapid deployment in existing vehicle control systems.

[0109] Specifically, the learning module acquires labeled trigger event data from the historical operation database, extracts features from vehicle state information and vehicle obstacle information in positive and negative samples to obtain feature combinations that describe scene features and driving behavior features. Based on the distribution differences between positive and negative samples under the same or similar feature combinations, the corresponding offset is determined.

[0110] The offset is used to characterize whether the current default confidence threshold should be adjusted towards a more sensitive or more robust direction in a specific scenario. Specifically, if the learning module analysis finds that, under a certain feature combination, the proportion of negative samples is high, meaning that false alarms are frequently triggered in this type of scenario, it indicates that the current default confidence threshold is too low, causing the warning triggering condition to be overly sensitive. In this case, the learning module can generate a positive offset, so that the offset, when summed with the original confidence threshold, yields a higher adjusted confidence threshold, thereby increasing the subsequent triggering threshold. Conversely, if the learning module analysis finds that, under a certain feature combination, the proportion of positive samples is high, and there are multiple real erroneous operation risk events, it indicates that the current default confidence threshold is too high, causing the system's sensitivity to identify this type of risk scenario to be insufficient. In this case, the learning module can generate a negative offset, so that the offset, when summed with the original confidence threshold, yields a lower adjusted confidence threshold, thereby improving the warning sensitivity for this type of scenario.

[0111] In the above embodiments, the adjusted confidence threshold can be expressed as: Adjusted confidence threshold = Original confidence threshold + Offset Δ

[0112] The offset Δ can be positive, negative, or zero. A positive offset Δ indicates that the default threshold needs to be increased in this scenario; a negative offset Δ indicates that the default threshold needs to be decreased in this scenario; and a zero offset Δ indicates that the current default threshold does not need to be adjusted in this scenario. The calculation of the offset can comprehensively consider factors such as the ratio of positive and negative samples, the driver's corrective behavior after the alarm, the degree of obstacle risk, the dynamic change trend of the vehicle, and the reliability of the sample size, in order to avoid over-adjustment of the threshold due to occasional isolated cases.

[0113] In the above implementation, the learning module can differentiate the confidence threshold according to the granularity of feature combinations. For example, if the learning module finds that most of the historical trigger events of a certain driving object in reverse gear R, at low speed, and with insufficient obstacle risk, are marked as false alarms, the learning module can determine that the default trigger condition in this type of scenario is too sensitive and generate a positive offset for this feature combination, thereby increasing the model output probability threshold for the corresponding scenario from the original value. Conversely, if the learning module finds that multiple trigger events of a certain driving object in neutral gear N, with a large slope, are marked as valid alarms, a negative offset can be generated for this feature combination, thereby decreasing the model output probability threshold for the corresponding scenario from the original value. Thus, the system can form differentiated threshold configurations for different driving features and different scenario features.

[0114] In the above embodiments, the learning module can be implemented in ways that are not limited to simple statistical analysis. It can also employ machine learning classification modules, clustering analysis modules, Bayesian update modules, online learning modules, or rule-assisted learning modules. Any module that can extract scene features based on positive and negative samples and output an offset used to correct the confidence threshold is applicable.

[0115] In another embodiment, this application embodiment also displays the current gear and suggested gear simultaneously through the central control screen and the head-up display module (HUD) to enhance visual guidance. The current gear is used to clearly show the driver the current gear status of the vehicle, and the suggested gear is used to prompt the driver to take the more appropriate gear adjustment direction in the current scenario, so as to reduce the risk of misoperation.

[0116] In some embodiments, the recommended gear is determined jointly by the results of the misoperation scenario identification, the vehicle's direction of travel, the location of the obstacle, and the current risk category. Specifically, a corresponding recommended gear strategy can be generated based on the currently identified misoperation scenario category. For example, when a dangerous backward rushing scenario is identified, if the vehicle is in reverse gear R and the obstacle behind is too close, the recommended gear can be parking gear P or neutral gear N; when a slope rolling scenario is identified, if the vehicle is in neutral gear N, the recommended gear can be forward gear D or reverse gear R that matches the desired driving direction, and the driver is prompted to cooperate with braking or parking operations; when a low-speed continuous approach scenario is identified, the driver can be prompted to maintain the current gear and prioritize applying the brakes, or when the intention to stop is clear, it can be suggested to switch to parking gear P.

[0117] In the above implementation, the current gear and suggested gear are simultaneously displayed on the central control screen and the head-up display (HUD), transforming abstract risk warnings into more specific operational guidance information. Compared to simply indicating the existence of a risk without clearly explaining the corrective direction, this approach directly shows the driver the current status and suggested adjustment target, which helps shorten the driver's understanding and response time.

[0118] It should be noted that the steps shown in the above process or in the flowchart of the accompanying figures can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0119] This application also provides a vehicle, including: a controller, sensors, actuators, an in-vehicle terminal, a cloud service platform, and a communication link connected in communication, wherein the controller is used to execute the control method described in the above embodiments.

[0120] A controller is typically an integrated electronic control module containing a microprocessor, memory, input / output interfaces, and other components. It can receive data from various sensors and systems in real time and make corresponding control decisions based on preset algorithms and logic. Controllers are generally installed in the vehicle's cockpit or engine compartment to ensure a relatively safe location that facilitates electrical connections with other vehicle systems.

[0121] The vehicle provided in this embodiment is used to execute the control method described above, and therefore can achieve the same effect as the implementation method described above.

[0122] The beneficial effects of the above embodiments can be referred to the beneficial effects of the corresponding methods provided above, and will not be repeated here.

[0123] In addition, combined Figures 1 to 4 The vehicle control method described in this application embodiment can be implemented by an electronic device.

[0124] Electronic devices may include processors and memory storing computer program instructions.

[0125] Specifically, the processor may include a central processing unit, an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0126] The memory may include a large-capacity storage device for data or instructions. For example, and not limitingly, the memory may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory may include removable or non-removable (or fixed) media. Where appropriate, the memory may be internal or external to the data processing device. In a particular embodiment, the memory is non-volatile memory. In a particular embodiment, the memory includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0127] Memory can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor.

[0128] The processor implements any of the vehicle control methods described in the above embodiments by reading and executing computer program instructions stored in the memory.

[0129] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0130] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for controlling a vehicle, characterized in that, The method includes: Real-time acquisition of vehicle status information and information on obstacles on the vehicle body; The vehicle status information and vehicle obstacle information are identified using an intent recognition model to determine whether the vehicle is in a misoperation scenario. The intent recognition model performs parallel judgment based on a rule engine and a gradient boosting algorithm, and combines the judgment results of the two to determine whether the vehicle is in a misoperation scenario. If so, multi-level intervention operations are performed, and the multi-level intervention operations provide early warning through information prompts at least.

2. The control method according to claim 1, characterized in that, The rule engine is configured with a mapping relationship between vehicle status information, vehicle obstacle information and misoperation scenarios; The vehicle state information and vehicle obstacle information are identified using an intent recognition model, including: Based on the mapping relationship, determine whether the current vehicle status information and vehicle obstacle information match the misoperation scenario, thereby determining whether the vehicle is in a misoperation scenario and obtaining the rule judgment result.

3. The control method according to claim 2, characterized in that, The gradient boosting algorithm is obtained by pre-training based on vehicle state information and vehicle obstacle information under a preset misoperation scenario; The method of using an intent recognition model to identify the vehicle state information and vehicle obstacle information also includes: Based on the gradient boosting algorithm, the current vehicle status information and vehicle obstacle information are identified to determine whether the vehicle is in a misoperation scenario and output the enhanced judgment confidence.

4. The control method according to claim 3, characterized in that, Determining whether a vehicle is in a misoperation scenario includes: If the rule judgment result is yes, and the enhanced judgment confidence exceeds the confidence threshold, then the vehicle is judged to be in a misoperation scenario, and the trigger event is recorded.

5. The control method according to claim 2, characterized in that, The rule engine determines whether the vehicle status information and vehicle obstacle information correspond to an erroneous operation scenario by looking up a table. If so, it determines that the vehicle is in an erroneous operation scenario.

6. The control method according to any one of claims 1 to 5, characterized in that, The execution of multi-level intervention operations includes: At least a Level 1 warning should be issued via voice prompts and / or icon prompts, and After setting the adjustment time, the intent recognition model is invoked to identify the current vehicle status information and vehicle obstacle information, and then it is determined again whether the vehicle is in a misoperation scenario: If the vehicle is again determined to be in a misoperation scenario, a secondary warning will be issued at least through voice prompts and / or a buzzer alarm, and the triggering event will be recorded again; If it is determined again that the vehicle is not in a misoperation scenario, then it is determined that the driver has performed corrective actions.

7. The control method according to claim 6, characterized in that, If, after the set adjustment period, it is determined again that the vehicle is in a misoperation scenario, and the vehicle status information or vehicle obstacle information meets the braking conditions, then the vehicle will automatically apply braking control.

8. The control method according to claim 7, characterized in that, The method further includes: Obtain the vehicle status information and vehicle obstacle information corresponding to the triggering event, and add a label for the triggering event indicating whether it is a valid alarm or a false alarm; Triggering events marked as valid alarms are used as positive samples, and triggering events marked as false alarms are used as negative samples to build a historical operation database; The confidence threshold is adjusted based on the historical operation database.

9. The control method according to claim 8, characterized in that, Adjusting the confidence threshold based on the historical operation database includes: The learning module is used to extract features from the vehicle state information and vehicle obstacle information of the negative and positive samples to determine the offset. The offset is then summed with the confidence threshold to obtain the adjusted confidence threshold.

10. A vehicle, characterized in that, Includes a controller for performing the control method as described in any one of claims 1 to 9.

Images