A vehicle control method, device, electronic equipment and vehicle

Abstract

Embodiments of the present application provide a vehicle control method and device, electronic equipment and vehicle, which obtain sensor physical state information and perception data of multiple sensors, determine a data quality index of the perception data, determine a health score of each sensor in the multiple sensors based on the sensor physical state information of the multiple sensors and the data quality index, calculate a system-level health state according to the health score of each sensor in the multiple sensors, and generate a vehicle control strategy based on a mapping relationship between the system-level health state and a safety boundary parameter to control the vehicle, thereby realizing cascading quantitative evaluation of the perception attenuation degree from a single component to a system as a whole, and dynamically adjusting the physical motion boundary of the vehicle based on the system overall health state, so as to effectively avoid the safety risks caused by blind driving or sudden exit of the autonomous driving system under the condition of environmental abnormalities or hardware degradation.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, and in particular to a vehicle control method, a vehicle control device, an electronic device, and a computer-readable storage medium. Background Technology

[0002] With the development of autonomous driving technology, the perception of the external environment by vehicles relies heavily on the collaborative work of multiple sensors, such as images, point clouds, and electromagnetic sensors. However, during actual vehicle operation, the perception performance of each sensor often degrades or deteriorates to varying degrees due to adverse weather conditions (such as rain, snow, fog, and mud) or hardware factors (such as aging of electronic components, severe temperature drift, and signal interference), seriously threatening the safety and reliability of autonomous driving.

[0003] First, the relevant technologies lack refined quantitative methods for assessing the multidimensional degradation state of sensors. The hardware diagnostic mechanisms of these technologies tend to rely on a binary, crude judgment of "normal operation" or "complete failure." When the signal quality of a sensor is "partially damaged" due to environmental pollution or changes in its microscopic physical state, the system struggles to quantitatively assess the actual health status of individual components in real time and continuously.

[0004] Furthermore, because it is impossible to accurately assess the macroscopic impact of damage to individual heterogeneous sensors on the overall vehicle perception system, autonomous driving or assisted driving systems often cannot provide a global confidence assessment when faced with hardware performance degradation. This leads to two extremes in vehicle control strategies: first, blindly maintaining the original high-risk control logic even when the perception part is damaged, creating potential collision hazards; second, adopting a "one-size-fits-all" strategy to directly report errors and forcibly disengage autonomous driving, causing sudden system malfunctions and easily triggering safety accidents where the driver cannot take over in time.

[0005] The present invention provides a vehicle control method, apparatus, electronic device, and computer-readable storage medium to overcome or at least partially solve the above-mentioned problems.

[0006] This invention discloses a vehicle control method, comprising: Acquire sensor physical state information and sensing data from multiple sensors; Determine the data quality indicators of the perceived data; Based on the physical state information of the sensors from multiple sensors and the data quality indicators, a health score is determined for each of the multiple sensors. The system-level health status is calculated based on the health score of each of the multiple sensors, and a vehicle control strategy is generated to control the vehicle based on the mapping relationship between the system-level health status and the safety boundary parameters.

[0007] Optionally, the sensor type includes at least an image sensor, a point cloud sensor, an electromagnetic or acoustic wave sensor, and an inertial measurement unit, and the step of determining the data quality indicators of the sensed data includes: The local contrast and high-frequency energy of the sensor data extracted by the image sensor are obtained. Obtain the point cloud density and reflection intensity of the sensing data extracted by the point cloud sensor. The signal-to-noise ratio and echo attenuation rate of the sensing data from the electromagnetic or acoustic wave sensor are obtained. Obtain the zero-bias drift of the sensing data from the inertial measurement unit; The local contrast and high-frequency energy, the point cloud density and reflection intensity, the signal-to-noise ratio and echo attenuation rate, and the zero-bias drift are determined as the data quality indicators of the corresponding sensors.

[0008] Optionally, determining the health score of each of the multiple sensors based on the sensor physical state information and the data quality indicators includes: Obtain a state baseline performance model that reflects the sensing capability of the sensor under normal conditions; Obtain the vehicle interior dew point temperature, and calculate the sum of the vehicle interior dew point temperature and the preset safety margin to determine the low temperature threshold; Identify abnormal areas where the temperature of the sensor is below the low-temperature threshold; The proportion of the frosted area is calculated by comparing the area of ​​the abnormal region with the total area of ​​the sensor. Generate an average temperature gradient that reflects the temperature difference from the edge to the center of the abnormal region; Based on the sensor type, the signal quality characteristic parameters corresponding to the sensor are extracted from the data quality indicators; The signal quality characteristic parameters are analyzed using the state benchmark performance model to generate a performance attenuation coefficient that characterizes the residual proportion of the sensor's sensing capability. The area ratio of the frosted region and the average temperature gradient are used as exponential decay terms by using a nonlinear mapping function, and combined with the performance decay coefficient as a multiplicative correction term for fusion calculation to obtain a health score for expressing the sensor.

[0009] Optionally, it also includes: Based on the health score, the confidence weights or detection decision thresholds of the multiple sensors in the perception fusion are dynamically adjusted to generate adaptive perception results. The step of dynamically adjusting the confidence weights or detection decision thresholds of the multiple sensors in perception fusion based on the health score to generate adaptive perception results includes: The observation noise covariance matrix in the fusion filter is scaled and adjusted based on the health score to reduce or increase the observation weight of the corresponding sensor in state estimation. The confidence threshold of the target detection algorithm is increased based on the health score, and low-confidence detection results are filtered out. Based on the adjusted observation weights and filtered detection results, multi-sensor data association and state updates are performed to generate adaptive fusion perception results.

[0010] Optionally, the step of controlling the vehicle based on the control strategy includes: The dynamic observation weights and health scores of multiple sensors are aggregated, and a system confidence index reflecting the overall reliability of system environment modeling is generated through weighted average calculation. The system confidence index is input into a preset safety constraint mapping function to calculate and generate dynamic safety boundary parameters including the maximum driving speed limit, target following distance, and lane change permission instructions. Extract the spatiotemporal attribute information of each obstacle from the adaptive fusion perception results; Combining the speed and spacing constraints set by the dynamic safety boundary parameters, a local planning trajectory that meets the safety margin is generated through spatiotemporal joint optimization; The system confidence index is compared with preset multi-level risk thresholds in real time, and the vehicle is controlled based on the comparison results.

[0011] Optionally, the step of controlling the vehicle based on the comparison result includes: When the system confidence index is determined to be higher than the first risk threshold, longitudinal acceleration control commands and lateral turning control commands are generated based on the local planning trajectory for the vehicle's power output and direction control actuators.

[0012] Optionally, it also includes: When the system confidence index is determined to be lower than the second risk threshold and no driver takeover signal is received within a preset time, the system uses the lane topology and side obstacle distribution information in the adaptive fusion perception results to generate a graded minimum risk strategy for the vehicle's braking and steering control actuators, and controls the vehicle to decelerate and stop at the target position based on the graded minimum risk strategy.

[0013] This invention also discloses a vehicle control device, comprising: The data acquisition module is used to acquire sensor physical state information and sensing data from multiple sensors; A data quality index determination module is used to determine the data quality index of the perceived data; A health score conversion module is used to determine the health score of each of the multiple sensors based on the sensor physical state information and the data quality indicators. The vehicle control module is used to calculate the system-level health status based on the health score of each of the plurality of sensors, and to generate a vehicle control strategy to control the vehicle based on the mapping relationship between the system-level health status and the safety boundary parameters.

[0014] This invention also discloses an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes a program stored in the memory, it implements the method described in the embodiments of the present invention.

[0015] This invention also discloses a vehicle that includes the electronic equipment described above.

[0016] This invention also discloses a computer-readable storage medium storing instructions that, when executed by one or more processors, cause the processors to perform the methods described in this invention.

[0017] The embodiments of the present invention have the following advantages: This invention, through a combination of the hardware physical state and underlying data quality of various heterogeneous sensors, achieves a cascaded quantitative assessment of the degree of perception attenuation from individual components to the entire system. Based on the overall health of the system, it dynamically adjusts the physical motion boundary of the vehicle, effectively avoiding the safety risks caused by blind driving or sudden disengagement of the autonomous driving system under abnormal environmental conditions or hardware degradation. Attached Figure Description

[0018] Figure 1 This is a flowchart of the steps of a vehicle control method provided in an embodiment of the present invention; Figure 2 This is a flowchart of another vehicle control method provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a vehicle control device provided in an embodiment of the present invention; Figure 4 This is a hardware structure block diagram of an electronic device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of a computer-readable medium provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Reference Figure 1 The diagram illustrates a flowchart of a vehicle control method provided in an embodiment of the present invention, which may specifically include the following steps: Step 101: Obtain sensor physical state information and sensing data from multiple sensors; Step 102: Determine the data quality indicators of the perceived data; Step 103: Based on the sensor physical state information of multiple sensors and the data quality index, determine the health score of each of the multiple sensors; Step 104: Calculate the system-level health status based on the health score of each of the multiple sensors, and generate a vehicle control strategy to control the vehicle based on the mapping relationship between the system-level health status and the safety boundary parameters.

[0021] In a specific implementation, the embodiments of the present invention can acquire sensor physical state information and sensing data from multiple sensors to simultaneously collect the external physical environment characteristics and internal sensing information of the sensors, providing basic raw data for subsequent cross-verification of the degree of frost.

[0022] Sensors refer to hardware devices installed on vehicles to detect the external environment, such as cameras, lidar, and millimeter-wave radar.

[0023] Sensor physical state information refers to non-sensory parameters that reflect the sensor hardware itself and its surrounding physical environment, such as the infrared thermal image temperature of the sensor surface, the real-time temperature matrix of the lens area, or the ambient dew point temperature.

[0024] Perception data refers to the raw signal stream collected by sensors during normal operation and used for environmental modeling, such as image / video frames output by cameras, point cloud data output by lidar, or a list of detected targets output by millimeter-wave radar.

[0025] In specific implementations, embodiments of the present invention can determine the data quality indicators of the sensed data in order to extract feature quantities that can reflect the interference of the signal from the data level, and preliminarily quantify the "degree of damage" of the sensed information.

[0026] Data quality metrics are quantitative parameters used to measure the clarity, integrity, and reliability of perceived signals, such as the local contrast and high-frequency energy ratio of an image, or the density and reflection intensity of point cloud data.

[0027] In a specific implementation, embodiments of the present invention can determine the health score of each of the multiple sensors based on the sensor physical state information and the data quality index of the multiple sensors, so as to calculate the hardware physical anomalies (external factors) and data quality loss (internal factors) of a single sensor and generate a standardized scalar score, which can be used to accurately quantify the current actual functional integrity of each component.

[0028] A health score can be a normalized value (such as a continuous scalar between 0 and 1) calculated for a single sensor. It is used to intuitively express the residual ratio between the effective range, accuracy, or sensitivity of the current sensor and its standard optimal state. The lower the value, the more severely the hardware function is impaired.

[0029] For example, when the system faces two independent hardware components, the "forward-facing camera" and the "left-side lidar," it calls their respective nonlinear quantization functions in parallel and without overlap: 1. Independent quantization for the "Front image sensor (Camera_Front)"; The input physical state information is that the lens infrared temperature measurement is -5°C. After image analysis, it is determined that the area of ​​the abnormally low temperature region (probable frosting region) at the edge of the lens accounts for 35%, and the average temperature gradient from the edge to the center of this region is ▽T=1.2°C / mm.

[0030] The input data quality indicators are that the local contrast of the extracted image has decreased by 30% compared to the baseline, and the high-frequency energy has decreased sharply. The current performance degradation coefficient K is calculated to be 0.70.

[0031] Nonlinear quantization mapping calculation is the system directly converting these three elements belonging to the Camera. Front The self-developed indicators are substituted into its proprietary nonlinear exponential decay formula for closed-door calculation: Hs_camera=e^-(a·S+b▽T)·K Through a mapping link of pure mathematical functions, the independent health score of the front-facing camera, Hs_camera=0.45 (representing severe functional impairment), is directly output.

[0032] 2. Independent quantization for the "left-side point cloud sensor (LiDAR_Left)"; Input physical state information: The base temperature of the outer protective cover of the lidar is 0°C, the area of ​​the local abnormal low temperature zone is S=10%, and the average temperature gradient is ▽T=0.3°C / mm.

[0033] The input data quality indicators show that the real-time point cloud density is basically normal, with only slight fluctuations in reflection intensity. The performance degradation coefficient K = 0.95 was calculated by comparing with the state baseline model.

[0034] Nonlinear quantization mapping calculation is a system that does not reference camera data at all, but only inputs the parameters of the LiDAR itself into the corresponding quantization model: Hs_lidar=e^-(a·Sb·▽T)·K The final mapping yields the independent health score of the left-side lidar: Hs_lidar=0.92 (representing basic functional health).

[0035] In a specific implementation, embodiments of the present invention can calculate the system-level health status based on the health score of each of the multiple sensors, and generate a vehicle control strategy to control the vehicle based on the mapping relationship between the system-level health status and the safety boundary parameters. This will refine the scattered component-level health status into a global system-level assessment and establish an adaptive mapping from perception capability to the vehicle's physical motion limits, ensuring that the system can actively tighten the safety boundary when the perception board is damaged, thereby achieving smooth degraded control of the vehicle.

[0036] System-level health status is a global indicator that reflects the overall reliability of the vehicle environment modeling and the trust level of the vehicle perception system, calculated by integrating the health scores of all heterogeneous sensors in the vehicle and their respective topological weights in the system.

[0037] Safety boundary parameters are a set of thresholds that limit the physical movement safety limits of a vehicle, including but not limited to the maximum permissible driving speed limit under the current perception level, the minimum target following distance that must be maintained, and lane change permission instructions that allow lane change behavior.

[0038] The mapping relationship is a mathematical function or logical comparison matrix preset in the controller, which is used to determine the physical quantification degree to which the safety boundary parameters should be tightened for each step of the system-level health state (for example, if the overall credibility is perceived to decrease, the maximum speed limit will be reduced proportionally and the following distance will be increased).

[0039] The vehicle control strategy is a longitudinal acceleration / deceleration and lateral steering planning trajectory and control command generated by the decision planning layer in combination with current environmental obstacle information and under the constraints of safety boundary parameters.

[0040] This invention, by integrating the hardware physical state and underlying data quality of various heterogeneous sensors, achieves a cascaded quantitative assessment of the degree of perception attenuation from individual components to the entire system. Based on the overall health of the system, it dynamically adjusts the physical motion boundary of the vehicle, effectively avoiding the safety risks caused by blind driving or sudden disengagement of the autonomous driving system under abnormal environmental conditions or hardware degradation.

[0041] Based on the above embodiments, modified embodiments of the above embodiments are proposed. It should be noted that, in order to keep the description brief, only the differences from the above embodiments are described in the modified embodiments.

[0042] In an optional embodiment of the present invention, the sensor type includes at least an image sensor, a point cloud sensor, an electromagnetic or acoustic wave sensor, and an inertial measurement unit, and the step of determining the data quality index of the sensed data includes: The local contrast and high-frequency energy of the sensor data extracted by the image sensor are obtained. Obtain the point cloud density and reflection intensity of the sensing data extracted by the point cloud sensor. The signal-to-noise ratio and echo attenuation rate of the sensing data from the electromagnetic or acoustic wave sensor are obtained. Obtain the zero-bias drift of the sensing data from the inertial measurement unit; The local contrast and high-frequency energy, the point cloud density and reflection intensity, the signal-to-noise ratio and echo attenuation rate, and the zero-bias drift are determined as the data quality indicators of the corresponding sensors.

[0043] In a specific implementation, embodiments of the present invention can obtain the local contrast and high-frequency energy of the perception data extracted by the image sensor in order to extract key features that can reflect the impairment of visual clarity, and use them to quantify the degree of optical obstruction or blurring caused by frost in the camera.

[0044] An image sensor is a device that converts optical images into electronic signals; in this solution, it mainly refers to an in-vehicle camera.

[0045] Local contrast refers to the degree of difference in pixel brightness within a local area of ​​an image; the physical occlusion caused by frosting will scatter light, making the image appear "hazy" and thus significantly reducing local contrast.

[0046] High-frequency energy is the high-frequency signal component in an image that represents edges, details, and textures; frosting causes image edges to soften (out-of-focus effect), and the proportion of high-frequency energy extracted by Fourier transform and other methods will decrease due to frosting.

[0047] In a specific implementation, embodiments of the present invention can obtain the point cloud density and reflection intensity of the sensing data extracted by the point cloud sensor in order to quantify the physical loss of the laser signal when penetrating the frost layer, thereby evaluating the attenuation of the lidar in obstacle detection accuracy.

[0048] A point cloud sensor is a device that acquires three-dimensional spatial information by emitting laser beams and receiving reflected signals; it is also known as lidar (LiDAR).

[0049] Point cloud density is the number of effective detection points per unit area or space. When the sensor surface is frosted, some laser beams cannot penetrate the frost layer or are severely deflected, resulting in a reduction in the number of effective points returned.

[0050] Reflection intensity is the energy intensity of the laser pulse echo received by the sensor; frost will absorb and scatter the laser energy, causing the energy of the returned signal from a target with high reflectivity to be greatly attenuated.

[0051] In specific implementations, embodiments of the present invention can obtain the signal-to-noise ratio and echo attenuation rate of the sensing data of the electromagnetic or acoustic wave sensor to evaluate the gain interference of the frost / ice layer on the transmission and reception of wave signals, and quantify the detection sensitivity of radar and ultrasonic equipment.

[0052] Electromagnetic or acoustic wave sensors are devices that use electromagnetic waves or ultrasonic waves to measure distance and speed, such as millimeter-wave radar and ultrasonic radar.

[0053] Signal-to-noise ratio (SNR) is the ratio of the intensity of the normal target echo signal to the intensity of the background noise. Frosting introduces additional environmental clutter and reduces the effective signal strength, resulting in a decrease in SNR.

[0054] The echo attenuation rate is the proportion of energy lost during the transmission and return of a signal. A thick layer of frost will change the dielectric impedance, causing the transmitted pulse to attenuate at the moment of transmission, or the echo to be absorbed by the frost layer.

[0055] In a specific implementation, the embodiments of the present invention can obtain the zero-bias drift of the sensing data of the inertial measurement unit in order to assess the accuracy of the vehicle's own motion attitude assessment by monitoring the impact of the low temperature environment accompanied by frost on the internal electronic characteristics of the sensor.

[0056] An inertial measurement unit (IMU) is a device that measures the three-axis attitude angles (or angular rates) and acceleration of an object.

[0057] Zero drift is the deviation value of the sensor output signal when it is stationary and changes with temperature and time. Frosting is usually accompanied by drastic environmental temperature differences, which can cause slight deformation of the internal structure of the IMU, resulting in zero drift and affecting positioning accuracy.

[0058] In specific implementations, embodiments of the present invention can determine the local contrast and high-frequency energy, the point cloud density and reflection intensity, the signal-to-noise ratio and echo attenuation rate, and the zero-bias drift as data quality indicators, so as to summarize the underlying signal features scattered in various dimensions into a unified evaluation system, and provide standard input for subsequent calculation of cross-modal sensor health scores.

[0059] Data quality indicators are a set of multi-dimensional vector parameters that represent the evaluation criteria for the signal effectiveness of each sensing channel of the whole vehicle perception system under the current frosting condition.

[0060] This invention, through customized extraction of quality indicators for sensors with different underlying detection principles, achieves multi-dimensional and precise monitoring of the degree of impact of frost on the heterogeneous perception system of the entire vehicle, ensuring that the system's diagnosis of perception failure can penetrate to the signal feature level.

[0061] In practical applications, "perception fusion" (also known as "multi-sensor fusion" in the fields of autonomous driving or robotics) simply means stitching together data collected by multiple different types of sensors on a vehicle, processing it through algorithms, and combining it into a more accurate and comprehensive global environmental profile than any single sensor.

[0062] In an optional embodiment of the present invention, the method further includes: Based on the health score, the confidence weights or detection decision thresholds of the multiple sensors in perception fusion are dynamically adjusted to generate adaptive perception results, so as to realize the dynamic self-healing and adaptive fault tolerance of the system at the perception level and break the direct transmission chain between hardware degradation and global perception deterioration.

[0063] Optionally, the step of dynamically adjusting the confidence weights or detection decision thresholds of the multiple sensors in perception fusion based on the health score to generate adaptive perception results includes: The observation noise covariance matrix in the fusion filter is scaled and adjusted based on the health score to reduce or increase the observation weight of the corresponding sensor in state estimation. The confidence threshold of the target detection algorithm is increased based on the health score, and low-confidence detection results are filtered out. Based on the adjusted observation weights and filtered detection results, multi-sensor data association and state updates are performed to generate adaptive fusion perception results.

[0064] In this embodiment of the invention, the observation noise covariance matrix in the fusion filter can be scaled and adjusted based on the health score to reduce or increase the observation weight of the corresponding sensor in state estimation. At the underlying state estimation level (such as target tracking), the trust mechanism of the component-level health real-time intervention filter can be used to enable the system to automatically reduce its dependence on degraded hardware signals during data fusion and prevent "dirty data" pollution.

[0065] Fusion filters are algorithmic architectures used to synthesize multi-source space probe data to eliminate random noise and compute optimal state estimates, such as extended Kalman filters, particle filters, or factor graph optimization models.

[0066] The observation noise covariance matrix is ​​a mathematical matrix used in the fusion filter to quantitatively describe the severity of fluctuations in sensor measurement errors; for example, the larger the value of this matrix, the more unreliable the system considers the sensor's current measurement result to be.

[0067] Scaling adjustment is an online correction operation that dynamically increases or decreases the resistance value of the corresponding element in the matrix (by multiplication or addition) using the health score as the independent variable. For example, when the health score decreases, the observed noise is scaled up proportionally to make the filter more resistant.

[0068] The corresponding sensor refers to the specific individual sensor hardware that is currently having its data processed by the filter and has a corresponding health score input.

[0069] State estimation is the process by which an autonomous driving system calculates the physical state of surrounding dynamic obstacles, such as position, velocity, and acceleration, based on historical trajectories and current observation data. The observation weights in state estimation are the trust proportions or contributions of the data output by the corresponding sensors during the state update of the entire fusion filter.

[0070] In this embodiment of the invention, the output confidence threshold of the target detection algorithm can be increased according to the health score, and low confidence detection results can be filtered out. In order to adaptively increase the detection threshold of obstacles at the upper-level deep learning target recognition level, the spread of false alarms caused by hardware degradation (such as mirror stains and noise) can be blocked and filtered from the source.

[0071] Object detection algorithms are deep learning networks or traditional computer vision algorithms deployed on computing platforms to identify and locate specific categories of obstacles (such as pedestrians, vehicles, and other obstacles) from raw perceptual data streams (images, point clouds, etc.).

[0072] The output confidence threshold is a scoring threshold used by the target detection algorithm to determine whether a detected suspected target is legitimate; only when the probability that the algorithm considers "it is a real obstacle" is higher than this threshold is it allowed to be output to the downstream.

[0073] The filtering confidence test results are data cleaning operations that automatically remove, block, or discard test candidate results that do not meet safety standards based on the set threshold limits.

[0074] Low confidence detection results are data on suspected obstacles that are detected with scores below the adjusted output confidence threshold. This data is often triggered by interference noise caused by sensor hardware degradation.

[0075] In this embodiment of the invention, based on the adjusted observation weights and filtered detection results, multi-sensor data association and state updates can be performed to generate adaptive fusion perception results. This integrates heterogeneous perception data from various channels after microscopic purification and dynamic weighting, completes cross-sensor temporal and spatial matching, and finally refreshes and outputs an adaptive big-screen image that can realistically and safely reflect the surrounding environment.

[0076] The adjusted observation weights are the latest filter confidence quotas, which have been scaled by the matrix and are now matched with the actual hardware degradation of the current sensor.

[0077] The filtered detection results are a clean and highly reliable set of target bounding boxes / dot matrixes that have been adjusted by adaptively raising the threshold and removing soft false alarms and degenerate noise.

[0078] Multi-sensor data association is an algorithmic matching process that pairs spatiotemporal data from different sensors (such as objects detected by cameras and objects tracked by lidar) to determine whether they belong to the same physical obstacle in the real world (such as the Hungarian algorithm and joint probabilistic data association).

[0079] Performing multi-sensor data association and state update is the process by which the fusion filter uses the current associated data to mathematically correct and refresh the predicted physical position and trajectory of the obstacle from the previous moment.

[0080] The adaptive fusion perception result is a three-dimensional structured image of the vehicle's surrounding environment, which is generated by integrating information from all heterogeneous sensors in the vehicle and undergoing in-depth self-adjustment and noise reduction based on the hardware health degradation status throughout the entire calculation process.

[0081] This invention achieves low-level self-healing and fault tolerance of perception data under conditions of degraded individual hardware performance by using micro-adaptive control of health scores in two dimensions: the noise matrix of the fusion filter and the confidence threshold of target detection. This ensures that the autonomous driving system can still output adaptive fusion perception results with high anti-interference capability.

[0082] In an optional embodiment of the present invention, the step of controlling the vehicle based on the control strategy includes: The observation noise covariance matrix in the Kalman filter is adjusted in real time using the health score, and the dynamic observation weights of the sensor are calculated. The output confidence threshold of the target detection network is dynamically adjusted based on the health score, and an adaptive fusion perception result is generated based on the adjusted output confidence threshold and the perception data. A control strategy is generated based on the dynamic observation weights and the adaptive fusion perception results.

[0083] In a specific implementation, the embodiments of the present invention can use the health score to adjust the observation noise covariance matrix in the Kalman filter in real time, calculate the dynamic observation weight of the sensor, so as to realize the dynamic allocation of sensor data reliability, ensure that the system can automatically reduce its dependence on damaged sensor signals during the data fusion stage, and improve the robustness of the sensing system.

[0084] Kalman filtering is an algorithm that uses the state equations of a linear system to make an optimal estimate of the system state through system input and output observation data; it is often used in autonomous driving for continuous tracking of the position and speed of obstacles.

[0085] The observation noise covariance matrix can be used as a parameter to measure the degree of error fluctuation between sensor observations and true values; the larger the matrix value, the less reliable the system considers the sensor's observation results to be.

[0086] Real-time adjustment of the observation noise covariance matrix in Kalman filtering is an operation that synchronously corrects the matrix parameters online based on the currently calculated health score; that is, the lower the health score, the higher the assigned observation noise value.

[0087] Dynamic observation weight is the proportion of influence of each sensor on the final state estimation result during the multi-sensor fusion process; this weight fluctuates in real time as the health status changes.

[0088] In a specific implementation, embodiments of the present invention can dynamically adjust the output confidence threshold of the target detection network according to the health score, and generate an adaptive fusion perception result based on the adjusted output confidence threshold and the perception data, so as to effectively filter false alarm targets caused by frost (such as frost layer being mistaken for an obstacle) through sensitivity adaptive control, thereby improving the accuracy of the perception result.

[0089] Object detection networks are deep learning-based algorithms used to identify and locate obstacles such as vehicles and pedestrians from perceptual data such as images or point clouds.

[0090] The output confidence threshold is the scoring threshold for determining whether the detection result is valid; the algorithm will only output a result when the probability that "the object is an obstacle" is higher than this threshold.

[0091] The dynamic adjustment of the target detection network's output confidence threshold means that when the health level decreases (severe frost), the system automatically raises the confidence threshold to prevent noise caused by frost from being incorrectly identified as target objects.

[0092] The adaptive fusion perception result is the final description of the vehicle's surrounding environment formed by integrating information from multiple sensors and filtering and weighting it, including obstacle type, distance, speed, etc.

[0093] In a specific implementation, embodiments of the present invention can generate a control strategy based on the dynamic observation weights and the adaptive fusion perception results, so as to combine the "credibility of each sensor" and the "final environmental profile" to formulate a driving scheme that matches the current perception capability.

[0094] The control strategy is to calculate the next action the vehicle should take based on the quality and content of the perceived information, such as maintaining the current lane, decelerating, or preparing to take over.

[0095] This invention achieves adaptive fault tolerance of the sensing system for frost conditions by finely adjusting the observation noise and detection threshold at the algorithm level through health scores. This ensures that even when some sensors are blocked, the system can still output reliable adaptive fusion sensing results and generate reasonable control strategies through weight redistribution.

[0096] In an optional embodiment of the present invention, the step of controlling the vehicle based on the control strategy includes: The dynamic observation weights and health scores of multiple sensors are aggregated, and a system confidence index reflecting the overall reliability of system environment modeling is generated through weighted average calculation. The system confidence index is input into a preset safety constraint mapping function to calculate and generate dynamic safety boundary parameters including the maximum driving speed limit, target following distance, and lane change permission instructions. Extract the spatiotemporal attribute information of each obstacle from the adaptive fusion perception results; Combining the speed and spacing constraints set by the dynamic safety boundary parameters, a local planning trajectory that meets the safety margin is generated through spatiotemporal joint optimization; The system confidence index is compared with preset multi-level risk thresholds in real time, and the vehicle is controlled based on the comparison results.

[0097] In a specific implementation, embodiments of the present invention can aggregate the dynamic observation weights and health scores of multiple sensors, and generate a system confidence index that reflects the overall reliability of system environment modeling through weighted average calculation, so as to aggregate the scattered single sensor evaluations into a global system evaluation and quantify the overall credibility of the current description of the external environment by the whole vehicle perception system.

[0098] Aggregating the dynamic observation weights and health scores of multiple sensors involves extracting the weight parameters and health scores distributed across different processors or algorithm channels and centralizing them at the decision center.

[0099] Weighted average is a mathematical calculation method that calculates a comprehensive average value based on the importance (weight) of each sensor in the sensing system and its own health status (score).

[0100] System environment modeling is the creation of a three-dimensional environment model of the vehicle's surroundings in digital space using sensor data by an autonomous driving system. This model includes curbs, lane lines, and obstacles.

[0101] The system confidence index represents a quantitative indicator of how confident the entire autonomous driving system is in believing that the current perceived environment is real; the higher the value, the higher the system's confidence in modeling the current environment.

[0102] In a specific implementation, the embodiments of the present invention can input the system confidence index into a preset safety constraint mapping function to calculate and generate dynamic safety boundary parameters including the maximum driving speed limit, target following distance and lane change permission instruction, so as to realize the logical mapping of "perception ability determines driving range" and dynamically lock the physical movement limit of the vehicle according to the strength of system confidence.

[0103] A safety constraint mapping function is a pre-defined logical relationship model used to determine the correspondence between "confidence level" and "safety limit" (e.g., a 20% decrease in confidence level corresponds to a 30% reduction in speed limit).

[0104] The maximum driving speed limit is the highest cruising speed that a vehicle is allowed to reach under current operating conditions, in order to allow for a longer braking distance in the event of perceived damage.

[0105] The target following distance is the time interval between the vehicle and the target object in front; the more severe the frost, the longer this distance should be set to increase safety redundancy.

[0106] The lane change permission instruction is a logic switch used to determine whether high-risk lateral maneuvers such as lane changes are currently permitted.

[0107] Dynamic safety boundary parameters are a set of physical constraints that change in real time with the system state, defining the "cage" for safe vehicle operation.

[0108] In a specific implementation, embodiments of the present invention can extract the spatiotemporal attribute information of each obstacle from the adaptive fusion perception result to obtain the physical state of the surrounding dynamic obstacles, providing a reference for coordinate and time dimensions for subsequent obstacle avoidance planning.

[0109] Obstacles are objects in or around a vehicle's path that may pose a collision risk, such as other vehicles, pedestrians, construction fences, etc.

[0110] Spatiotemporal attribute information includes comprehensive data in both spatial (position, size, heading angle) and temporal (velocity, acceleration, predicted trajectory) dimensions.

[0111] In a specific implementation, embodiments of the present invention can combine the speed and spacing constraints set by the dynamic safety boundary parameters to generate a local planning trajectory that meets the safety margin through spatiotemporal joint optimization, so as to calculate a driving path that can avoid obstacles and meet the current perception reliability without exceeding the safety boundary.

[0112] Speed ​​and spacing constraints refer to the speed limits and following distance requirements that can be generated in the specific implementation of this invention, which serve as hard boundaries that trajectory planning must adhere to.

[0113] Spatiotemporal joint optimization is a high-dimensional planning algorithm that simultaneously searches for the optimal path on both the time and space axes, ensuring that the trajectory is physically smooth and temporally unconflicting.

[0114] Safety margin is the extra safety clearance maintained between the track and obstacles or lane edges.

[0115] Local planning trajectory: A sequence of timestamped coordinates for a vehicle to travel in the next few seconds.

[0116] In a specific implementation, the embodiments of the present invention can compare the system confidence index with the preset multi-level risk thresholds in real time, control the vehicle according to the comparison results, establish a final defense mechanism, and decide whether to perform normal autonomous driving actions or mandatory safety degradation actions based on the severity of damage to the perception system.

[0117] Multi-level risk thresholds are a set of pre-defined hierarchical indicators (such as excellent, good, average, and poor) used to define different risk levels from "fully automated driving" to "mandatory stopping in the emergency lane".

[0118] The comparison results can be used to characterize which threshold range the confidence index falls within.

[0119] Based on the comparison results, the vehicle control system executes corresponding low-level control actions according to the risk level, including normal longitudinal and lateral control or emergency braking.

[0120] This invention transforms global perception reliability into dynamic safety boundaries and trajectory constraints, enabling the autonomous driving decision layer to physically address the risk of frost, ensuring that the vehicle's driving plan is always strictly limited to the current perception level, and fundamentally eliminating the risk of speeding or exceeding distance due to blind confidence.

[0121] In an optional embodiment of the present invention, the step of controlling the vehicle based on the comparison result includes: When the system confidence index is determined to be higher than the first risk threshold, longitudinal acceleration control commands and lateral turning control commands are generated based on the local planning trajectory for the vehicle's power output and direction control actuators.

[0122] In a specific implementation, embodiments of the present invention may allow physical-level power and steering output only when the system confidence index is determined to be higher than the first risk threshold.

[0123] The first risk threshold is a set of values ​​that can be preset by the system to define the safety level (e.g., safety level, warning level, danger level); here it specifically refers to the minimum threshold that represents "safe driving".

[0124] In a specific implementation, embodiments of the present invention can generate longitudinal acceleration control commands and lateral turning angle control commands for the vehicle's power output and direction control actuators based on the local planning trajectory, so as to establish the preconditions for executing normal autonomous driving control, and transform the virtual driving path into physical action commands that the vehicle hardware can recognize, thereby realizing the final execution of autonomous driving.

[0125] The power output and steering control actuators are the vehicle chassis hardware, including the engine / motor drive controller (responsible for acceleration), the electronic braking system (responsible for deceleration), and the electronic power steering system (responsible for steering).

[0126] Longitudinal acceleration control commands are commands that act on the vehicle's direction of travel (forward or backward). By controlling the throttle opening or braking pressure, the vehicle's acceleration or deceleration actions are achieved.

[0127] Lateral steering control commands are commands that act on the vehicle's steering system (left and right). By controlling the rotation angle of the steering motor, the vehicle can be turned or made to turn around.

[0128] By establishing a confidence-based admission mechanism at the execution layer, this invention ensures that the vehicle performs complex autonomous driving actions only when the perception system is highly reliable. This achieves a deep closed loop between control commands and perception quality, greatly reducing the probability of erroneous operations due to impaired perception accuracy.

[0129] In an optional embodiment of the present invention, it further includes: When the system confidence index is determined to be lower than the second risk threshold and no driver takeover signal is received within a preset time, the system uses the lane topology and side obstacle distribution information in the adaptive fusion perception results to generate a graded minimum risk strategy for the vehicle's braking and steering control actuators, and controls the vehicle to decelerate and stop at the target position based on the graded minimum risk strategy.

[0130] In a specific implementation, the embodiments of the present invention can forcibly activate the system self-protection when it is determined that the system confidence index is lower than the second risk threshold and the driver has not taken over, which means that under the dual risks of the perception system being untrustworthy and the failure of manual takeover, the system self-protection is activated.

[0131] The system confidence index is a quantitative value that reflects the overall realism of the environmental description after the fusion of all vehicle sensors.

[0132] The second risk threshold is a set of preset safety limits; here it specifically refers to the lowest level of "safety red line", once it is broken, it means that the system can no longer guarantee basic autonomous driving safety.

[0133] The failure to receive a driver takeover signal within a preset time is used to characterize the state where, after the system issues a takeover request (such as an audible and visual alarm), the driver is not detected to intervene in steering, braking, or pressing the accelerator pedal within a preset time.

[0134] In specific implementation, embodiments of the present invention can determine the critical conditions for triggering the emergency safety mechanism, and utilize the lane topology and lateral obstacle distribution information in the adaptive fusion perception results to fully explore the "residual value" in the damaged perception system, and use the key information that is limited but still usable to provide navigation basis for risk avoidance actions.

[0135] The adaptive fusion perception result is the output of environmental structured information after integrating data from multiple sensors and undergoing adaptive adjustment.

[0136] Lane topology refers to the spatial geometric relationship and connection logic of road lane lines, which guides vehicles to stay within the lane or move towards the roadside during parking.

[0137] Lateral obstacle distribution information is a spatial description of whether there are other vehicles, curbs, or guardrails in the adjacent lanes on the left and right sides of the vehicle, and is used to assess the safety of lane change for hazard avoidance.

[0138] In a specific implementation, embodiments of the present invention can generate a graded minimum risk strategy for the braking and steering control actuators of a vehicle, so as to formulate an orderly set of degraded action instructions according to the urgency of the risk, and terminate the dangerous driving state in the smoothest and safest way.

[0139] Braking and steering control actuators are the chassis control system of a vehicle, including the electronic brake control unit (which controls deceleration) and the steering control unit (which controls orientation).

[0140] The Minimum Risk Maneuver (MRM) is an emergency response logic that executes different safety actions based on the perceived level of damage. For example, Level 1 involves slowing down and stopping within the lane, while Level 2 involves changing lanes to the shoulder / emergency lane and stopping.

[0141] In specific implementations, embodiments of the present invention can control the vehicle to decelerate and stop at the target position based on the graded minimum risk strategy, so as to ultimately perform a risk avoidance action and ensure that the vehicle can still stop safely and orderly in the preset refuge area under extreme conditions of perception failure.

[0142] Controlling vehicle deceleration involves applying braking force to the wheels through an electronic braking system, causing the vehicle speed to decrease smoothly to a standstill.

[0143] The target location is the final avoidance point in the strategic planning, which is usually a safe location within the current lane or an emergency stopping area on the side of the road.

[0144] For example, suppose the system presets three risk thresholds, and divides the system confidence index (ranging from 0% to 100%) into four intervals from high to low, each corresponding to different judgment conditions and control strategies: Level 1 risk threshold (safety red line T_1) = 80%; Level 2 risk threshold (warning red line T_2=50%) Level 3 risk threshold (collapse threshold T_3 = 20%) The first working scenario is to determine whether the conditions for "normal autonomous driving" are met; The input status is sunny weather, and all vehicle sensors are in extremely high health.

[0145] The calculated system confidence index is P_sys = 90%.

[0146] The system compares 90% with the threshold set level by level, determining that the system confidence index is higher than the first-level risk threshold T_1 (i.e., falls within the absolute safety range of [80%, 100%]). The system then generates a normal local planning trajectory to control the vehicle's normal driving.

[0147] The second working scenario is determined to meet the condition of "dynamic boundary tightening (speed limit / increased distance)"; The input status is that there is a sudden light fog ahead, and the data quality indicators of some cameras and LiDAR have declined, resulting in a loss of health.

[0148] The system confidence index calculation result is P_sys=65%.

[0149] After comparison, the system found that the indicator was between the first and second level thresholds. It determined that the system confidence index was lower than the first risk threshold T_1 and higher than the second risk threshold T_2 (i.e., falling into the low-risk warning range of [50%, 80%)). The system triggered an adaptive mechanism to calculate and generate dynamic safety boundary parameters, including the maximum driving speed limit and the target following distance.

[0150] Scenario 3 is determined to meet the conditions for "triggering the graded minimum risk strategy (safe docking)"; The input condition is a sudden, extremely severe working condition, where the surface of all vehicle sensors is severely obscured by mud.

[0151] The system confidence index calculation result is P_sys = 35%.

[0152] After comparison, the system determined that the indicator had fallen below the secondary threshold but remained above the tertiary threshold. Therefore, the system confidence index was determined to be below the secondary risk threshold T_2 (specifically falling into the high-risk emergency zone of [20%, 50%)), and the driver did not respond after the takeover signal was issued. The system then forcibly generated a tiered minimum risk strategy, controlling the vehicle to decelerate and safely stop at the target location.

[0153] This invention introduces a tiered minimum risk strategy in the extreme state of system confidence collapse, achieving closed-loop guidance from "perception failure" to "safe parking". This ensures that even when frost causes severe vision impairment and no human intervention is available, the vehicle can still use residual perception information to perform smooth avoidance actions, avoiding the risk of secondary accidents caused by sudden braking or blind driving.

[0154] To enable those skilled in the art to better understand the embodiments of the present invention, a complete example is used below to illustrate the embodiments of the present invention.

[0155] refer to Figure 2 , Figure 2 This is a flowchart of another vehicle control method provided in this embodiment of the invention; its core process includes three layers: sensor health assessment layer, fusion perception adaptive layer, and system safety decision layer.

[0156] Step S1: Sensor Health Assessment - In-depth Logic and Examples Logical objective: To convert the sensor's physical state (frost) into a computable, standardized health score H_s.

[0157] Detailed process (using a front-view camera as an example): Inputs: Infrared thermal imager temperature matrix T_glass(x,y), vehicle interior dew point temperature T_dew, camera raw image I_raw, and cleanliness baseline model M_base.

[0158] Sub-step S1.1 Frosting condition diagnosis: Logic: Scan the T_glass matrix to identify continuous regions R_frost with temperatures below T_dew + ΔT (ΔT being a safety margin). Calculate the area ratio A_ratio and the average temperature gradient (temperature difference from edge to center) G_avg for this region.

[0159] Output: Frost area mask M_frost, preliminary severity index [A_ratio, G_avg].

[0160] Example: Assume T_dew = 2°C, ΔT = 1°C. The temperature in the lower left area of ​​the windshield is detected to be between -1°C and 1°C, covering 30% of the camera's field of view (A_ratio = 0.3). The center of this area is the coldest (-1°C), and the edges are warmer (1°C), resulting in G_avg = 2°C. This is determined to be localized moderate frost.

[0161] Sub-step S1.2 Performance degradation quantification: Logic: Within the region marked M_frost, extract an image patch from I_raw. Calculate the key quality metrics for this image patch: Q_contrast: Local contrast (normalized variance).

[0162] Q_hf: High-frequency energy (the proportion of high-frequency component energy after Fourier transform of the image patch).

[0163] Q_feat: Number and stability of feature points (using the FAST corner detector to compare the matching rate of feature points between the current frame and the previous frame).

[0164] K_contrast = Q_contrast_current / Q_contrast_base K_hf = Q_hf_current / Q_hf_base K_feat = Q_feat_match_rate (directly used as a coefficient, 0-1) K_img = α*K_contrast + β*K_hf + γ*K_feat (α+β+γ=1, which are learnable or preset weights) Output: Image quality attenuation coefficient K_img(0-1).

[0165] Example: In the frosted area, Q_contrast_current=0.1, Q_contrast_base=0.5 → K_contrast=0.2. K_hf=0.15, K_feat=0.1. Assuming weights α=0.4, β=0.4, γ=0.2, then K_img = 0.4 * 0.2 + 0.4 * 0.15 + 0.2 * 0.1 = 0.16. This indicates that the image quality in this area has degraded to 16% of the baseline.

[0166] Sub-step S1.3 Health score synthesis: Logic: Combining the severity of frosting and performance degradation, a final health level is synthesized. A feasible non-linear mapping function is as follows: H_s = K_img * exp(-λ * A_ratio * (1 + G_avg / G0)) Where λ is the decay factor and G0 is the reference temperature gradient. A large A_ratio (large frost area) or a large G_avg (thick frost, large temperature difference) will decrease the H_s exponent. K_img is a multiplicative term that directly reflects the current mass.

[0167] Output: Camera health score H_cam.

[0168] Example: Continuing from the previous example, K_img=0.16, A_ratio=0.3, G_avg=2, let λ=1, G0=5. Then H_cam =0.16 * exp(-1 * 0.3 * (1+2 / 5)) = 0.16 * exp(-0.42) ≈ 0.16 * 0.657 ≈ 0.105. The health score is approximately 0.1 (out of 1.0), indicating severe performance degradation.

[0169] Step S2: Health-Based Fusion-Aware Adaptation - In-Depth Logic and Examples H_s is transformed into dynamic weights in perceptual fusion to optimize the fusion result.

[0170] Detailed process (taking forward target tracking as an example, fusing camera and millimeter-wave radar): Input: Camera target list List_cam (including position, speed, and confidence level), radar target list List_radar, camera health H_cam, radar health H_radar (radar is generally unaffected by frost, and H_radar is usually ≈ 1.0).

[0171] Sub-step S2.1 Dynamic fusion weight calculation: In data association and tracking filtering (such as Kalman filtering), the reciprocal of the observation noise covariance matrix R for each sensor is equivalent to its weight. The observation weights of sensor s are defined as: W_s ∝ H_s^2 / σ_s^2 Where σ_s^2 is the variance of the baseline observation noise of the sensor under ideal conditions. H_s^2 reflects the increased uncertainty caused by the decline in health (the variance increases to σ_s^2 / H_s^2).

[0172] Specific operation: In the Kalman filter update step, the observation noise covariance matrix is ​​temporarily adjusted to R_s' = R_s / H_s^2. The filter will automatically assign higher weights to healthier sensors (those with larger H_s).

[0173] Example: The variance of the observation noise at the camera reference position is σ_cam^2 = 0.1 m², and the variance of the radar is σ_radar^2 = 0.5 m². Currently, H_cam = 0.1 and H_radar = 1.0.

[0174] Effective noise variance of the camera: R_cam' = 0.1 / (0.1^2) = 10 m² Radar effective noise variance: R_radar' = 0.5 / (1.0^2) = 0.5 m² In the filter, the weight of radar observations will be 20 times that of camera observations, and the system will trust radar data more.

[0175] Sub-step S2.2: Adjusting the parameters of the perception algorithm: Logic: For sensors with low health, increase the decision threshold of their internal sensing algorithms and reduce the injection of unreliable data into the fusion center.

[0176] Specific operation: For the camera target detection network, dynamically increase its output confidence threshold from Th_high=0.7 to Th_low=0.9. Th = Th_high + (1 - H_cam) * (Th_low - Th_high).

[0177] Example: H_cam=0.1, then Th = 0.7 + (1-0.1)*(0.9-0.7) = 0.88. Only detected targets with a confidence level higher than 0.88 will be sent to the fusion module, which greatly reduces the probability of false alarm targets caused by image blurring entering the downstream.

[0178] Step S3: System Security Decisions and Control - In-depth Logic and Examples Logical goal: To map system-level health states into actionable security decisions.

[0179] Detailed process: Input: The set of health status of all sensors H_s, the current system status (vehicle speed V, following distance TTC, lane information, etc.), and the current driving mode.

[0180] Sub-step S3.1 Dynamic safety boundary adjustment: Logic: Define the overall system health index H_sys, for example, H_sys = min(H_s) or H_sys = ∏(H_s)^w_s (geometric mean). Establish a continuous mapping function from H_sys to the safety boundary parameters.

[0181] Example: Assume only the forward-facing camera is frosted, H_cam = 0.25, and all other sensors are healthy, H_sys = 0.25. Looking up the table, the system will limit the vehicle speed to 60 km / h, increase the following distance requirement to 3.5 seconds, and disable automatic lane changing.

[0182] Sub-step S3.2 Minimum Risk Strategies (MRM) Triggered: Logic: Set a series of decreasing thresholds θ1>θ2>θ3, corresponding to different levels of MRM.

[0183] Strategy Flow: Continuous monitoring: Real-time calculation of H_sys(t).

[0184] Conditional judgment and triggering: If H_sys(t) < θ1 and continues for ΔT1: trigger Level 1 warning. The affected sensor area is highlighted on the HUD, and a voice prompt reads, "Sensor field of view is limited, functionality is limited, please prepare to take over."

[0185] If H_sys(t) < θ2 or `H_sys(t) < θ1 and the driver does not take over within ΔT2: Level 2 intervention is triggered**. The system automatically and smoothly decelerates to a safe speed (e.g., 40 km / h), automatically activates hazard lights, and plans a route to change lanes to the rightmost lane at the safest time.

[0186] If H_sys(t) < θ3 or the vehicle is in an extremely dangerous state: Trigger Level 3 emergency takeover. Perform an emergency pullover (e.g., turn on the turn signal and stop in the emergency lane if it is safe to do so) and continue to issue a strong audible and visual alarm until the driver takes over.

[0187] Example: Let θ1=0.3, θ2=0.15, θ3=0.05, ΔT1=5s, ΔT2=10s. H_sys decreases from 0.4 to 0.25 (<θ1) and remains at that level for 5 seconds, triggering a Level 1 warning. If the driver does not respond after 10 seconds, and H_sys remains at 0.25, a Level 2 intervention is triggered, and the system automatically decelerates, activates hazard lights, and changes lanes to the right.

[0188] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0189] This application also provides a vehicle control device 30, see reference. Figure 3 The diagram shows a structural block diagram of a vehicle control device provided in an embodiment of the present invention, including: a physical state information and perception data acquisition module 310, a data quality index determination module 320, a health score conversion module 330, and a vehicle control module 340.

[0190] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0191] In addition, this application also provides an electronic device 40, please refer to... Figure 4 It includes a processor 410 and a memory 420, wherein the memory 410 is used to store computer programs; and the processor 420 is used to execute the programs stored in the memory 410 to implement the vehicle control method described in any embodiment of this application.

[0192] This invention also discloses a vehicle that includes the electronic equipment described above.

[0193] like Figure 5 As shown, in another embodiment of the present invention, a computer-readable storage medium 501 is also provided, which stores instructions that, when executed on a computer, cause the computer to perform the vehicle control method described in the above embodiments.

[0194] In this application, "multiple" refers to two or more.

[0195] In this application, unless otherwise expressly defined, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0196] The terms “first,” “second,” “third,” “fourth,” etc., in this application (if present) are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0197] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0198] Unless otherwise specified, all steps in this application may be performed sequentially or randomly. For example, if the method includes steps A and B, it means that the method may include steps A and B performed sequentially, or it may include steps B and A performed sequentially. For example, if the method may also include step C, it means that step C may be added to the method in any order. For example, the method may include steps A, B, and C, or it may include steps A, C, and B, or it may include steps C, A, and B, etc.

[0199] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A vehicle control method, characterized in that, include: Acquire sensor physical state information and sensing data from multiple sensors; Determine the data quality indicators of the perceived data; Based on the physical state information of the sensors from multiple sensors and the data quality indicators, a health score is determined for each of the multiple sensors. The system-level health status is calculated based on the health score of each of the multiple sensors, and a vehicle control strategy is generated to control the vehicle based on the mapping relationship between the system-level health status and the safety boundary parameters.

2. The method according to claim 1, characterized in that, The sensor types mentioned include at least image sensors, point cloud sensors, electromagnetic or acoustic wave sensors, and inertial measurement units. The step of determining the data quality indicators of the sensed data includes: The local contrast and high-frequency energy of the sensor data extracted by the image sensor are obtained. Obtain the point cloud density and reflection intensity of the sensing data extracted by the point cloud sensor. The signal-to-noise ratio and echo attenuation rate of the sensing data from the electromagnetic or acoustic wave sensor are obtained. Obtain the zero-bias drift of the sensing data from the inertial measurement unit; The local contrast and high-frequency energy, the point cloud density and reflection intensity, the signal-to-noise ratio and echo attenuation rate, and the zero-bias drift are determined as the data quality indicators of the corresponding sensors.

3. The method according to claim 2, characterized in that, The method of determining the health score of each of the multiple sensors based on the sensor physical state information and the data quality indicators includes: Obtain a state baseline performance model that reflects the sensing capability of the sensor under normal conditions; Obtain the vehicle interior dew point temperature, and calculate the sum of the vehicle interior dew point temperature and the preset safety margin to determine the low temperature threshold; Identify abnormal areas where the temperature of the sensor is below the low-temperature threshold; The proportion of the frosted area is calculated by comparing the area of ​​the abnormal region with the total area of ​​the sensor. Generate an average temperature gradient that reflects the temperature difference from the edge to the center of the abnormal region; Based on the sensor type, the signal quality characteristic parameters corresponding to the sensor are extracted from the data quality indicators; The signal quality characteristic parameters are analyzed using the state benchmark performance model to generate a performance attenuation coefficient that characterizes the residual proportion of the sensor's sensing capability. The area ratio of the frosted region and the average temperature gradient are used as exponential decay terms by using a nonlinear mapping function, and combined with the performance decay coefficient as a multiplicative correction term for fusion calculation to obtain a health score for expressing the sensor.

4. The method according to claim 3, characterized in that, Also includes: Based on the health score, the confidence weights or detection decision thresholds of the multiple sensors in the perception fusion are dynamically adjusted to generate adaptive perception results. The step of dynamically adjusting the confidence weights or detection decision thresholds of the multiple sensors in perception fusion based on the health score to generate adaptive perception results includes: The observation noise covariance matrix in the fusion filter is scaled and adjusted based on the health score to reduce or increase the observation weight of the corresponding sensor in state estimation. The confidence threshold of the target detection algorithm is increased based on the health score, and low-confidence detection results are filtered out. Based on the adjusted observation weights and filtered detection results, multi-sensor data association and state updates are performed to generate adaptive fusion perception results.

5. The method according to claim 4, characterized in that, The steps of controlling the vehicle based on the control strategy include: The dynamic observation weights and health scores of multiple sensors are aggregated, and a system confidence index reflecting the overall reliability of system environment modeling is generated through weighted average calculation. The system confidence index is input into a preset safety constraint mapping function to calculate and generate dynamic safety boundary parameters including the maximum driving speed limit, target following distance, and lane change permission instructions. Extract the spatiotemporal attribute information of each obstacle from the adaptive fusion perception results; Combining the speed and spacing constraints set by the dynamic safety boundary parameters, a local planning trajectory that meets the safety margin is generated through spatiotemporal joint optimization; The system confidence index is compared with preset multi-level risk thresholds in real time, and the vehicle is controlled based on the comparison results.

6. The method according to claim 5, characterized in that, The step of controlling the vehicle based on the comparison result includes: When the system confidence index is determined to be higher than the first risk threshold, longitudinal acceleration control commands and lateral turning control commands are generated based on the local planning trajectory for the vehicle's power output and direction control actuators.

7. The method according to claim 5 or 6, characterized in that, Also includes: When the system confidence index is determined to be lower than the second risk threshold and no driver takeover signal is received within a preset time, the system uses the lane topology and side obstacle distribution information in the adaptive fusion perception results to generate a graded minimum risk strategy for the vehicle's braking and steering control actuators, and controls the vehicle to decelerate and stop at the target position based on the graded minimum risk strategy.

8. A vehicle control device, characterized in that, include: The data acquisition module is used to acquire sensor physical state information and sensing data from multiple sensors; A data quality index determination module is used to determine the data quality index of the perceived data; A health score conversion module is used to determine the health score of each of the multiple sensors based on the sensor physical state information and the data quality indicators. The vehicle control module is used to calculate the system-level health status based on the health score of each of the plurality of sensors, and to generate a vehicle control strategy to control the vehicle based on the mapping relationship between the system-level health status and the safety boundary parameters.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes a program stored in the memory, it implements the method as described in any one of claims 1-12 or 13.

10. A vehicle, characterized in that, It includes the electronic device as described in claim 9.

Images