An automatic driving vehicle minimum risk strategy adjustment method based on severe weather

Abstract

This invention discloses a method and apparatus for adjusting the minimum risk strategy of autonomous vehicles based on severe weather. The method includes: determining the type and level of severe weather based on onboard multi-sensor data and vehicle-road cooperative data; determining a risk prediction result based on the severe weather level, vehicle surrounding environment data, and vehicle driving status; adjusting the trigger threshold of the minimum risk strategy based on the severe weather level; increasing the triggering methods of the minimum risk strategy based on the risk prediction result and the data effectiveness of the onboard multi-sensor combination; and dynamically adjusting the execution parameters of the minimum risk strategy according to the severe weather type, severe weather level, and risk prediction result. This method, by adjusting the trigger threshold, triggering methods, and execution parameters of the minimum risk strategy, enables each action in the minimum risk strategy to adapt to the current environment and risk state, significantly improving the execution safety and reliability of the minimum risk strategy under severe weather conditions.

Description

Technical Field

[0001] This invention relates to the field of autonomous vehicle control technology, and in particular to a method and apparatus for adjusting the minimum risk strategy of autonomous vehicles based on adverse weather conditions. Background Technology

[0002] Level 3 autonomous vehicles can autonomously complete dynamic driving tasks within their designed operating domain. When the system malfunctions, exceeds the designed operating domain, or the driver fails to respond to the takeover request in a timely manner, the Minimum Risk Maneuver (MRM) must be triggered immediately to safely bring the vehicle to a stationary state or a stable takeover state.

[0003] Currently, the minimum risk strategy of existing autonomous vehicles mainly relies on onboard sensors for environmental perception and executes corresponding operations such as deceleration and moving to the side of the road according to preset rules. However, in real-world driving environments, especially in adverse weather conditions such as rain, snow, fog, and strong winds, the performance of onboard sensors is often significantly affected, leading to a decrease in the accuracy, continuity, and stability of environmental perception data. In such scenarios, existing minimum risk strategies, lacking the ability to adapt to environmental changes, struggle to guarantee the timeliness of strategy triggering and the safety of action execution, posing certain safety risks. Summary of the Invention

[0004] To improve the adaptability and reliability of the minimum risk strategy for autonomous vehicles under adverse weather conditions, this invention provides a method and apparatus for adjusting the minimum risk strategy for autonomous vehicles based on adverse weather conditions.

[0005] In a first aspect, embodiments of the present invention provide a method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather, which may include: Based on data from multiple vehicle sensors and vehicle-road cooperative data, the type and level of severe weather are determined. Based on the aforementioned severe weather level, vehicle surrounding environment data, and vehicle driving status, the risk prediction result is determined; Based on the severity of the severe weather, adjust the trigger threshold of the minimum risk strategy; Based on the risk prediction results and the data effectiveness of the vehicle multi-sensor combination, the triggering method of the minimum risk strategy is increased; The execution parameters of the minimum risk strategy are dynamically adjusted based on the type of severe weather, the level of severe weather, and the risk prediction results.

[0006] In one or more optional embodiments of this application, adjusting the trigger threshold of the minimum risk strategy based on the severe weather level includes: Based on the severity of the weather, determine the preset threshold adjustment range; Depending on the type of the trigger threshold, the trigger threshold of the minimum risk strategy is lowered or raised according to the preset threshold adjustment range.

[0007] In one or more optional embodiments of this application, the risk prediction result includes multiple risk types and a risk level corresponding to each risk type; the risk level includes high, medium, and low; the vehicle-mounted multi-sensor combination includes lidar and millimeter-wave radar; The triggering methods for increasing the minimum risk strategy based on the risk prediction results and the data effectiveness of the combined vehicle multi-sensor system include: If the data effectiveness of both the lidar and the millimeter-wave radar is lower than the preset sensor failure threshold, the minimum risk strategy is triggered. If any risk type in the risk prediction results corresponds to a high risk level, then the minimum risk strategy is triggered.

[0008] In one or more optional embodiments of this application, the data efficiency of the lidar or the millimeter-wave radar is determined by the following method: The data effectiveness is determined based on the data integrity, data accuracy, and data stability of the lidar or the millimeter-wave radar.

[0009] In one or more optional embodiments of this application, the step of dynamically adjusting the execution parameters of the minimum risk strategy based on the severe weather type, the severe weather level, and the risk prediction result includes: If the severe weather type is rain, snow, or fog, then a preset deceleration range is determined based on the severe weather level and the vehicle weight. Based on the road friction, the vehicle is controlled in real time to decelerate within the preset deceleration range, so that the vehicle decelerates smoothly. If the severe weather type is rain, snow, or strong wind, then increase the steering damping; If the severe weather type is rain, snow, or strong wind, the steering force will be adjusted in real time according to the vehicle's side angle. If the severe weather type is heavy fog or satellite navigation signal loss, the inertial navigation module and high-precision map data are integrated to plan a parking trajectory to the emergency lane position, and the curvature of the parking trajectory does not exceed a preset curvature threshold. If the risk type in the risk prediction result is a rear-end collision, then the deceleration operation will be performed first; If the risk type in the risk prediction result is sideslip, then lane centering control will be implemented first. If the risk type in the risk prediction result is location loss, then location calibration should be performed first, and then the docking operation should be performed.

[0010] In one or more optional embodiments of this application, before determining the severe weather type and severe weather level based on vehicle-mounted multi-sensor data and vehicle-road cooperative data, the method further includes: The data is collected based on a combination of vehicle-mounted sensors; wherein, the combination of vehicle-mounted sensors includes multiple of the following: lidar, millimeter-wave radar, vision camera, rain sensor, visibility sensor, temperature and humidity sensor, and wind speed sensor.

[0011] In one or more optional embodiments of this application, after determining the type and level of severe weather based on vehicle-mounted multi-sensor data and vehicle-road cooperative data, the method further includes: If the data effectiveness of the lidar is lower than a preset switching threshold, the main sensor in the vehicle used for environmental perception and positioning will be switched from the lidar to the millimeter-wave radar and inertial navigation module.

[0012] In one or more optional embodiments of this application, after determining the type and level of severe weather based on vehicle-mounted multi-sensor data and vehicle-road cooperative data, the method further includes: Based on the severe weather type, perform hardware self-cleaning or dynamic calibration on the lidar, the millimeter-wave radar, and the vision camera; The data collected by the lidar, millimeter-wave radar, and visual camera are enhanced using the enhancement algorithm corresponding to the severe weather type.

[0013] Secondly, embodiments of the present invention provide a minimum risk strategy adjustment device for autonomous vehicles based on severe weather, which may include: The first determination module is used to determine the type and level of severe weather based on data from onboard multi-sensor systems and vehicle-road cooperative data. The second determining module is used to determine the risk prediction result based on the severe weather level, vehicle surrounding environment data and vehicle driving status; The first adjustment module is used to adjust the trigger threshold of the minimum risk strategy based on the severe weather level. The second adjustment module is used to increase the triggering method of the minimum risk strategy based on the risk prediction results and the data effectiveness of the vehicle multi-sensor combination. The third adjustment module is used to dynamically adjust the execution parameters of the minimum risk strategy based on the severe weather type, the severe weather level, and the risk prediction result.

[0014] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather as described above.

[0015] Fourthly, embodiments of the present invention provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method for adjusting the minimum risk strategy of an autonomous vehicle based on severe weather as described above.

[0016] Fifthly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-described method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather.

[0017] In a sixth aspect, embodiments of the present invention provide a minimum risk strategy adjustment device for autonomous vehicles based on severe weather, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, it implements the minimum risk strategy adjustment method for autonomous vehicles based on severe weather as described above.

[0018] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following: This invention provides a method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather. This method accurately identifies the type and level of severe weather by fusing data from multiple onboard sensors and vehicle-road cooperative data, and dynamically adjusts the trigger threshold of the minimum risk strategy accordingly. This avoids the problem of fixed thresholds triggering too early or too late in severe weather. Simultaneously, it dynamically adjusts execution parameters based on the type, level, and risk prediction results of the severe weather, ensuring that each action in the minimum risk strategy adapts to the current environment and risk state. Furthermore, based on the risk prediction results and the data efficiency of onboard sensors, this method increases the strategy triggering methods, enabling the system to intervene promptly when perception performance deteriorates or high-risk predictions are made, improving the rationality of the minimum risk strategy triggering timing and significantly enhancing the execution safety and reliability of the minimum risk strategy under severe weather conditions.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 A flowchart illustrating the method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the minimum risk strategy adjustment device for autonomous vehicles based on severe weather, provided in an embodiment of the present invention. Detailed Implementation

[0022] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0023] The inventors discovered that in existing technologies, Level 3 autonomous driving falls under the category of conditional autonomous driving. Its core characteristic is that within a pre-defined Operational Design Domain (ODD), the vehicle can autonomously complete dynamic driving tasks, including core operations such as speed control, lane keeping, and following other vehicles. A takeover request is only issued to the driver when the vehicle exceeds the ODD, malfunctions, or encounters a sudden risk. When the system malfunctions, exceeds the ODD, or the driver fails to respond to the takeover request in a timely manner (e.g., takeover timeout, driver fatigue or distraction), a minimum risk strategy must be immediately triggered to safely bring the vehicle to a stationary or stable takeover state. Common minimum risk strategy actions include graded deceleration, lane keeping, emergency parking, and activating hazard lights. In existing technologies, most minimum risk strategies use fixed control parameters and preset action sequences, failing to adequately consider the significant negative impact of adverse weather on the vehicle's perception system. However, adverse weather is precisely a typical scenario where minimum risk strategies are frequently triggered in Level 3 autonomous driving scenarios. Specifically, rain and snow can easily lead to blurred images and increased image noise in visual cameras, and a significant reduction in the detection range of lidar (the effective recognition distance can decrease by more than 50% under heavy rain conditions, and the point cloud density is significantly reduced); foggy weather can cause sensor signal scattering, leading to a more than 30% increase in the false recognition rate of millimeter-wave radar targets, and visual cameras can almost lose their long-range recognition capabilities; strong winds will disrupt the stability of vehicle driving posture, easily causing vehicle swerving and tail-wagging, especially affecting commercial vehicles more significantly; high dust and heavy water fog will further exacerbate the degradation of sensor performance, leading to a decrease in the accuracy of minimum risk strategy execution and highlighting safety hazards, which may even cause rear-end collisions and other traffic accidents in severe cases. Based on this, the inventors have further developed this invention, providing a method and device for adjusting the minimum risk strategy of autonomous vehicles based on severe weather conditions.

[0024] Example 1 Embodiment 1 of the present invention provides a method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather, referring to... Figure 1 As shown, the method may include the following steps S101-S105: S101: Based on onboard multi-sensor data and vehicle-road cooperative data, determine the type and level of severe weather.

[0025] S102: Determine the risk prediction result based on the severity of severe weather, vehicle surrounding environment data, and vehicle driving status.

[0026] S103: Adjust the trigger threshold of the minimum risk strategy based on the severity of severe weather.

[0027] S104: Based on the risk prediction results and the data effectiveness of the vehicle multi-sensor combination, increase the triggering method of the minimum risk strategy.

[0028] S105: Dynamically adjust the execution parameters of the minimum risk strategy based on the type and level of severe weather and the risk prediction results.

[0029] This invention provides a method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather. This method accurately identifies the type and level of severe weather by fusing data from multiple onboard sensors and vehicle-road cooperative data, and dynamically adjusts the trigger threshold of the minimum risk strategy accordingly. This avoids the problem of fixed thresholds triggering too early or too late in severe weather. Simultaneously, it dynamically adjusts execution parameters based on the type, level, and risk prediction results of the severe weather, ensuring that each action in the minimum risk strategy adapts to the current environment and risk state. Furthermore, based on the risk prediction results and the data efficiency of onboard sensors, this method increases the strategy triggering methods, enabling the system to intervene promptly when perception performance deteriorates or high-risk predictions are made, improving the rationality of the minimum risk strategy triggering timing and significantly enhancing the execution safety and reliability of the minimum risk strategy under severe weather conditions.

[0030] In this embodiment of the application, before performing step S101, step S100 is further included: collecting vehicle multi-sensor data based on a combination of vehicle multi-sensors. The combination of vehicle multi-sensors includes multiple sensors selected from LiDAR, millimeter-wave radar, vision camera, rain sensor, visibility sensor, temperature and humidity sensor, and wind speed sensor.

[0031] Specifically, this can be achieved by using a LiDAR device deployed on the vehicle's roof to collect 3D point cloud data, which is then used to obtain distance, orientation, and contour information of obstacles around the vehicle. Millimeter-wave radars deployed on the front bumper and four corners of the vehicle collect relative distance, relative speed, and azimuth information of obstacles. A vision camera deployed behind the windshield collects road image information, used to identify lane lines, traffic signs, pedestrians, and vehicles. A rain sensor deployed on the exterior of the vehicle collects rainfall data. A visibility sensor deployed on the roof or front grille collects visibility data. A temperature and humidity sensor deployed on the vehicle body collects ambient temperature and relative humidity data. A wind speed sensor deployed on the roof or rearview mirrors collects wind speed and direction data. The sampling frequency of the above multi-sensor data is no less than 10 Hz to ensure the real-time performance of this method.

[0032] In step S101 above, the type and level of severe weather are determined based on vehicle-mounted multi-sensor data and vehicle-road cooperative data.

[0033] Specifically, it can be to receive roadside sensor data from roadside equipment via Vehicle-to-Infrastructure (V2I) and weather warning data from the meteorological service center via Vehicle-to-Network (V2N), combine them to obtain vehicle-road cooperative data, and jointly construct a three-in-one environmental perception network of vehicle-end perception, roadside perception, and weather warning with the vehicle-mounted multi-sensor data collected in step S100 above.

[0034] Among them, vehicle-to-infrastructure (V2I) data is mainly used to supplement the blind spots of vehicle-mounted sensors in adverse weather conditions, and to obtain global environmental information such as overall weather conditions, temporary traffic control information, and the distribution of obstacles ahead. The transmission of V2I data adopts 5G communication technology, and the data transmission latency is controlled within 100 milliseconds.

[0035] Based on the rainfall, visibility, wind speed, ambient temperature, and relative humidity collected in step S100, and combined with V2N weather warning information from vehicle-to-everything (V2N) data, the current severe weather type of the vehicle is identified. Severe weather types include rain / snow, fog, strong winds, high dust, and heavy water fog. For example, when rainfall exceeds a corresponding preset threshold, it is determined to be rain / snow; when visibility and relative humidity are both below the corresponding preset thresholds, it is determined to be fog; and when wind speed exceeds the corresponding preset threshold, it is determined to be strong winds.

[0036] The rainfall, visibility, wind speed collected in step S100, and the road surface friction force measured or estimated by the vehicle-mounted sensor are used as input parameters and input into the preset weather level evaluation model to obtain the severe weather level.

[0037] The preset weather level evaluation model is constructed based on fuzzy comprehensive evaluation method and machine learning.

[0038] Specifically, based on the impact of severe weather on autonomous driving, severe weather is divided into four levels: extremely high risk, high risk, relatively high risk, and general risk, with each level corresponding to a specific quantitative indicator. For example, rainfall exceeding 20 millimeters per hour is considered extremely high risk, 10 to 20 millimeters per hour is considered high risk, 5 to 10 millimeters per hour is considered relatively high risk, and less than 5 millimeters per hour is considered general risk.

[0039] The fuzzy comprehensive evaluation method quantifies and assesses the above parameters—rainfall, visibility, wind speed, and road surface friction—to obtain scores for each parameter, resulting in a comprehensive risk score. A higher comprehensive risk score indicates a greater impact of current weather conditions on autonomous driving.

[0040] Machine learning continuously optimizes the scoring rules and weighting coefficients of each parameter in the fuzzy comprehensive evaluation method based on historical weather data and road accident statistics, making the mapping relationship between the comprehensive risk score and the actual accident risk more accurate. By comparing the optimized comprehensive risk score with multiple preset thresholds, severe weather can be divided into four levels: extremely high, high, relatively high, and moderate. Each level corresponds to a clear quantitative indicator, and the judgment error does not exceed 10%.

[0041] Furthermore, the input parameters of the preset weather level assessment model also include historical weather data and road accident statistics. The historical weather data is used to adapt to the weather characteristics of different seasons and regions, while the road accident statistics are used to optimize the accuracy of risk prediction. The model parameters are continuously optimized through machine learning algorithms to improve the accuracy of level classification and adapt to the weather impact characteristics of different road sections (such as the difference in the impact of fog on mountainous road sections and plain road sections).

[0042] In this embodiment of the application, after completing the above step S101, step S1010 is further included, specifically including the following steps S10101-S10103: S10101: If the data efficiency of the LiDAR is lower than the preset switching threshold, the main sensor used for environmental perception and positioning in the vehicle will be switched from LiDAR to millimeter-wave radar and inertial navigation module.

[0043] Specifically, a dual-redundancy architecture with primary and backup sensors can be adopted, with the preset sensor priority as follows: LiDAR takes precedence over millimeter-wave radar, millimeter-wave radar takes precedence over vision camera, and vision camera takes precedence over inertial navigation module. LiDAR serves as the primary sensor for environmental perception and localization, enabling long-range, high-precision obstacle detection; millimeter-wave radar serves as the first backup sensor, handling adverse weather conditions such as rain, snow, and fog; the vision camera serves as the second backup sensor, assisting in target recognition; and the inertial navigation module serves as an emergency backup sensor, used for locating lost targets.

[0044] The system monitors the effectiveness of the LiDAR data in real time. When the LiDAR data effectiveness falls below a preset switching threshold, the main sensors used for environmental perception and positioning in the vehicle are switched from LiDAR to millimeter-wave radar and inertial navigation module. The switching process is completed within 50 milliseconds, with a switching delay of less than or equal to 50 milliseconds, achieving a seamless switching and ensuring uninterrupted environmental perception data acquisition.

[0045] After switching, the millimeter-wave radar serves as the primary sensor for environmental perception and positioning, detecting the distance and relative speed of obstacles in front of and around the vehicle. This millimeter-wave radar is a 4D millimeter-wave radar, capable of outputting high-density point cloud data. Based on this, a fusion technology combining 4D millimeter-wave point cloud simultaneous localization and mapping (SLAM) and inertial navigation is employed. The point cloud data collected by the 4D millimeter-wave radar is fused with the position data collected by the inertial navigation module, and a fusion filtering algorithm is used to optimize positioning accuracy. Even in scenarios where satellite navigation signals are lost, such as when the vehicle is traveling in a tunnel or in heavy fog causing a loss of GPS signal, this fusion technology can still maintain centimeter-level positioning accuracy, reaching ±2 cm, with positioning drift controlled within ±5 cm. This ensures the accurate execution of the minimum-risk strategy under conditions of limited positioning signals.

[0046] The data effectiveness of lidar is determined by the following method: based on the data integrity, data accuracy, and data stability of lidar or millimeter-wave radar, the data effectiveness is determined.

[0047] Specifically, this can involve real-time monitoring of the data integrity, accuracy, and stability of the LiDAR-acquired data. Data integrity assesses the continuity of the LiDAR point cloud data packets; data accuracy measures the consistency between the LiDAR measurements and the actual values; and data stability examines the degree of fluctuation in the LiDAR output data over time. Each of these three indicators is quantified and scored, with values ​​ranging from 0 to 100%, and then a weighted summation method is used to calculate the data effectiveness rate. For example, when the data integrity score is higher than 90%, or the accuracy score is higher than 85%, and the stability score is higher than 80%, the overall calculated data effectiveness rate is higher than a preset switching threshold.

[0048] S10102: Perform hardware self-cleaning or dynamic calibration for lidar, millimeter-wave radar, and vision cameras based on severe weather type.

[0049] Specifically, when the severe weather type is rain, snow, high dust, or heavy fog, the sensor self-cleaning module can be activated. This sensor self-cleaning module adopts a dual mode of heating and spraying. In rain, snow, or high dust weather, the heating / spraying snow removal and decontamination function is activated to remove the attachments on the sensor surface and ensure that the sensor detection is unobstructed.

[0050] In strong winds, the attitude adaptive calibration module adjusts the sensor's installation attitude in real time to reduce the impact of wind resistance on the sensor's installation attitude and ensure the accuracy of the sensor's detection angle.

[0051] Furthermore, it also includes a dynamic calibration module, which periodically calibrates the accuracy of each sensor, with a calibration cycle of no more than 5 minutes, to ensure the accuracy of the sensor-collected data.

[0052] In the spray mode, the spray pressure is controlled at 0.3 MPa for 5 seconds, and in the heating mode, the heating temperature is controlled at 50 degrees Celsius for 10 seconds.

[0053] S10103: Enhance the data collected by lidar, millimeter-wave radar and vision camera using enhancement algorithms corresponding to severe weather types.

[0054] Specifically, when the severe weather type is rain or snow, a deep learning-based image de-raining and de-snowing algorithm is used for the images captured by the visual camera, and the dark channel prior algorithm is optimized to improve the image clarity after de-raining and de-snowing; a distance compensation algorithm is used for the point cloud data collected by the lidar, and the detection error is corrected according to the attenuation law of rainfall and lidar detection distance to ensure lidar detection accuracy.

[0055] When the severe weather type is heavy fog, a signal enhancement algorithm is applied to the signals collected by the millimeter-wave radar to improve the target recognition accuracy of the millimeter-wave radar by filtering out scattered interference signals. At the same time, 4D millimeter-wave point cloud, SLAM and inertial navigation data are integrated to compensate for the blind spots of visual sensors and achieve centimeter-level positioning in scenarios where satellite navigation signals are lost.

[0056] When the severe weather type is strong wind, a fusion filtering algorithm is used for the data collected by lidar, millimeter-wave radar and vision camera. Kalman filtering algorithm can be selected to remove abnormal perception data caused by wind resistance and improve the stability of the data.

[0057] In this embodiment of the application, the method, through the above step S1010, adopts a triple anti-interference scheme of "hardware protection + algorithm optimization + redundancy backup" to comprehensively ensure the continuity and accuracy of environmental perception data acquisition, and solves the core pain point of sensor performance degradation under severe weather conditions.

[0058] In step S102 above, the risk prediction result is determined based on the severity of the weather, the surrounding environment data of the vehicle, and the vehicle's driving status.

[0059] Specifically, the severe weather level determined in step S101 can be used as input, combined with data on the vehicle's surrounding environment and real-time vehicle driving status, to jointly predict risks. The vehicle's surrounding environment data includes obstacle distribution, lane status, traffic control information, and road surface material, while the vehicle driving status includes vehicle speed, vehicle posture, road surface friction, vehicle type, and vehicle weight.

[0060] Based on the severity of severe weather, assess the overall impact of severe weather on driving safety. Based on the obstacle distribution in the vehicle's surrounding environment data, determine whether there are stationary or moving obstacles in front, to the sides, and behind, and their relative distance and speed. Based on the vehicle's speed and road friction during its driving state, assess the vehicle's braking distance and stability margin.

[0061] Based on the above information, we predict the types of risks the vehicle may face under real-time driving conditions and the corresponding risk levels for each type. Risk types include, but are not limited to, rear-end collision risk, skidding risk, collision risk, and location loss risk. Risk levels are categorized into high, medium, and low. The risk types and their corresponding risk levels constitute the risk prediction results.

[0062] For example, when the severe weather level is extremely high, there is a stationary obstacle ahead, and the current vehicle speed is high enough to result in insufficient braking distance, the risk level of a rear-end collision is predicted to be high. When the severe weather level is high, the road surface friction is low, and the vehicle body shows a tendency to swerve, the risk level of skidding is predicted to be medium or high. When the severe weather type is dense fog and satellite navigation signal is lost, the risk level of positioning loss is predicted to be high.

[0063] In step S103 above, the trigger threshold for the minimum risk strategy is adjusted based on the severity of the severe weather. Specifically, this includes the following steps S1031-S1032: S1031: Determine the preset threshold adjustment range based on the severity of severe weather.

[0064] Specifically, when the severe weather level is extremely high, the corresponding preset threshold adjustment range is 30% to 40%. When the severe weather level is high, the corresponding preset threshold adjustment range is 20% to 30%. When the severe weather level is relatively high, the corresponding preset threshold adjustment range is 10% to 20%. When the severe weather level is moderate, the corresponding preset threshold adjustment range is to keep the basic trigger threshold unchanged, that is, the adjustment range is 0.

[0065] S1032: Depending on the type of trigger threshold, the trigger threshold of the minimum risk strategy is lowered or raised according to the preset threshold adjustment range.

[0066] Specifically, the trigger threshold for the minimum risk strategy can include multiple types, with different types of thresholds adjusted in different directions based on their physical meaning. When the trigger threshold type is a distance threshold, such as "triggering the minimum risk strategy when the obstacle distance is less than or equal to a preset distance threshold," then the preset distance threshold should be increased according to the preset threshold adjustment range so that the vehicle can trigger earlier from a greater distance. When the trigger threshold type is an angle threshold, such as "triggering the minimum risk strategy when the vehicle body side-slip angle is greater than or equal to a preset angle threshold," then the preset angle threshold should be decreased according to the preset threshold adjustment range so that the vehicle can trigger at a smaller side-slip angle.

[0067] Regardless of the adjustment direction adopted, the core objective is to lower the trigger threshold during periods of severe weather, making the minimum risk strategy more easily triggered and thus enabling early hazard avoidance, allowing vehicles sufficient time to avoid risks. Simultaneously, the trigger threshold adjustment employs a linear adjustment method to ensure a smooth adjustment process and avoid false triggering of the minimum risk strategy due to sudden threshold changes.

[0068] In step S104 above, based on the risk prediction results and the data effectiveness of the combined vehicle multi-sensor system, a triggering method for the minimum risk strategy is added. The risk prediction results include multiple risk types and the corresponding risk level for each risk type. The risk levels include high, medium, and low. Step S104 specifically includes the following steps S1041-S1042: S1041: If the data effectiveness of both lidar and millimeter-wave radar is lower than the preset sensor failure threshold, the minimum risk strategy is triggered.

[0069] Specifically, this can involve real-time monitoring of the data effectiveness of both LiDAR and millimeter-wave radar. When the data effectiveness of both LiDAR and millimeter-wave radar falls below a preset sensor failure threshold, it is determined that the onboard environmental perception system can no longer provide reliable obstacle detection information. At this point, regardless of whether a system malfunction is detected, whether the driver has failed to take over within the allotted time, or the risk prediction result, the minimum risk strategy is directly triggered, and a safe stopping operation is performed. The preset sensor failure threshold can be set to 60%. This triggering method allows the vehicle to proactively enter a minimum risk state even in extreme situations where severe weather causes both primary and backup perception sensors to fail simultaneously, thus avoiding safety accidents caused by missing perception information.

[0070] The data effectiveness rate of lidar or millimeter-wave radar is determined as follows: based on the data integrity, accuracy, and stability of the lidar or millimeter-wave radar, the data effectiveness rate is determined. The specific calculation method is explained in step S10101 above and will not be repeated here.

[0071] S1042: If any risk type in the risk prediction results corresponds to a high risk level, then the minimum risk strategy is triggered.

[0072] Specifically, the risk prediction results can include multiple risk types and the corresponding risk level for each risk type. Each risk type's risk level is checked one by one. If any risk type is determined to be high, the minimum risk strategy is triggered directly without waiting for a system failure, driver takeover timeout, or other triggering conditions to be met.

[0073] For example, when the risk assessment indicates a high risk level for a rear-end collision, the minimum risk strategy is immediately triggered to perform deceleration and hazard avoidance. When the risk assessment indicates a high risk level for skidding, the minimum risk strategy is immediately triggered to perform lane centering control. This addition of triggering methods allows for proactive risk avoidance by intervening in advance when the vehicle anticipates an impending high-risk event.

[0074] Furthermore, by combining historical data from the execution of the minimum risk strategy, the preset threshold adjustment range determined based on the severe weather level in step S103 and the preset sensor failure threshold in step S104 are corrected in real time using machine learning or deep learning algorithms.

[0075] Specifically, this can involve collecting historical data from each execution of the minimum risk strategy, including the weather scenario at the time of triggering, the triggering timing, the execution effect, and whether an accident occurred. This historical data is used as training samples to train a model using a neural network algorithm, establishing a threshold correction model. This model aims to adjust the preset threshold adjustment range and preset sensor failure threshold based on the differences in severe weather characteristics across different regions and seasons, such as the regional differences between rainy southern regions and snowy northern regions. For example, in rainy southern regions, where vehicles and drivers are relatively more adaptable to rainy conditions, the preset threshold adjustment range based on severe weather levels or the preset sensor failure threshold can be appropriately lowered to avoid premature triggering and unnecessary frequent interventions. In snowy northern regions, where the friction coefficient of icy and snowy roads is low, the preset threshold adjustment range can be appropriately expanded or the preset sensor failure threshold can be increased, making the minimum risk strategy easier to trigger and allowing for early intervention to ensure safety. Through this dynamic threshold correction mechanism, the accuracy of minimum risk strategy triggering and its scenario adaptability are further improved.

[0076] In this embodiment of the application, the above steps S103-S104 dynamically adjust the trigger threshold and trigger logic of the minimum risk strategy based on the severity of severe weather, risk prediction results and the data effectiveness of sensors, so as to avoid the problems of false triggering and missed triggering of the minimum risk strategy in all aspects and ensure that the minimum risk strategy can be triggered accurately at the appropriate time.

[0077] In step S105 above, the execution parameters of the minimum risk strategy are dynamically adjusted based on the type and level of severe weather and the risk prediction results. Specifically, this includes the following steps S1051-S1058: S1051: If the severe weather type is rain, snow or fog, the preset deceleration range is determined based on the severe weather level and vehicle weight.

[0078] Specifically, when the severe weather type is rain, snow, or fog, the road surface friction is low, which can easily lead to rear-end collisions or skidding accidents. Therefore, a progressive graded deceleration control mode is adopted. The preset deceleration range is determined based on the severity of the severe weather and the vehicle's weight.

[0079] For example, when the severe weather level is extremely high or high, the preset deceleration range can be set to 0.3g to 0.4g; when the severe weather level is relatively high or moderate, the preset deceleration range can be set to 0.4g to 0.5g, where g is the gravitational acceleration, approximately 9.8 m / s².

[0080] For passenger vehicles, which are lighter and more agile, the default preset deceleration range used in the example above is used to control the total deceleration range of 0.3g to 0.5g under all severe weather conditions. For commercial vehicles, due to their greater weight and inertia, the preset deceleration range is correspondingly reduced to extend the deceleration time, ensure the stability of the avoidance control, and prevent the vehicle from overturning due to sudden braking. The total deceleration range of commercial vehicles under all severe weather conditions can be controlled to be 0.25g to 0.4g. For extremely high or high weather conditions, the preset deceleration range can be set to 0.25g to 0.35g, and for relatively high or normal weather conditions, it can be set to 0.35g to 0.4g.

[0081] S1052: Based on road friction, the vehicle is controlled in real time to decelerate within a preset deceleration range so that the vehicle decelerates smoothly.

[0082] Specifically, this can be achieved by monitoring road surface friction in real time during the execution of the minimum risk strategy. The actual deceleration is controlled within the preset deceleration range determined in S1051, and adjusted in real time according to changes in road surface friction, extending the deceleration time accordingly. When road surface friction is low, the deceleration is controlled to tend towards the lower limit of the preset deceleration range, making the deceleration process smoother and avoiding wheel lock-up or vehicle skidding due to sudden braking. When road surface friction is high, the deceleration is controlled to tend towards the upper limit of the preset deceleration range, improving deceleration efficiency while ensuring safety. Real-time deceleration control can be implemented using a closed-loop control (Proportional-Integral-Derivative, PID) algorithm to ensure the stability and response speed of deceleration adjustment. Through this method, the vehicle can achieve smooth deceleration under different road surface conditions, reducing the risk of rear-end collisions and skidding.

[0083] S1053: If the severe weather type is rain, snow, or strong wind, increase the steering damping.

[0084] Specifically, when the weather conditions are rain, snow, or strong winds, vehicles are prone to lateral deviation during driving, and the steering system experiences significantly increased external interference. In such situations, a lane centering control strategy can be employed, increasing steering damping to generate greater resistance when the steering wheel is turned, preventing unexpected steering wheel deflection due to uneven road surfaces or crosswinds. The increase in steering damping can be adjusted according to the severity of the weather; the higher the severity, the greater the increase in steering damping. By increasing steering damping, the vehicle maintains a more stable driving posture on slippery roads or in strong crosswind conditions, mitigating safety risks caused by sharp turns and preventing loss of control due to panicked steering by the driver or the autonomous driving system.

[0085] S1054: If the severe weather type is rain, snow, or strong wind, the steering force will be adjusted in real time according to the vehicle's side angle.

[0086] Specifically, this could involve real-time monitoring of the vehicle's lateral tilt angle during severe weather conditions such as rain, snow, or strong winds. The lateral tilt angle can be obtained through an inertial navigation module or a vehicle attitude sensor, representing the deviation between the vehicle's driving direction and its pointing direction.

[0087] As described in step S1053 above, when the severe weather type is rain, snow, or strong wind, the vehicle is prone to lateral deviation during driving, and the external interference to the steering system increases significantly. Therefore, a small-amplitude steering control strategy can be adopted during steering. When an increase in the vehicle's lateral deviation angle is detected, which will cause a deviation in the driving trajectory, the steering force is adjusted according to the magnitude of the lateral deviation angle based on the PID algorithm to pull the vehicle back to the normal driving trajectory.

[0088] S1055: If the severe weather type is heavy fog or satellite navigation signal loss, the inertial navigation module and high-precision map data will be integrated to plan the parking trajectory to the emergency lane position, and the curvature of the parking trajectory will not exceed the preset curvature threshold.

[0089] Specifically, in situations where severe weather conditions such as heavy fog or vehicles traveling through tunnels cause satellite navigation signals to be lost, the perception capabilities of visual cameras and LiDAR are severely limited, making it difficult to determine the vehicle's precise location using satellite positioning signals. In such cases, inertial navigation module and high-precision map data are integrated for positioning and path planning. The inertial navigation module provides the vehicle's acceleration, angular velocity, and attitude information to calculate the vehicle's relative position change; high-precision map data provides precise geometric information about the road, including lane line positions, emergency lane positions, and road curvature. This determines the vehicle's current position and plans a stopping trajectory from the current position to the emergency lane position. Furthermore, the curvature of the planned stopping trajectory is controlled to not exceed a preset curvature threshold, which can be set to 0.05m. 1 This ensures a smooth parking process, preventing vehicle tilting or passenger discomfort caused by excessive curvature of the parking trajectory. Through these methods, even in conditions of extremely low visibility or loss of positioning signal, the vehicle can still accurately identify the emergency lane location and smoothly enter the parking area, completing a safe parking operation with minimal risk.

[0090] S1056: If the risk type in the risk prediction result is rear-end collision, then the deceleration operation shall be performed first.

[0091] Specifically, when the risk assessment result includes a rear-end collision risk, it is determined that there is a collision hazard between the vehicle and the vehicle in front or the obstacle ahead. In this case, during the implementation of the minimum risk strategy, deceleration is the highest priority action, prioritizing reducing the vehicle speed to increase the distance between the vehicle and the obstacle ahead, thus avoiding a rear-end collision. After the deceleration operation is completed and the risk of a rear-end collision is no longer present, other actions are then performed according to the actual situation, such as lane keeping or pulling over.

[0092] S1057: If the risk type in the risk prediction result is sideslip, then lane centering control shall be implemented first.

[0093] Specifically, when the risk assessment result includes a risk of sideslip, it is determined that the vehicle has a tendency to deviate from its lane or lose control under the current driving conditions. In this case, during the execution of the minimum risk strategy, lane centering control is the highest priority action, prioritizing the suppression of the vehicle's sideslip tendency and adjusting the vehicle's attitude to the center of the lane. Lane centering control includes increasing steering damping and adjusting steering force in real time according to the vehicle's sideslip angle; specific execution methods can be found in S1053 to S1054. After the sideslip risk is effectively suppressed and the vehicle's driving attitude returns to stability, other operations such as deceleration or pulling over are then performed according to the actual situation.

[0094] S1058: If the risk type in the risk prediction result is location loss, then location calibration should be performed first, and then docking operation should be performed.

[0095] Specifically, when the risk assessment results include the risk of location loss, it can be determined that the vehicle cannot currently obtain accurate location information. Directly performing a pullover operation might not allow it to accurately enter the emergency lane or stop in a safe location. In this case, during the implementation of the minimum risk strategy, location calibration is the highest priority action. The inertial navigation module and high-precision map data are integrated first for location calibration to restore the vehicle's positioning accuracy. After location calibration is completed and the vehicle's location information is restored accurately, a parking trajectory is planned and the pullover operation is performed to ensure the vehicle accurately stops in a safe location within the emergency lane.

[0096] Furthermore, this application also includes a step of constructing a minimum risk strategy action library: Pre-collecting minimum risk strategy execution data for different severe weather types, road conditions, and vehicles, covering multiple scenarios in rainy / snowy weather, foggy weather, strong wind weather, high dust weather, and heavy water fog weather, with a data volume of no less than 100,000 sets to ensure the coverage of the minimum risk strategy action library. The minimum risk strategy action library is generated through machine learning training. Specifically, the machine learning uses reinforcement learning algorithms to generate optimal execution parameters and action sequences for different scenarios through continuous iterative optimization. Thus, a minimum risk strategy action library is obtained, containing optimal execution parameters and action sequences for different scenarios, used for rapid matching and dynamic adjustment of minimum risk strategy action sequences and basic execution parameters, improving strategy response speed, with a response time of no more than 200 milliseconds.

[0097] Based on the severe weather type and the risk type in the risk prediction results, a complete action sequence and basic execution parameters are matched from a pre-built minimum risk strategy action library. The action sequence defines the temporal order and priority relationship of each action. Then, according to step S105 above, the basic execution parameters of the action sequence are adjusted and modified. Through the above fast matching mechanism, the matching time does not exceed 50 milliseconds, ensuring that the minimum risk strategy can quickly adapt to changes in the scenario.

[0098] Meanwhile, the minimum risk strategy action library supports online updates and can be continuously optimized based on new weather scenarios and strategy execution data to adapt to new severe weather conditions, ensuring the timeliness and adaptability of the minimum risk strategy action library. Machine learning continuously optimizes the action sequences and execution parameters in the action library based on historical execution results, achieving adaptive iteration of action sequences.

[0099] Furthermore, this method also includes a meteorological early warning linkage step. Specifically, this can involve connecting to the meteorological early warning system in the area where the vehicle is located, establishing a real-time data exchange channel, with a data exchange frequency of no less than once per minute. When severe weather warning information is received, steps S103 and S105 are executed, and the sensor anti-interference preprocessing procedure in step S10102 is initiated to achieve advance prediction and preparation of the minimum risk strategy, avoiding the problem of untimely risk avoidance caused by sudden severe weather, and reserving sufficient preparation time of no less than 3 seconds for the triggering of the minimum risk strategy.

[0100] In this embodiment of the application, after completing the above step S105, step S106 is also included, in which the dynamically adjusted minimum risk strategy is executed, and feedback correction and full-process log saving are performed.

[0101] Specifically, this could involve implementing a dynamically adjusted minimum-risk strategy. During execution, real-time data collection is performed on vehicle driving data, environmental change data, and the operational status data of each sensor in the onboard multi-sensor combination, with a sampling frequency of at least 10 Hz, to construct a real-time feedback closed loop. Vehicle driving data includes vehicle speed, deceleration, steering angle, and vehicle attitude; environmental change data includes weather level, obstacle movement trajectory, and road surface friction change curves.

[0102] The execution parameters of the minimum-risk strategy are adjusted in real time through a dual-feedback correction mechanism combining a PID algorithm and machine learning. The PID algorithm is used for real-time fine-tuning of execution parameters to ensure accuracy. The machine learning algorithm optimizes action parameter settings based on scene changes and execution results, ensuring real-time matching of actions with environmental changes and vehicle driving data. This dual-feedback correction mechanism continues to operate until the vehicle is safely stationary or the driver successfully takes over.

[0103] When the driver initiates takeover, the system gradually exits the minimum risk strategy control and enters a takeover transition period. During this transition period, auxiliary control of steering and deceleration is maintained to give the driver a buffer time to take over. The driver's operating status is continuously monitored, and once the driver's operation is stable and there are no abnormal operations for 3 seconds, the minimum risk strategy control mode is completely exited to ensure a smooth and safe takeover process.

[0104] The entire process log of the minimum risk strategy is synchronously retained. The log content includes onboard multi-sensor data from 10 seconds before triggering to the end of execution, severe weather type and severity level, vehicle-to-infrastructure (V2I) data, execution parameters of the minimum risk strategy, action sequence, and vehicle driving status. The logs are stored encrypted using the Advanced Encryption Standard (AES-256) and for at least 6 months. The retained full-process logs can be integrated with the back-end management systems of traffic management departments and vehicle manufacturers for accident liability tracing and strategy optimization. They comply with expected functional safety standards and autonomous vehicle terminology standards, ensuring compliance and traceability.

[0105] To facilitate understanding of this method by those skilled in the art, the following three specific examples further illustrate the method in detail. Each example is based on the technical solution of this method and focuses on different severe weather scenarios: Example 1: Minimum risk strategy adjustment under moderate rain and snow (high severity weather level) A Level 3 autonomous passenger vehicle (model: mid-size sedan, weight 1500kg) was traveling on a highway (road surface: asphalt, speed limit 120 km / h) at a speed of 100 km / h. The rain sensor detected 15 mm / h of rainfall. The image captured by the vision camera (resolution 1920×1080 pixels) showed significant rain streaks, reducing image clarity by 60%. The LiDAR (detection range 150m) detection range was reduced to 60% of normal operating range (i.e., 90m). The visibility sensor detected 80m of visibility. The millimeter-wave radar (detection range 200m) showed no significant anomalies, with a data validity rate of 92%. According to the fusion of vehicle-road cooperative data, the road section was experiencing moderate rain and snow, with road surface friction reduced by 30% (from 0.8 under normal operating conditions to 0.56). There was no temporary traffic control 500 meters ahead, and no large vehicles in the surrounding lanes. The preset weather level assessment model determined the severe weather level to be high, and the predicted risk types were skidding and rear-end collision, both at a medium risk level.

[0106] S101+S102: A pre-defined weather level assessment model using "fuzzy comprehensive evaluation method + machine learning correction" was implemented. Input parameters included rainfall of 15 mm / hour, visibility of 80 meters, wind speed of 5 m / s, and road surface friction of 0.56. The fuzzy comprehensive evaluation method quantified and scored each parameter (3 points for rainfall, 3 points for visibility, 1 point for wind speed, and 3 points for road surface friction, for a total of 10 points). After machine learning correction, the final severe weather level was determined to be high. Simultaneously, combined with vehicle speed of 100 km / h, vehicle driving status with no large vehicles nearby, and data on the surrounding environment, the risk prediction results identified skidding and rear-end collision as the risk types, both at a medium risk level. Vehicle-road cooperative data was simultaneously used to obtain road segment weather conditions, confirming that the entire road segment experienced moderate rain and snow, with no sudden heavy rain or strong winds, providing a stable scenario basis for subsequent adjustments to the minimum risk strategy.

[0107] S1010: Initiating the sensor anti-interference processing flow: A deep learning-based rain removal algorithm is used on the data acquired by the visual camera to remove rain streaks and improve image clarity from 40% to 85%, ensuring the visual camera can accurately identify obstacles ahead. A distance compensation algorithm is used on the data acquired by the lidar, adjusting the detection distance from 90 mm to 108 mm based on the attenuation pattern of 15 mm / hour rainfall, with a correction error of 8%, ensuring accurate obstacle recognition. The sensor self-cleaning module (spray rain removal mode) is activated, with the spray pressure controlled at 0.3 MPa for 5 seconds to remove rainwater adhering to the sensor surface, ensuring unobstructed sensor detection. The lidar serves as the primary sensor for environmental perception and positioning, while the millimeter-wave radar serves as a backup sensor. The data effectiveness of the lidar is monitored in real time (currently 85%, higher than the preset switching threshold of 80%, no switching of the primary environmental perception and positioning sensor is required); the inertial navigation module is simultaneously activated for warm-up (3 seconds) to prepare for potential positioning fluctuations and ensure a rapid response from the positioning system.

[0108] S103: Increase the minimum risk strategy trigger threshold by 15% (corresponding to a higher level of severe weather). The original trigger method for the minimum risk strategy was "triggered when the obstacle distance is less than or equal to 50 meters". The adjusted trigger method is "triggered when the obstacle distance is less than or equal to 57.5 meters". At this time, the minimum risk strategy will be triggered when a stationary obstacle is detected 57.5m ahead.

[0109] The content of S104 is not special in this example and will not be repeated here.

[0110] S105: Dynamically Adjusted Minimum Risk Strategy Action Parameters (Adapted to Passenger Vehicle Characteristics): Adopting a "progressive graded deceleration" control mode, the deceleration is adjusted to 0.45g, corresponding to an actual deceleration of 4.41 m / s², with the deceleration time extended to 10 seconds. The vehicle speed is gradually reduced from 100 km / h to 36 km / h, avoiding the risk of rear-end collisions caused by sudden braking. The steering force control strategy is adjusted to a "lane centering control" mode, increasing the steering force from the conventional 15 Nm to 25 Nm to suppress vehicle body roll and keep the body roll angle within 3°, avoiding sharp lane changes (slippery roads in rain or snow can easily cause skidding). To achieve a smooth stop in the emergency lane, the emergency lane position is locked (emergency lane width 3.5 meters), and a smooth stopping trajectory is planned with a curvature of 0.04m. -1 This ensures a smooth and slip-free parking process.

[0111] S106: Executes a minimum risk strategy, collecting vehicle speed, road friction, and LiDAR perception data in real time at a sampling frequency of 10Hz. Through a dual feedback correction mechanism, the execution parameters of the minimum risk strategy are adjusted in real time: when the road friction further decreases to 60% of the normal friction (i.e., 0.34), the deceleration is corrected to 0.4g (3.92 m / s²), and the deceleration time is extended to 12 seconds to ensure a smooth deceleration process without sideslip. Simultaneously, a full-process log is maintained in real time. During execution, the driver does not issue a takeover command, and the minimum risk strategy continues to be executed until the vehicle smoothly stops in the emergency lane, the hazard warning lights are activated (flashing frequency 1Hz), and the vehicle comes to a complete stop (speed less than or equal to 0.5 km / h). The minimum risk strategy is then completed, and the stopping information (stopping location, stopping time, weather conditions) is simultaneously uploaded to the roadside management system via the vehicle-to-infrastructure network, completing the hazard avoidance control. The entire hazard avoidance process is free of any safety hazards.

[0112] Example 2: Minimum risk strategy adjustment under severe fog (extremely high weather severity). A Level 3 autonomous commercial vehicle (model: heavy-duty truck, weight 10,000 kg) was traveling on an urban expressway (road surface: cement, speed limit 80 km / h) at a speed of 70 km / h. The visibility sensor recorded a visibility of 50 m. The vision camera (resolution 1920×1080 pixels) could not identify obstacles more than 10 meters ahead, with image clarity less than 20%. The millimeter-wave radar detection signal was interfered with by heavy fog, resulting in large data fluctuations (data effectiveness 75%). The lidar (detection distance 150 meters) had its detection distance reduced to 40% of normal operating conditions (i.e., 60 meters), with a data effectiveness of 78%. According to the fusion of vehicle-road cooperative data, the road section was in heavy fog. There was a temporary construction barrier (1.8 meters high and 5 meters wide) 200 meters ahead, and the road surface friction decreased by 40% (from 0.75 under normal operating conditions to 0.45). The preset weather level assessment model determined the severe weather level to be extremely high, and the predicted risk types were collision and location loss, both of which were high risk levels.

[0113] S101+S102: Using a pre-set weather level assessment model, input parameters include rainfall of 0 mm / h, visibility of 50 m, wind speed of 3 m / s, and road surface friction of 0.45. Fuzzy comprehensive evaluation is used to quantify the scores (1 point for rainfall, 5 points for visibility, 1 point for wind speed, and 5 points for road surface friction, for a total of 12 points). After machine learning correction, the final severe weather level is determined to be extremely high. Simultaneously, combining the location information of the construction site barrier ahead and the vehicle speed of 70 km / h, the risk prediction results determine the risk types as collision and location loss, both with high risk levels. At the same time, the precise location information of the construction site barrier ahead (longitude and latitude errors less than or equal to 1 meter) is obtained through a vehicle-road cooperative network, providing accurate data support for the path planning of the minimum risk strategy, ensuring that the minimum risk strategy can avoid the construction site barrier.

[0114] S1010: Initiate the sensor anti-interference processing procedure: Employ a signal enhancement algorithm to the millimeter-wave radar sensing signal, filtering out scattered interference signals and increasing the data effectiveness from 75% to 82%, ensuring that the millimeter-wave radar can accurately identify the construction site enclosure ahead. Integrate 4D millimeter-wave point cloud SLAM and inertial navigation data, and use a Kalman filter algorithm to optimize positioning accuracy, achieving centimeter-level positioning in scenarios where satellite navigation signals are lost, avoiding the risk of positioning loss; activate the sensor self-cleaning module (heating defogging mode), with the heating temperature controlled at 50℃ for 10 seconds, to remove fog adhering to the sensor surface, avoiding the impact of fog adhering on sensing accuracy. Trigger the sensor redundancy switching procedure. Because the lidar data effectiveness (78%) is lower than the preset switching threshold of 80%, the main environmental perception and positioning sensor is automatically and seamlessly switched to the "millimeter-wave radar + inertial navigation module" backup combination (switching delay of 40 milliseconds, imperceptible), ensuring uninterrupted environmental perception data acquisition and guaranteeing the continuous execution of the minimum risk strategy.

[0115] S103+S104: The minimum risk strategy trigger threshold is increased by 35% (corresponding to an extremely high severe weather level). The original trigger method for the minimum risk strategy was "triggered when the obstacle distance is less than or equal to 60m," and the adjusted trigger method is "triggered when the obstacle distance is less than or equal to 81m." A new trigger method, "risk prediction trigger," is added. If a construction barrier is predicted ahead (high risk of collision), and the current vehicle speed is 70 km / h with low road friction, waiting for obstacle detection to trigger would not allow sufficient time for hazard avoidance. Therefore, the minimum risk strategy is triggered directly in advance without waiting for obstacle detection, allowing sufficient time (no less than 10 seconds) for hazard avoidance control. At the same time, the weather warning linkage process is activated to continuously monitor weather changes. If it is confirmed that the severe fog on this road section will continue for at least 30 minutes with no signs of weather improvement, the current minimum risk strategy adjustment status will be maintained.

[0116] S105: Dynamically Adjusted Minimum Risk Strategy Action Parameters (Adapted to Commercial Vehicle Characteristics): Adopting a "progressive graded deceleration" control mode, given the large weight and inertia of commercial vehicles, the deceleration is adjusted to 0.3g (2.94 m / s²), with the deceleration time extended to 15 seconds, gradually decreasing from the current speed of 70 km / h to 11 km / h to avoid the risk of vehicle rollover or rear-end collisions caused by sudden braking. The steering force control strategy is adjusted to a "small steering + lane centering" mode, controlling the vehicle's side deviation angle within 5°, slowly moving towards the emergency lane (4 meters wide) to avoid sharp turns (due to poor visibility in foggy weather, sharp turns easily cause collisions); combining 4D millimeter-wave point cloud SLAM and inertial navigation positioning data, a precise parking trajectory is planned, with a parking trajectory curvature of 0.03m. -1 Avoid construction barriers ahead (leave a safe distance of ≥10 meters), locate a safe stopping point on the emergency lane, and ensure that the stopping position is accurate.

[0117] S106: Implements a minimum risk strategy, collecting real-time data on vehicle speed, steering angle, positioning, and changes in the location of construction barriers at a sampling frequency of 10Hz. Through a dual feedback correction mechanism combining PID algorithm and machine learning, it adjusts steering force and deceleration in real time: when the vehicle approaches the edge of the emergency lane (less than or equal to 0.5m from the edge), the steering force is adjusted to reduce the steering angle to within 2° to ensure a smooth stop. When the road surface friction is detected to further decrease to 0.4, the deceleration is corrected to 0.28g (2.74 m / s²) to ensure a smooth deceleration process. Meanwhile, encrypted logs of the entire process are stored in real time for accident liability tracing. During the execution, the driver issues a takeover command (triggered by the steering wheel button), and the system enters the takeover transition period, retaining the assisted steering and deceleration functions. The assisted steering force is maintained at 20 Nm, and the assisted deceleration is maintained at 0.28g until the driver's operation is stable (no abnormal operation for 3 seconds, and the vehicle speed is controlled within 10 km / h). Then, the system completely exits the minimum risk strategy control mode and completes the avoidance control. The entire avoidance process is smooth and orderly, and no collision or skidding occurs.

[0118] Example 3: Adaptive adjustment of minimum risk strategy under strong winds (high severity weather level) When an L3 autonomous passenger vehicle (compact sedan, 1200 kg) was traveling on a highway (asphalt road surface, speed limit 120 km / h) at a speed of 110 km / h, the wind speed sensor detected a wind speed of 18 m / s (high risk, corresponding to level 7 wind). The vehicle attitude sensor detected a slight side-slip phenomenon with a side-slip angle of 3°. The visual camera image showed no significant abnormalities, with image clarity ≥90%. The LiDAR data had an effectiveness rate of 88%, and the millimeter-wave radar data had an effectiveness rate of 90%. According to the fusion of vehicle-road cooperative data, the road section was in strong wind weather, with some lanes having crosswind warnings (crosswind speeds up to 20 m / s). There were no obstacles 500m ahead, and the road surface friction was 0.7 (normal operating conditions). The preset weather level assessment model determined the severe weather level to be high, and the predicted risk types were side-slip and loss of control, both at a medium risk level.

[0119] S101+S102: Using a pre-set weather level assessment model, input parameters include rainfall of 0 mm / h, visibility of 200m, wind speed of 18 m / s, and road surface friction of 0.7. A fuzzy comprehensive evaluation method is used to quantify the scores (rainfall 1 point, visibility 1 point, wind speed 5 points, road surface friction 2 points, total 9 points). After machine learning correction, the final severe weather level is determined to be high risk. Simultaneously, combined with real-time vehicle body sideslip angle of 3° and vehicle speed of 110 km / h, the predicted risk type is sideslip and loss of control, both at a medium risk level. Simultaneously, vehicle-road cooperative data is used to determine the crosswind distribution information of the road segment, confirming that the high crosswind area is located 500m to the right of the current lane, with crosswind speeds reaching 20 m / s, providing accurate crosswind information for the minimum risk strategy's route planning.

[0120] S1010: Initiating the sensor anti-interference processing flow: A fusion filtering algorithm (Kalman filter algorithm) is used on the attitude sensor's perceived data to eliminate abnormal data caused by wind resistance, improving data stability from 82% to 95%, enhancing attitude detection accuracy, and ensuring accurate capture of changes in the vehicle's side slip angle. The sensor attitude dynamic calibration module is activated to adjust the installation attitude of the LiDAR and vision camera in real time, with an adjustment angle range of ±2°, reducing the impact of wind resistance on the sensor's installation attitude and ensuring the accuracy of the sensor's detection angle. The LiDAR serves as the primary sensor, and the millimeter-wave radar as a backup sensor, with real-time monitoring data effectiveness (both above 80%), eliminating the need for redundant switching; inertial navigation data is simultaneously fused to assist in vehicle attitude adjustment, controlling the detection error of the vehicle's side slip angle within ±0.5°, ensuring timely capture of side slip changes.

[0121] S103: The trigger threshold for the minimum risk strategy is reduced by 25% (corresponding to a high level of severe weather). The original trigger method for the minimum risk strategy was "trigger with a side slip angle ≥ 5°", and the adjusted trigger method is "trigger with a side slip angle ≥ 3.75°". The trigger method is specifically set to "trigger for abnormal vehicle posture". When the side slip angle of the vehicle exceeds the threshold (e.g., 3.75°), the minimum risk strategy is immediately triggered. At the same time, wind speed changes are continuously monitored at a sampling frequency of 10Hz to predict the trend of crosswind risk changes. It is confirmed that the wind speed in the high crosswind area will remain stable at around 20 m / s without any further increase, providing a basis for adjusting the execution parameters of the minimum risk strategy.

[0122] The content of S104 is not special in this example and will not be repeated here.

[0123] S105: Dynamically Adjusted Minimum Risk Strategy Action Parameters (Adapted to Passenger Vehicle Characteristics): Adopting a "Lane Centering Enhancement + Small Deceleration" control mode, considering that significant deceleration in strong winds can easily lead to vehicle imbalance, the deceleration is adjusted to 0.4g (3.92 m / s²), and the deceleration time is extended to 8 seconds, gradually decreasing from the current speed of 110 km / h to 78 km / h to ensure a smooth and unbalanced deceleration process. The steering force adjustment strategy is a "Slip Suppression Mode," adjusting the steering force in real time according to the vehicle's posture. When the body slip angle increases to 3.5°, the steering force is increased to 30 Nm to suppress body slip, keeping the body slip angle within 3° and maintaining lane centering. It avoids high crosswind areas, plans a smooth driving trajectory, and controls the trajectory curvature to be below 0.035 m. -1 We will not pull over to the side of the road for the time being. In strong winds, pulling over to the side of the road may cause the vehicle to lose control due to crosswinds. We will wait until the wind speed decreases before pulling over to ensure safety.

[0124] S106: Executes a minimum risk strategy, collecting vehicle attitude, wind speed, and steering angle in real time at a sampling frequency of 10 Hz. Through a dual feedback correction mechanism of PID algorithm and machine learning, it adjusts steering force parameters and deceleration in real time: When the vehicle's sideslip angle increases to 4°, it immediately increases the counter-steering force to 35 Nm to quickly suppress sideslip and bring the sideslip angle back to within 3°. When the wind speed decreases to 12 m / s (general risk, corresponding to level 6 wind), it adjusts the minimum risk strategy execution parameters, gradually executing a pullover maneuver. The deceleration is adjusted to 0.45g (4.41 m / s²), and the deceleration time is extended to 6 seconds, gradually decreasing from 78 km / h to 0 km / h, smoothly stopping in the emergency lane (emergency lane width 3.5m). The entire process log is stored in real time, and the log content fully records the parameter changes throughout the entire hazard avoidance process; after execution, the hazard avoidance control is completed by turning on the hazard warning lights (flashing frequency 1 Hz). No loss of vehicle control occurred during the entire hazard avoidance process, which fully demonstrates the adaptability and safety of the invention in strong wind weather.

[0125] In this embodiment, compared with the prior art, the method adopts a triple multi-sensor anti-interference design of "hardware protection + algorithm optimization + redundancy backup". It combines vehicle-road cooperative data with 4D millimeter-wave point cloud SLAM fusion positioning technology to comprehensively solve the technical problems of sensor performance degradation, data misjudgment and omission, and positioning loss under severe weather conditions. It significantly improves the accuracy and continuity of environmental perception and ensures the accuracy of the minimum risk strategy trigger condition determination. At the same time, it effectively avoids the core defects of existing patents that rely on a single sensor and lack systematic anti-interference design, makes up for the physical perception blind spots of vehicle sensors, enables vehicles to obtain global environmental information, improves the reliability of vehicle perception system in extreme environments, increases sensor data effectiveness to over 90%, and improves positioning accuracy to the centimeter level.

[0126] This method, in line with industry standards, adopts a four-level weather classification system, with each level corresponding to specific quantitative indicators. This enables dynamic iterative adjustment of the minimum risk strategy trigger threshold and execution parameters. It provides differentiated risk avoidance control actions for different types and levels of severe weather, overcoming the technical limitations of existing minimum risk strategies, such as fixed action parameters, coarse weather classification, and inability to adapt to severe weather. Furthermore, this method is fully compliant with the regulatory requirements for Level 3 autonomous driving, ensuring it meets the compliance conditions for commercial application and promoting the deployment of Level 3 autonomous driving in extreme environments.

[0127] This method significantly improves the response speed (no more than 200 milliseconds) and execution accuracy of the minimum risk strategy through a closed-loop control design of "prediction-execution-feedback-correction," a minimum risk strategy action library, and machine learning optimization technology, ensuring real-time matching of avoidance actions with environmental changes and vehicle driving status. The sensor redundancy backup design effectively avoids the safety hazard of minimum risk strategy execution interruption, with a switching delay of no more than 50 milliseconds, ensuring the continuous and stable execution of the minimum risk strategy. The dual feedback correction mechanism further enhances the rationality of the minimum risk strategy, effectively reducing the risk of secondary accidents such as rear-end collisions and skidding, reducing the accident rate by more than 40%, and significantly improving the safety of L3 autonomous driving in adverse weather conditions.

[0128] This method integrates meteorological early warning linkage and full-process log retention. The meteorological early warning linkage enables advance prediction and preparation for minimum-risk strategies, allowing sufficient time for hazard avoidance and mitigation, thus avoiding untimely hazard response due to sudden severe weather. Full-process log retention meets the requirements for liability traceability and compliance auditing for Level 3 autonomous driving. Furthermore, this method can be directly integrated into existing Level 3 autonomous driving systems without large-scale hardware architecture modifications, requiring only software algorithm optimization and the addition of a small number of sensor auxiliary modules. This reduces the cost of technology implementation, making it highly practical and adaptable to various Level 3 passenger and commercial vehicles. Its application scenarios cover core driving scenarios such as highways, urban expressways, and tunnels, demonstrating a wide range of applicability and high commercial application value.

[0129] This method, through the design of auxiliary control during the driver takeover transition period, provides the driver with a certain buffer time to take over, retaining assisted steering and deceleration functions until the driver's operation stabilizes and completely exits the minimum risk strategy control. This improves the safety and smoothness of human-machine interaction and avoids operational errors caused by sudden changes in vehicle state during driver takeover. Simultaneously, it fills the research gap in the field of cooperative control during the takeover transition period, further improving the overall safety level of L3 autonomous driving in adverse weather conditions and promoting the further improvement and upgrading of L3 autonomous driving technology.

[0130] Example 2 Based on the same inventive concept, embodiments of the present invention also provide a minimum risk strategy adjustment device for autonomous vehicles based on severe weather, referring to... Figure 2 As shown, the device includes: The first determining module 101 is used to determine the type and level of severe weather based on on-board multi-sensor data and vehicle-road cooperative data; The second determining module 102 is used to determine the risk prediction result based on the severe weather level, vehicle surrounding environment data and vehicle driving status; The first adjustment module 103 is used to adjust the trigger threshold of the minimum risk strategy based on the severe weather level. The second adjustment module 104 is used to increase the triggering method of the minimum risk strategy based on the risk prediction result and the data effectiveness of the vehicle multi-sensor combination. The third adjustment module 105 is used to dynamically adjust the execution parameters of the minimum risk strategy based on the type of severe weather, the level of severe weather, and the risk prediction result.

[0131] Example 3 Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather as described in Embodiment 1 above.

[0132] Example 4 Based on the same inventive concept, embodiments of the present invention also provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather as described in Embodiment 1 above.

[0133] Example 5 Based on the same inventive concept, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, it implements the method for adjusting the minimum risk strategy of autonomous vehicles based on severe weather as described in Embodiment 1 above.

[0134] Example 6 Based on the same inventive concept, this embodiment of the invention also provides a minimum risk strategy adjustment device for autonomous vehicles based on severe weather, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, it implements the minimum risk strategy adjustment method for autonomous vehicles based on severe weather as described in Embodiment 1 above.

[0135] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0136] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0137] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0139] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for adjusting the minimum risk strategy of an autonomous vehicle based on severe weather, characterized in that, include: Based on data from multiple vehicle sensors and vehicle-road cooperative data, the type and level of severe weather are determined. Based on the aforementioned severe weather level, vehicle surrounding environment data, and vehicle driving status, the risk prediction result is determined; Based on the severity of the severe weather, adjust the trigger threshold of the minimum risk strategy; Based on the risk prediction results and the data effectiveness of the vehicle multi-sensor combination, the triggering method of the minimum risk strategy is increased; The execution parameters of the minimum risk strategy are dynamically adjusted based on the type of severe weather, the level of severe weather, and the risk prediction results.

2. The method of claim 1, wherein, The adjustment of the trigger threshold for the minimum risk strategy based on the severe weather level includes: Based on the severity of the weather, determine the preset threshold adjustment range; Depending on the type of the trigger threshold, the trigger threshold of the minimum risk strategy is lowered or raised according to the preset threshold adjustment range.

3. The method of claim 1, wherein, The risk prediction results include multiple risk types and the corresponding risk level for each risk type; the risk levels include high, medium, and low; the vehicle-mounted multi-sensor combination includes lidar and millimeter-wave radar; The triggering methods for increasing the minimum risk strategy based on the risk prediction results and the data effectiveness of the combined vehicle multi-sensor system include: If the data effectiveness of both the lidar and the millimeter-wave radar is lower than the preset sensor failure threshold, the minimum risk strategy is triggered. If any risk type in the risk prediction results corresponds to a high risk level, then the minimum risk strategy is triggered.

4. The method of claim 3, wherein, The data efficiency of the lidar or the millimeter-wave radar is determined by the following method: The data effectiveness is determined based on the data integrity, data accuracy, and data stability of the lidar or the millimeter-wave radar.

5. The method of claim 1, wherein, The step of dynamically adjusting the execution parameters of the minimum risk strategy based on the severe weather type, the severe weather level, and the risk prediction result includes: If the severe weather type is rain, snow, or fog, then a preset deceleration range is determined based on the severe weather level and the vehicle weight. Based on the road friction, the vehicle is controlled in real time to decelerate within the preset deceleration range, so that the vehicle decelerates smoothly. If the severe weather type is rain, snow, or strong wind, then increase the steering damping; If the severe weather type is rain, snow, or strong wind, the steering force will be adjusted in real time according to the vehicle's side angle. If the severe weather type is heavy fog or satellite navigation signal loss, the inertial navigation module and high-precision map data are integrated to plan a parking trajectory to the emergency lane position, and the curvature of the parking trajectory does not exceed a preset curvature threshold. If the risk type in the risk prediction result is a rear-end collision, then the deceleration operation will be performed first; If the risk type in the risk prediction result is sideslip, then lane centering control will be implemented first. If the risk type in the risk prediction result is location loss, then location calibration should be performed first, and then the docking operation should be performed.

6. The method of claim 1, wherein, Before determining the type and level of severe weather based on onboard multi-sensor data and vehicle-road cooperative data, the following steps are also included: The data is collected based on a combination of vehicle-mounted sensors; wherein, the combination of vehicle-mounted sensors includes multiple of the following: lidar, millimeter-wave radar, vision camera, rain sensor, visibility sensor, temperature and humidity sensor, and wind speed sensor.

7. The method according to claim 6, characterized in that, After determining the type and level of severe weather based on onboard multi-sensor data and vehicle-road cooperative data, the following steps are also included: If the data effectiveness of the lidar is lower than a preset switching threshold, the main sensor in the vehicle used for environmental perception and positioning will be switched from the lidar to the millimeter-wave radar and inertial navigation module.

8. The method according to claim 6, characterized in that, After determining the type and level of severe weather based on onboard multi-sensor data and vehicle-road cooperative data, the following steps are also included: Based on the severe weather type, perform hardware self-cleaning or dynamic calibration on the lidar, the millimeter-wave radar, and the vision camera; The data collected by the lidar, millimeter-wave radar, and visual camera are enhanced using the enhancement algorithm corresponding to the severe weather type.

9. A minimum risk strategy adjustment device for autonomous vehicles based on severe weather, characterized in that, include: The first determination module is used to determine the type and level of severe weather based on data from onboard multi-sensor systems and vehicle-road cooperative data. The second determining module is used to determine the risk prediction result based on the severe weather level, vehicle surrounding environment data and vehicle driving status; The first adjustment module is used to adjust the trigger threshold of the minimum risk strategy based on the severe weather level. The second adjustment module is used to increase the triggering method of the minimum risk strategy based on the risk prediction results and the data effectiveness of the vehicle multi-sensor combination. The third adjustment module is used to dynamically adjust the execution parameters of the minimum risk strategy based on the severe weather type, the severe weather level, and the risk prediction result.

10. A device for adjusting the minimum risk strategy of an autonomous vehicle based on severe weather, comprising a memory and a processor, characterized in that, The memory stores computer program instructions that can be executed by the processor, and when the processor executes the computer program instructions, it implements the steps of the method as described in any one of claims 1 to 8.

Images