Automatic driving decision planning cooperation method and system based on fault prediction

Abstract

The invention relates to the technical field of automatic driving, and discloses an automatic driving decision planning cooperation method and system based on fault prediction, and the method comprises the steps: collecting gateway operation data, carrying out the abnormal trend detection and analysis of time sequence features, and generating a fault probability prediction curve; acquiring a load trigger signal and dynamic response data, and estimating a quality parameter on line by a recursive least square algorithm; the mass center position is calculated by combining suspension displacement data with a statics equation, and a kinetic parameter updating package is generated through fusion; comparing the gateway fault probability with a threshold judgment risk, sending an update packet and generating a check code at a low risk, and pre-synchronizing a storage snapshot at a high risk; updating the kinetic equation coefficient matrix and outputting to a decision domain and a chassis domain; when the gateway fails, each domain operates independently based on a snapshot, and versions are compared to verify consistency and output a collaborative identifier; according to the invention, each function domain can still cooperatively operate based on the same kinetic parameter reference during the gateway fault period.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and more specifically, to a collaborative method and system for autonomous driving decision-making and planning based on fault prediction. Background Technology

[0002] Autonomous taxis and freight vehicles generally employ a domain controller architecture, with each functional domain (perception domain, decision domain, and chassis domain) exchanging data through an onboard gateway. As a cross-domain communication hub, the onboard gateway plays a crucial role in synchronizing dynamic parameters between functional domains. During vehicle operation, passenger boarding / alighting or cargo loading / unloading causes changes in vehicle load. This necessitates online estimation of new dynamic parameters (including vehicle mass and center of gravity position), and the updated parameters are synchronized to each functional domain via the gateway. This ensures that the decision domain and chassis domain use a consistent parameter benchmark for trajectory planning and execution control.

[0003] The existing technology has the following technical problems: When the vehicle load changes, the updated dynamic parameters need to be synchronized to each functional domain through a gateway. If the gateway fails or degrades in performance during the parameter synchronization process, some functional domains may receive the new parameters while others continue to use the old parameters, resulting in inconsistent parameter versions. This inconsistency causes the decision domain to plan the trajectory based on the new mass parameters, while the chassis domain executes control based on the old parameters, resulting in a systemic planning-execution deviation. The existing technology treats gateway failure handling and dynamic parameter updating as independent issues, lacking a coordinated consideration of the timing of gateway failures and the parameter synchronization process, and failing to anticipate the potential threat to parameter synchronization during the gateway performance degradation phase. Summary of the Invention

[0004] This invention provides a collaborative method and system for autonomous driving decision-making and planning based on fault prediction, which solves the technical problem of uncontrollable gateway failure timing in related technologies.

[0005] This invention provides a collaborative decision-making and planning method for autonomous driving based on fault prediction, comprising the following steps: Acquire the operational status data of the vehicle gateway, analyze the time-series changes of health indicators using an abnormal trend detection algorithm, and generate a gateway failure probability prediction curve. Acquire load change event trigger signals and vehicle dynamic response data, and use the recursive least squares algorithm to estimate the current mass parameters of the vehicle online; Acquire suspension displacement sensor data, calculate the estimated value of the center of mass position using the static equilibrium equation, and combine the mass parameters and the center of mass position to generate a dynamic parameter update package; Obtain the parameter version identifier of each domain controller, compare the predicted gateway failure probability value with the threshold, and determine the gateway risk status. When the gateway is in a low-risk state, dynamic parameter update packets are sent to each domain controller through the gateway to generate cross-domain parameter consistency verification codes; when the gateway is in a high-risk state, parameter pre-synchronization is performed and parameter snapshot storage instructions are sent to each domain controller. Based on the updated dynamic parameters, the coefficient matrix of the dynamic equation is recalculated and output to the decision domain and chassis domain. When the gateway fails, each domain runs independently based on local parameter snapshots, uses a version identifier comparison algorithm to verify parameter consistency, and outputs a collaborative status identifier.

[0006] Furthermore, the step of acquiring the operating status data of the vehicle gateway, analyzing the time-series changes of health indicators using an abnormal trend detection algorithm, and generating a gateway failure probability prediction curve includes: Obtain the vehicle gateway's processing latency data, packet loss rate data, and temperature index data within the current time window; Trend features were extracted from processing delay data, packet loss rate data, and temperature index data, respectively. The trend characteristics of each health indicator are weighted and fused to generate a comprehensive failure probability prediction value; Arrange the comprehensive failure probability prediction values ​​in chronological order to form a gateway failure probability prediction curve that characterizes the probability distribution of gateway failure within a preset time period.

[0007] Furthermore, the acquisition of load change event trigger signals and vehicle dynamic response data, and the online estimation of the vehicle's current mass parameters using a recursive least squares algorithm, includes: Acquire load change event trigger signals, which include door opening / closing status signals and cargo compartment sensor status change signals; After detecting a load change event, acquire the vehicle's acceleration response data and driving torque data within a preset time period before and after the event; The longitudinal dynamics equation of the vehicle is established based on Newton's second law, and the acceleration response data and driving torque data are input into the recursive least squares algorithm for parameter identification. The recursive least squares algorithm uses a forgetting factor to adjust the weight ratio of historical data and current data, and outputs an estimated value of the vehicle's current quality parameters.

[0008] Furthermore, the process of acquiring suspension displacement sensor data, calculating the estimated center of gravity position using the static equilibrium equation, and merging the mass parameters and center of gravity position to generate a dynamic parameter update package includes: The displacement data of each wheel collected by the suspension displacement sensor is obtained, including the displacement values ​​of the front axle left wheel, the front axle right wheel, the rear axle left wheel, and the rear axle right wheel. Input the displacement data of each wheel and the suspension stiffness parameters into the static equilibrium equation to calculate the vertical load borne by each wheel; Based on the vertical loads of each wheel and the vehicle's geometric parameters, the estimated position coordinates of the center of mass in the longitudinal and lateral directions of the vehicle are obtained. The estimated mass parameters and the estimated centroid position are merged and packaged to generate a dynamic parameter update package containing a timestamp and version identifier.

[0009] Furthermore, the step of performing parameter pre-synchronization and sending parameter snapshot storage instructions to each domain controller when the gateway is in a high-risk state includes: Determine if the gateway is still available. If the gateway is still available, the dynamic parameter update package will be sent through the gateway to the perception domain controller, decision domain controller and chassis domain controller. Send parameter snapshot storage instructions to each domain controller; Upon receiving the storage instruction, each domain controller stores the latest parameter version and version identifier in its local non-volatile memory, forming a parameter snapshot.

[0010] Furthermore, the recalculation of the dynamic equation coefficient matrix based on the updated dynamic parameters and the output to the decision domain and chassis domain includes: Based on the updated mass parameters and centroid position parameters, the coefficient matrix of the vehicle dynamics equations is recalculated; The updated coefficient matrix is ​​input into the trajectory planning module of the decision domain. The trajectory planning module reconstructs the constraints for trajectory optimization based on the new coefficient matrix. The constraints include acceleration constraints, steering constraints, and stability constraints. The output parameter update notification is sent to the chassis domain controller, which adjusts the control gain parameter based on the new dynamic parameters. A parameter transition flag is generated. During the parameter transition, a conservative trajectory planning strategy is adopted in the decision domain. Once the chassis domain confirms that the control gain adjustment is complete, the parameter transition flag is cleared.

[0011] Furthermore, when a gateway fails, each domain operates independently based on a local parameter snapshot, uses a version identifier comparison algorithm to verify parameter consistency, and outputs a collaborative status identifier, including: When a gateway failure is detected that causes cross-domain communication to be interrupted, each domain controller switches to independent operation mode; Each domain controller reads a snapshot of parameters stored locally and continues to execute its respective functional tasks based on the dynamic parameters in the snapshot; The decision domain controller sends a version identifier query request to the chassis domain controller via a backup communication link; The chassis domain controller returns the parameter version identifier stored locally; The decision domain controller executes a version identifier comparison algorithm to compare the version identifier of this domain with the version identifier of the chassis domain; If the version identifiers are consistent, output a normal status indicator for planning-control coordination; if the version identifiers are inconsistent, output a warning indicator for inconsistent parameter versions and trigger a degraded operation strategy.

[0012] Furthermore, the degradation operation strategy includes: The decision domain adopts a conservative trajectory planning boundary to increase the safety margin and compensate for execution deviations caused by parameter inconsistencies.

[0013] This invention provides a fault prediction-based autonomous driving decision-planning collaborative system for executing the aforementioned fault prediction-based autonomous driving decision-planning collaborative method, comprising: The gateway status monitoring module is used to acquire the operating status data of the vehicle gateway, analyze the time-series changes of health indicators using an abnormal trend detection algorithm, and generate a gateway failure probability prediction curve. The dynamic parameter estimation module is used to acquire load change event trigger signals and vehicle dynamic response data, estimate the current mass parameters of the vehicle online using the recursive least squares algorithm, calculate the estimated value of the center of mass position using the static equilibrium equation, and generate a dynamic parameter update package. The risk assessment module is used to obtain the parameter version identifier of each domain controller, compare the predicted value of gateway failure probability with the threshold, and determine the risk status of the gateway. The parameter synchronization module is used to send dynamic parameter update packages to each domain controller through the gateway and generate cross-domain parameter consistency verification codes when the gateway is in a low-risk state, and to perform parameter pre-synchronization and send parameter snapshot storage instructions to each domain controller when the gateway is in a high-risk state. The planning and control update module is used to recalculate the coefficient matrix of the dynamic equations based on the updated dynamic parameters and output it to the decision domain and chassis domain. The collaborative operation module is used to control each domain to run independently based on local parameter snapshots when the gateway fails. It uses a version identifier comparison algorithm to verify parameter consistency and outputs a collaborative status identifier.

[0014] The beneficial effects of this invention are: overcoming the uncontrollable timing of gateway failures in the prior art, solving the technical problem of systematic planning-execution deviation caused by inconsistent parameter benchmarks between the decision domain and the chassis domain, and ensuring that each functional domain can still operate collaboratively based on the same dynamic parameter benchmarks during gateway failures. Attached Figure Description

[0015] Figure 1 This is a flowchart of the autonomous driving decision-making and planning collaborative method based on fault prediction according to the present invention; Figure 2This is a line graph of the gateway failure probability prediction curve of the present invention, showing the evolution trend of the gateway failure probability over time; Figure 3 This invention presents a series of line graphs showing the time-series changes in the standardized deviation of gateway health indicators, displaying the trend of Z-score standardized deviation of three health indicators—processing latency, packet loss rate, and temperature—over time, as well as the weighted and fused comprehensive anomaly score. Figure 4 This is a mixed graph of line graph and scatter plot of the vehicle mass recursive estimation process of the present invention, which shows the online estimation process of vehicle mass parameters by the recursive least squares algorithm (RLS); Figure 5 This is a bar chart of the suspension load distribution and center of gravity position calculation of the present invention, showing the vertical load distribution measured by the four-wheel suspension displacement sensor; Figure 6 This is a bar graph comparing the updates of the dynamic constraint parameters of the present invention, showing the comparative changes of the dynamic constraint parameters under empty and passenger-carrying conditions. Detailed Implementation

[0016] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0017] This implementation provides a collaborative decision-making and planning method for autonomous driving based on fault prediction, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the operating status data of the vehicle gateway, analyze the time-series changes of health indicators using an abnormal trend detection algorithm, and generate a gateway failure probability prediction curve.

[0018] The system acquires the operational status data of the vehicle gateway within the current time window. This operational status data includes processing latency data, packet loss rate data, and temperature index data. The aforementioned multi-dimensional health index data is input into an anomaly trend detection algorithm for time-series analysis. The anomaly trend detection algorithm performs a fusion analysis of the time-series changes of each health index and outputs a gateway failure probability prediction curve. This gateway failure probability prediction curve represents the probability distribution of gateway failure within a preset future time period.

[0019] The aforementioned abnormal trend detection algorithm includes the following sub-steps: Sub-step 1-1: Calculate the mean and standard deviation within the sliding window for the processed delayed data sequence. Divide the difference between the current value and the mean by the standard deviation to obtain the standardized deviation of the delay index.

[0020] Sub-steps 1-2: Calculate the mean and standard deviation within the sliding window for the packet loss rate data sequence. Divide the difference between the current value and the mean by the standard deviation to obtain the standardized deviation of the packet loss index.

[0021] Sub-steps 1-3: Calculate the mean and standard deviation within the sliding window for the temperature index data sequence, and divide the difference between the current value and the mean by the standard deviation to obtain the standardized deviation of the temperature index.

[0022] It should be noted that sub-steps 1-1 to 1-3 perform Z-score standardization on each health indicator to eliminate the dimensional differences between processing latency (milliseconds), packet loss rate (percentage), and temperature indicator (degrees Celsius), converting different physical quantities into dimensionless standardized deviations, enabling weighted fusion calculations of different types of indicators.

[0023] Furthermore, the duration of the sliding window is determined based on the gateway data sampling period and the fault evolution speed. The number of sampling points contained in the sliding window should be sufficient to reflect the statistical characteristics of the indicator. The sliding window slides forward over time to continuously track the changing trend of the health indicator.

[0024] Sub-steps 1-4: Weight the deviation of the delay index, the deviation of the packet loss index, and the deviation of the temperature index according to preset weights and sum them to obtain a comprehensive anomaly score.

[0025] Furthermore, each weight value in the preset weights is a non-negative number, and the sum of each weight value is equal to 1, to ensure that the scale of the comprehensive abnormal score is consistent with the deviation of the individual item.

[0026] Sub-steps 1-5: Map the comprehensive anomaly score to the [0,1] probability interval using the Sigmoid function to obtain the fault probability value at the current moment; repeat the above calculation for each moment within a preset future time period, and stitch them together to form the gateway fault probability prediction curve.

[0027] Furthermore, the specific form of the Sigmoid function is as follows: ,in This represents the probability of failure. To summarize abnormal scores, is a natural constant; when the comprehensive anomaly score is 0, the failure probability is 0.5. The larger the comprehensive anomaly score, the closer the failure probability is to 1, and the smaller the comprehensive anomaly score, the closer the failure probability is to 0.

[0028] Taking an autonomous taxi operating in an urban environment as an example, the vehicle has an empty weight of 1500 kg, a wheelbase of 2.85 m, and a track width of 1.56 m. After picking up passengers at a certain station, the vehicle continues its journey, during which the system continuously monitors the gateway's operating status. The raw data of the collected gateway health indicators are shown in Table 1: Table 1. Raw data of gateway health indicators The standardized deviation of each metric is calculated using the Z-score standardization formula: Delayed metric deviation: Packet loss metric deviation: Temperature index deviation: The weight vector is set as follows: Calculate the overall abnormality score: Substituting the comprehensive anomaly score into the Sigmoid function yields the fault probability value at the current moment: This value is included in the gateway failure probability prediction curve.

[0029] Step 2: Obtain the load change event trigger signal and vehicle dynamic response data, and use the recursive least squares algorithm to estimate the current mass parameters of the vehicle online.

[0030] The system acquires load change event trigger signals, including door opening / closing status signals and cargo compartment sensor status change signals. Upon detecting a load change event, it acquires the vehicle's acceleration response data and driving torque data for a preset time period before and after the event. The acceleration response data and driving torque data are then input into a recursive least squares algorithm to identify parameters based on the dynamic equations established using Newton's second law, outputting estimated values ​​for the vehicle's current mass parameters.

[0031] Furthermore, the specific form of the dynamic equation is as follows: the driving torque divided by the effective radius of the wheel yields the driving force, and the net force after subtracting rolling resistance and air resistance from the driving force is equal to the product of the vehicle mass and longitudinal acceleration. The recursive least squares algorithm identifies the vehicle mass parameters by solving multiple sets of acceleration and driving torque data pairs simultaneously.

[0032] It should be noted that the recursive least squares algorithm adopts a forgetting factor, which is used to adjust the weight ratio of historical data and current data, enabling the algorithm to quickly track changes in quality parameters.

[0033] Furthermore, the range of the forgetting factor is as follows: The closer the value is to 1, the lower the degree of forgetting of historical data; the lower the value is, the higher the degree of forgetting of historical data and the faster the response to current data.

[0034] In this taxi passenger pick-up scenario, the door sensors detect the sequential opening and closing of the rear left and rear right doors, triggering a load change event. The system collects dynamic response data during the initial acceleration phase after the passengers are seated and stable, including the effective wheel radius. The collected acceleration and driving torque data are shown in Table 2: Table 2 Vehicle Dynamic Response Data Setting a forgetting factor According to the dynamic equation The recursive least squares algorithm iteratively identifies the above four sets of data. Taking the first set of data as an example, the net driving force is: Preliminary estimate of quality After four iterations, the algorithm outputs an estimated current vehicle mass of 1710 kg, an increase of 210 kg compared to the empty mass of 1500 kg, corresponding to the load increase for three passengers.

[0035] Step 3: Obtain suspension displacement sensor data, calculate the estimated value of the center of mass position using the static equilibrium equation, and generate a dynamic parameter update package.

[0036] The system acquires wheel displacement data collected by suspension displacement sensors, including the displacement values ​​of the front left wheel, front right wheel, rear left wheel, and rear right wheel. The wheel displacement data and suspension stiffness parameters are input into the static equilibrium equation to calculate the vertical load on each wheel, thereby estimating the longitudinal and lateral coordinates of the center of gravity. The mass parameter estimates obtained in step 2 and the center of gravity position estimates obtained in this step are combined and packaged to generate a complete dynamic parameter update package, which also includes a timestamp and parameter version identifier.

[0037] Furthermore, the suspension stiffness parameters are known parameters calibrated at the vehicle's factory and stored in the vehicle parameter configuration file. The calculation method for the static equilibrium equation is as follows: multiply the suspension displacement value of each wheel by the corresponding suspension stiffness to obtain the vertical load of each wheel. Based on the principle of torque balance, the longitudinal position of the center of gravity is determined by the product of the front-to-rear axle load ratio and the wheelbase, and the lateral position of the center of gravity is determined by the product of the left-to-right side load ratio and the wheelbase.

[0038] After the passengers are seated, the system collects data from the suspension displacement sensors. The stiffness of each suspension component is 35 N / mm. The collected suspension displacement data for each wheel is shown in Table 3. Table 3 Suspension Displacement Sensor Data Calculate the total load on the front axle based on the static equilibrium equations. Rear axle total load Longitudinal position of the center of gravity (distance from the front axle) Total load on the left side: Total load on the right side: Lateral offset of centroid: That is, the center of mass is slightly offset to the left by 7mm. The system encapsulates the mass parameters 1710kg, longitudinal position of the center of mass 1.507m, and lateral position of the center of mass 0.007m into a dynamic parameter update package, and adds the timestamp "2024-11-15T14:32:18.652Z" and the parameter version identifier "v2024111514321865".

[0039] Step 4: Obtain the parameter version identifier of each domain controller, compare the predicted gateway failure probability value with the threshold, and determine the gateway risk status.

[0040] Obtain the parameter version identifiers currently held by the domain controllers in the perception domain, decision domain, and chassis domain. Extract the predicted fault probability value at the current moment from the gateway fault probability prediction curve generated in step 1, and compare the predicted fault probability value with a preset parameter synchronization trigger threshold. If the predicted fault probability value is lower than the threshold, the gateway is determined to be in a low-risk state, and step 5 is executed; if the predicted fault probability value is not lower than the threshold, the gateway is determined to be in a high-risk state, and step 6 is executed.

[0041] Furthermore, the parameter synchronization trigger threshold ranges from (0,1). The parameter synchronization trigger threshold is set based on the gateway's historical fault statistics and the system's requirements for parameter synchronization reliability. The lower the parameter synchronization trigger threshold, the lower the system's tolerance for gateway fault risks and the more inclined it is to perform pre-synchronization operations in advance.

[0042] Step 5: Under the low-risk state of the gateway, send dynamic parameter update packets to each domain controller through the gateway to generate cross-domain parameter consistency check codes.

[0043] Under low-risk gateway conditions, the dynamic parameter update package generated in step 3 is sequentially sent to the perception domain controller, decision domain controller, and chassis domain controller via the vehicle gateway. Synchronization confirmation responses are received from each domain controller, containing parameter version identifiers that each domain controller has received and stored. Consistency checks are performed on the parameter version identifiers returned by each domain. If all are consistent, a cross-domain parameter consistency check code is generated, which indicates that the parameters of each domain are currently in a synchronized state. Proceed to step 7.

[0044] Furthermore, the method for generating the cross-domain parameter consistency check code is as follows: the parameter version identifiers returned by each domain are concatenated in a fixed order and a hash value is calculated. The hash value is stored in the gateway as a cross-domain parameter consistency check code and is used to verify whether the parameter versions of each domain have changed in subsequent verification.

[0045] Step 6: When the gateway is in a high-risk state, perform parameter pre-synchronization and send parameter snapshot storage instructions to each domain controller.

[0046] If the gateway is in a high-risk state, determine if the gateway is still available. If the gateway is still available, prioritize the parameter pre-synchronization operation: send the dynamic parameter update package generated in step 3 to each domain controller through the gateway, and simultaneously send a parameter snapshot storage instruction to each domain controller. Upon receiving the storage instruction, each domain controller stores the latest parameter version and parameter version identifier in its local non-volatile memory, forming a parameter snapshot. The parameter snapshot is used to support the independent operation of each domain in the event of a subsequent gateway failure. Proceed to step 7.

[0047] Furthermore, if the gateway is already unavailable, each domain controller directly reads the locally stored parameter snapshot, continues to run based on the dynamic parameters in the snapshot, and jumps to execute the independent operation mode and parameter version identifier comparison process in step 8.

[0048] The system queries the parameter version identifier currently held by each domain controller. The perception domain, decision domain, and chassis domain all return the old version identifier "v2024111509153042" (corresponding to the no-load state parameter). The current fault probability prediction value is extracted from the gateway fault probability prediction curve generated in step 1 and is 0.735. The system's preset parameter synchronization trigger threshold is 0.6. Because... The system determines that the gateway is currently in a high-risk state. Since the gateway can still respond to communication requests, it performs a parameter pre-synchronization operation: sending the dynamic parameter update package to each domain controller through the gateway, and simultaneously broadcasting a parameter snapshot storage command. Upon receiving the parameter update package, each domain controller writes the new version parameters (mass 1710 kg, center of mass position [1.507 m, 0.007 m]) and the version identifier "v2024111514321865" into its local non-volatile memory, completing the parameter snapshot storage.

[0049] Step 7: Based on the updated dynamic parameters, recalculate the coefficient matrix of the dynamic equations and output it to the decision domain and chassis domain.

[0050] Based on the updated dynamic parameters (mass parameters and center of mass position parameters) obtained in steps 2 and 3, the coefficient matrix of the vehicle dynamics equations is recalculated. The updated coefficient matrix is ​​input into the trajectory planning module in the decision domain. The trajectory planning module reconstructs the constraints for trajectory optimization based on the new coefficient matrix, including acceleration constraints, steering constraints, and stability constraints. Simultaneously, the updated output parameters are notified to the chassis domain controller, which adjusts the control gain parameters based on the new dynamic parameters.

[0051] Furthermore, the elements in the coefficient matrix are jointly determined by the mass parameter and the centroid position parameter: the mass parameter affects the inertia term coefficient, the longitudinal position of the centroid affects the front and rear axle load distribution coefficient, and the lateral position of the centroid affects the lateral stability coefficient. The updating of the coefficient matrix makes the trajectory planning constraints match the actual dynamic characteristics of the current vehicle.

[0052] Furthermore, the acceleration constraint is specifically defined as follows: the maximum permissible longitudinal acceleration is equal to the maximum driving force divided by the current vehicle mass, and the maximum permissible deceleration is equal to the maximum braking force divided by the current vehicle mass; the steering constraint is specifically defined as follows: the maximum permissible lateral acceleration is determined by the rollover stability limit corresponding to the current center of gravity position; the stability constraint is specifically defined as follows: the absolute value of the difference between the front and rear axle slip angles does not exceed the critical instability threshold determined by the center of gravity position.

[0053] In this embodiment of the application, in order to ensure the coordinated operation of the decision domain and the chassis domain during the parameter transition, step 7 further includes: generating a parameter transition flag; during the parameter transition, the decision domain adopts a conservative trajectory planning strategy to reduce the utilization of dynamic limits; after the chassis domain confirms that the control gain adjustment is completed, the parameter transition flag is cleared and the normal planning strategy is restored.

[0054] Based on the updated mass parameters of 1710 kg and the longitudinal position of the center of gravity of 1.507 m, the system recalculates the trajectory planning constraints. The maximum driving force is set to 8500 N, and the maximum braking force to 12000 N. The updated constraint parameters are shown in Table 4. Table 4 Comparison of Updated Dynamic Constraint Parameters The decision domain trajectory planning module receives the updated coefficient matrix, adjusting the maximum permissible longitudinal acceleration from 5.67 m / s² to 4.97 m / s² and the maximum permissible deceleration from 8.00 m / s² to 7.02 m / s². The chassis domain controller adjusts the longitudinal control gain based on the new mass parameters, changing the proportional gain from: Adjusted to: To compensate for the impact of increased quality on the control response.

[0055] Step 8: When the gateway fails, each domain runs independently based on local parameter snapshots, uses a parameter version identifier comparison algorithm to verify parameter consistency, and outputs a collaborative status identifier.

[0056] In this embodiment of the application, the following steps are also included: When a gateway failure is detected, causing an interruption in cross-domain communication, each domain controller switches to independent operation mode. Each domain controller reads the local parameter snapshot stored in step 6 and continues to execute its respective functional tasks based on the dynamic parameters in the snapshot. The decision domain controller sends a parameter version identifier query request to the chassis domain controller through a backup communication link (such as a direct CAN bus connection). The chassis domain controller returns its locally stored parameter version identifier. The decision domain controller performs a parameter version identifier comparison, checking the string equality between its own domain parameter version identifier and the chassis domain parameter version identifier. If the parameter version identifiers match, it is determined that the parameters of each domain are in a synchronized state, and a "Planning-Control Coordination Normal" status flag is output; if the parameter version identifiers do not match, a "Parameter Version Inconsistency" warning flag is output, and a degraded operation strategy is triggered.

[0057] It should be noted that the degradation operation strategy includes: adopting a more conservative trajectory planning boundary for the decision domain and increasing the safety margin to compensate for execution deviations that may be caused by inconsistent parameters.

[0058] Furthermore, the specific way to increase the safety margin is to reduce the thresholds of acceleration constraints, steering constraints and stability constraints by a preset safety factor, wherein the safety factor is less than 1, so that the planned trajectory maintains a larger margin with the dynamic limit in each constraint dimension, thereby accommodating the control error that may be introduced by parameter inconsistency.

[0059] Approximately 15 seconds after parameter pre-synchronization, the gateway failure probability steadily increased to 0.98 and eventually a communication interruption occurred. Upon detecting a lost network key, each domain controller automatically switched to independent operation mode and read a snapshot of the parameters from its local non-volatile memory. The decision domain controller sent a parameter version identifier query frame to the chassis domain controller via a direct CAN bus link, and the chassis domain controller returned the parameter version identifier stored locally. The comparison results of the parameter version identifiers for each domain are shown in Table 5. Table 5 Comparison Results of Version Identifiers for Each Domain Parameter The decision domain controller performs a string equality check between its own domain version identifier "v2024111514321865" and the version identifier returned by the chassis domain. If the result is consistent, the system outputs a "Planning-Control Coordination Normal" status flag. Each domain continues to execute its respective functional tasks based on the dynamic parameters (mass 1710kg, center of mass position [1.507m, 0.007m]) in the parameter snapshot, ensuring that the trajectory planned by the decision domain remains consistent with the parameter benchmark used by the chassis domain for control during the gateway failure.

[0060] It is understood that data preprocessing methods known to those skilled in the art include data cleaning, data transformation, and data reduction. Data transformation includes type conversion and normalization and standardization. Although the dimensions and types of data were omitted in the description of the preceding embodiments, data preprocessing is a technical knowledge known to those skilled in the art and a prerequisite step in data processing. Therefore, the previously described well-known data preprocessing steps were not described independently.

[0061] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A collaborative decision-making and planning method for autonomous driving based on fault prediction, characterized in that, Includes the following steps: Acquire the operational status data of the vehicle gateway, analyze the time-series changes of health indicators using an abnormal trend detection algorithm, and generate a gateway failure probability prediction curve. Acquire load change event trigger signals and vehicle dynamic response data, and use the recursive least squares algorithm to estimate the current mass parameters of the vehicle online; Acquire suspension displacement sensor data, calculate the estimated value of the center of mass position using the static equilibrium equation, and combine the mass parameters and the center of mass position to generate a dynamic parameter update package; Obtain the parameter version identifier of each domain controller, compare the predicted gateway failure probability value with the threshold, and determine the gateway risk status. When the gateway is in a low-risk state, dynamic parameter update packets are sent to each domain controller through the gateway to generate cross-domain parameter consistency check codes. When the gateway is in a high-risk state, perform parameter pre-synchronization and send parameter snapshot storage instructions to each domain controller; Based on the updated dynamic parameters, the coefficient matrix of the dynamic equation is recalculated and output to the decision domain and chassis domain. When the gateway fails, each domain runs independently based on local parameter snapshots, uses a version identifier comparison algorithm to verify parameter consistency, and outputs a collaborative status identifier.

2. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 1, characterized in that, The process of acquiring the operational status data of the vehicle gateway, analyzing the time-series changes of health indicators using an abnormal trend detection algorithm, and generating a gateway failure probability prediction curve includes: Obtain the vehicle gateway's processing latency data, packet loss rate data, and temperature index data within the current time window; Trend features were extracted from processing delay data, packet loss rate data, and temperature index data, respectively. The trend characteristics of each health indicator are weighted and fused to generate a comprehensive failure probability prediction value; Arrange the comprehensive failure probability prediction values ​​in chronological order to form a gateway failure probability prediction curve that characterizes the probability distribution of gateway failure within a preset time period.

3. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 1, characterized in that, The acquisition of load change event trigger signals and vehicle dynamic response data, and the online estimation of the vehicle's current mass parameters using a recursive least squares algorithm, includes: Acquire load change event trigger signals, which include door opening / closing status signals and cargo compartment sensor status change signals; After detecting a load change event, acquire the vehicle's acceleration response data and driving torque data within a preset time period before and after the event; The longitudinal dynamics equation of the vehicle is established based on Newton's second law, and the acceleration response data and driving torque data are input into the recursive least squares algorithm for parameter identification. The recursive least squares algorithm uses a forgetting factor to adjust the weight ratio of historical data and current data, and outputs an estimated value of the vehicle's current quality parameters.

4. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 1, characterized in that, The process of acquiring suspension displacement sensor data, calculating the estimated center of gravity position using static equilibrium equations, and merging mass parameters and center of gravity position to generate a dynamic parameter update package includes: The displacement data of each wheel collected by the suspension displacement sensor is obtained, including the displacement values ​​of the front axle left wheel, the front axle right wheel, the rear axle left wheel, and the rear axle right wheel. Input the displacement data of each wheel and the suspension stiffness parameters into the static equilibrium equation to calculate the vertical load borne by each wheel; Based on the vertical loads of each wheel and the vehicle's geometric parameters, the estimated position coordinates of the center of mass in the longitudinal and lateral directions of the vehicle are obtained. The estimated mass parameters and the estimated centroid position are merged and packaged to generate a dynamic parameter update package containing a timestamp and version identifier.

5. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 1, characterized in that, The step of performing parameter pre-synchronization and sending parameter snapshot storage instructions to each domain controller when the gateway is in a high-risk state includes: Determine if the gateway is still available. If the gateway is still available, the dynamic parameter update package will be sent through the gateway to the perception domain controller, decision domain controller and chassis domain controller. Send parameter snapshot storage instructions to each domain controller; Upon receiving the storage instruction, each domain controller stores the latest parameter version and version identifier in its local non-volatile memory, forming a parameter snapshot.

6. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 1, characterized in that, The process of recalculating the coefficient matrix of the dynamic equations based on the updated dynamic parameters and outputting it to the decision domain and chassis domain includes: Based on the updated mass parameters and centroid position parameters, the coefficient matrix of the vehicle dynamics equations is recalculated; The updated coefficient matrix is ​​input into the trajectory planning module of the decision domain. The trajectory planning module reconstructs the constraints for trajectory optimization based on the new coefficient matrix. The constraints include acceleration constraints, steering constraints, and stability constraints. The output parameter update notification is sent to the chassis domain controller, which adjusts the control gain parameter based on the new dynamic parameters. A parameter transition flag is generated. During the parameter transition, a conservative trajectory planning strategy is adopted in the decision domain. Once the chassis domain confirms that the control gain adjustment is complete, the parameter transition flag is cleared.

7. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 1, characterized in that, When a gateway fails, each domain operates independently based on a local parameter snapshot, uses a version identifier comparison algorithm to verify parameter consistency, and outputs a collaborative status identifier, including: When a gateway failure is detected that causes cross-domain communication to be interrupted, each domain controller switches to independent operation mode; Each domain controller reads a snapshot of parameters stored locally and continues to execute its respective functional tasks based on the dynamic parameters in the snapshot; The decision domain controller sends a version identifier query request to the chassis domain controller via a backup communication link; The chassis domain controller returns the parameter version identifier stored locally; The decision domain controller executes a version identifier comparison algorithm to compare the version identifier of this domain with the version identifier of the chassis domain; If the version identifiers are consistent, output a normal status indicator for planning-control coordination; if the version identifiers are inconsistent, output a warning indicator for inconsistent parameter versions and trigger a degraded operation strategy.

8. The collaborative method for autonomous driving decision-making and planning based on fault prediction according to claim 7, characterized in that, The degradation operation strategy includes: The decision domain adopts a conservative trajectory planning boundary to increase the safety margin and compensate for execution deviations caused by parameter inconsistencies.

9. A collaborative system for autonomous driving decision-making and planning based on fault prediction, used to execute the collaborative method for autonomous driving decision-making and planning based on fault prediction as described in any one of claims 1-8, characterized in that, include: The gateway status monitoring module is used to acquire the operating status data of the vehicle gateway, analyze the time-series changes of health indicators using an abnormal trend detection algorithm, and generate a gateway failure probability prediction curve. The dynamic parameter estimation module is used to acquire load change event trigger signals and vehicle dynamic response data, estimate the current mass parameters of the vehicle online using the recursive least squares algorithm, calculate the estimated value of the center of mass position using the static equilibrium equation, and generate a dynamic parameter update package. The risk assessment module is used to obtain the parameter version identifier of each domain controller, compare the predicted value of gateway failure probability with the threshold, and determine the risk status of the gateway. The parameter synchronization module is used to send dynamic parameter update packages to each domain controller through the gateway and generate cross-domain parameter consistency verification codes when the gateway is in a low-risk state, and to perform parameter pre-synchronization and send parameter snapshot storage instructions to each domain controller when the gateway is in a high-risk state. The planning and control update module is used to recalculate the coefficient matrix of the dynamic equations based on the updated dynamic parameters and output it to the decision domain and chassis domain. The collaborative operation module is used to control each domain to run independently based on local parameter snapshots when the gateway fails. It uses a version identifier comparison algorithm to verify parameter consistency and outputs a collaborative status identifier.

Images