Method for checking correction of system model in kalman filter

Abstract

The invention relates to a method for checking a correction of a system model in a Kalman filter (1), which is part of a filter network (2) for determining positioning data in a motor vehicle, and determining the positioning data using a first set of sensor data (3), the method comprises at least the following steps: a) detecting a stationary situation of the motor vehicle by means of a stationary detection (12) during operation, thereby generating a stationary signal (13); b) performing a calibration function for the parameters of the system model in the Kalman filter (1) during the stationary situation, the corrected parameters (5) being ascertained; c) when a stationary signal (13) is present, performing a check of the corrected parameter (5) by comparing the corrected parameter (5) with a comparison parameter (6), the comparison parameter (6) being ascertained using a further data source (10), the further data source (10) creating the comparison parameter (6) using the second set of sensor data (4), the comparison parameter (6) being ascertained using the second set of sensor data (4); the second group of sensor data (4) is reduced relative to the first group of sensor data (3), d) if the check performed in step c) results in a negative result, performing an error function (7).

Description

Technical Field

[0001] This invention relates to a novel method for examining the parameterization of system models in Kalman filters. Background Technology

[0002] A Kalman filter (also known as a Kalman-Bucy filter, Stratoovich-Kalman-Bucy-Filter, or Kalman-Bucy-Sratonovich-Filter) is a mathematical filtering model used to iteratively estimate the state of a system based on erroneous input data, particularly sensor data.

[0003] Kalman filters are used to estimate system parameters that cannot be directly measured while optimally minimizing observational errors. A Kalman filter typically maintains an internal mathematical model as an additional condition for the input data; this model flows into the parameter estimation and takes into account the dynamic relationships between system parameters. Equations of motion are stored, for example, in the mathematical model, and variable positions and velocities (which flow into the Kalman filter as input data) are correlated using these equations of motion. Therefore, accurate estimates can be created using (erroneous) input data for position and velocity.

[0004] Kalman filters are particularly useful for iterative estimation of system states based on often erroneous observations. In this context, Kalman filters have proven especially advantageous in several applications where sensor information from different sensors must be combined (or fused) with model information. Furthermore, Kalman filters are commonly used in embedded systems because their computation is advantageous, accurate, and robust. Additionally, microcontrollers can advantageously and efficiently perform Kalman filter computations.

[0005] Kalman filters are particularly used to fuse data from different sensors (which can be used to determine the vehicle's position) to obtain high-precision positioning data. Positioning data here specifically refers to location data as well as data involving velocity and acceleration. Sensors (whose data can be processed or fused using Kalman filters) include, for example, GNSS sensors used to determine positioning via GNSS satellites, inertial sensors, and, for example, wheel sensors and steering angle sensors (used in motor vehicles) to monitor the vehicle's movement via its chassis. Therefore, these sensors acquire prior data for positioning, which is used as input data in the Kalman filter to determine high-precision positioning data. High-precision positioning data can be generated from the described input data by considering the data together in the Kalman filter using the system model, parameters, and / or system states stored in the Kalman filter.

[0006] Kalman filters are particularly useful for obtaining high-precision positioning data for highly automated driving functions of motor vehicles, especially autonomous driving functions.

[0007] In a Kalman filter, it is preferable to continuously correct the internal parameters and system state. The term "corrected parameters" is used here in its entirety to refer to the corrections that occur in a Kalman filter. The quality of the Kalman filter's internal parameters and / or system state is important for improving positioning data by using a Kalman filter.

[0008] However, there is a fundamental need for improvement in the accuracy and confidence of the positioning data obtained using Kalman filters, because for highly automated and autonomous driving applications, very high accuracy and high confidence of the positioning data are desired. Summary of the Invention

[0009] From this point on, a particularly advantageous method should be described for examining the design of the system model and its parameters in a Kalman filter.

[0010] This invention relates to a method for examining the design of a system model in a Kalman filter, which is part of a filtering network for determining positioning data in a motor vehicle, and using a first set of sensor data to determine the positioning data, wherein the method includes at least the following steps:

[0011] a) During operation, stationary recognition is used to identify the stationary status of motor vehicles, thereby generating a stationary signal;

[0012] b) In the Kalman filter, during the rest period, a calibration function is performed for the parameters of the system model, wherein the corrected parameters are obtained;

[0013] c) When a stationary signal is present, the calibrated parameters are checked by comparing them with a comparison parameter. The comparison parameter is obtained using another data source, which creates the comparison parameter using a second set of sensor data, where the second set of sensor data is reduced compared to the first set of sensor data.

[0014] d) If the check performed in step c) yields a negative result, execute the error function.

[0015] The principle of the method described here is to use a stationary phase of the vehicle to check or verify the self-calibration capability of the Kalman filter. The self-calibration capability of the Kalman filter specifically relates to its ability to self-calibrate its internal parameters.

[0016] Navigation systems typically use a stationary phase to perform isolated calibration of individual sensors (e.g., calibrating an inertial sensor system). This approach is based on the premise that, for example, the inertial sensor system should not output acceleration during the stationary phase. Therefore, erroneous signals from the inertial sensor system can be identified during the stationary phase, such as zero-point drift, and offset parameters to correct this zero-point drift can be imported if necessary.

[0017] The method described here goes a step further. If a quiescent phase exists, then according to step c), the system model of the filter itself is checked or verified as a whole in the Kalman filter.

[0018] To date, the system model of the Kalman filter has been validated, particularly during development, using a (vehicle-mounted) reference system. Therefore, checks on the discrepancies estimated by the filter (and consequently the system model) are performed only during development and not in the navigation filter located in the field. With the method proposed here, validation can be performed during runtime for each navigation device located in the field. Thus, changes in the aging decisions of the measuring equipment and sensor failures can be detected to a greater extent.

[0019] Calibration of the Kalman filter is typically performed continuously due to its characteristics. The calibration function described in step b) is preferably a (common) self-calibration of the Kalman filter, which is continuously performed during Kalman filter operation. The calibration according to step b) is a process inherent to the Kalman filter itself, performed by the filter itself in a static state, and preferably also in other operational phases (non-static phases). In particular, calibration is achieved through automatic error compensation generated by the structure of the Kalman filter itself. In other words, errors that occur are implicitly identified, leading to changes in the system model, stored in the system model, and used in the future to correct sensor signals and generate improved positioning data.

[0020] Here, due to the structure of the Kalman filter, the stationary phase plays a crucial role because the ratios of the various signal sources processed by the Kalman filter differ in these phases compared to the moving phase. For example, inertial sensors typically provide no data or only data based on sensor inaccuracies. Therefore, the system state of the system model in the Kalman filter changes, particularly in the stationary phase, thereby correcting the parameters of the system model. The parameters of the system model that change based on the internal correction mechanism in the Kalman filter are here "generally" referred to as the corrected parameters.

[0021] The parameters of a Kalman filter (and consequently the corrected parameters) are typically the individual values ​​of the subsequent matrix and vector:

[0022] - The covariance matrix of the Kalman filter;

[0023] - The estimated state vector (x) of the Kalman filter;

[0024] - The parameters and / or structure of the system model of the Kalman filter; and

[0025] - Noise vector, which describes the system noise of the Kalman filter.

[0026] In the sense of the first and second groups, the multiple sets of sensor data refer to the sensor data themselves. These multiple sets of data are preferably further processed using filters, such as Kalman filters. The signals flowing into the first group preferably overlap with the signals flowing into the second group. Particularly preferably, the second group is a subset of the first group.

[0027] Compared to the calibration described in step b), the check performed in step c) does not interfere with the Kalman filter. The check does not change the internal parameters (system state) of the Kalman filter. Instead, it compares the internal (corrected) parameters of the Kalman filter with other parameters. The check is preferably binary and can produce either a positive or negative result. If necessary, the check can also output more than two (binary) result values. However, it is preferable that there is a negative result value that triggers the error function in step d), and a positive result value that does not trigger the error function in step d). If the check yields a positive result, the filter network works as expected. The error function in step d) does not need to be triggered. If the check yields a negative result, the filter network does not work as expected. In particular, the self-calibration of the continuously corrected parameters or the output corrected parameters of the Kalman filter is compromised.

[0028] The purpose of the method is to trigger the erroneous function explained in step d) if, according to step c), the Kalman filter is identified as not performing sufficiently good self-calibration. For example, an erroneous function could be identified as follows: the positioning data obtained using the Kalman filter is generally compromised and should be discarded or disregarded for highly automated or autonomous driving functions.

[0029] For the comparison according to step c), it is preferable to obtain the comparison parameters. This utilizes a different data source than the Kalman filter. In the simplest case, the comparison parameters can be parameters obtained directly using sensors (e.g., inertial sensors). Inertial sensors should generally not record the vehicle's motion and / or acceleration or velocity during the stationary phase (for safety identification). Therefore, deviations can be identified (when data from inertial sensors is used directly as comparison parameters). These deviations can be used to infer errors in the Kalman filter's system model.

[0030] Preferably, the other data source is designed such that the comparison parameters available from the data source ensure simple comparability with the calibrated parameters.

[0031] The method described is based on performing genuine and reliable stillness identification in step a).

[0032] It has been confirmed that wheel sensor signals alone are generally insufficient to safely identify a stationary state using the described method. According to the described method, the calibrated parameters of the Kalman filter are examined only when the object is stationary.

[0033] The method is particularly advantageous if the first set of sensor data used by the Kalman filter includes at least GNSS signals from at least one GNSS sensor and inertial sensor signals from an inertial sensor.

[0034] Particularly preferably, the sensor data in the first group also includes WSS signals from the wheel sensors, and even more preferably, it also includes data from sensors that monitor the steering angle of the vehicle.

[0035] Furthermore, the method is advantageous if changes in the aging determination of the GNSS sensor and / or inertial sensor are identified by examining the calibrated parameters in step c).

[0036] In particular, the described method can be used to identify changes in the aging determination of the sensor. Specifically, it can also identify changes in the aging determination of the sensor that were not considered when designing the Kalman filter.

[0037] Furthermore, this method is advantageous if data from GNSS sensors are not considered when identifying stationary conditions in step a).

[0038] Furthermore, the method is advantageous if sensor data from at least one inertial sensor and / or a rotary sensor are used to identify the stationary state in step a).

[0039] To identify stationary states, it is preferable to fuse data from multiple sensors. For example, the following data could be considered for identifying stationary states:

[0040] - Camera images, where, for example, a camera image that is completely or partially still can be evaluated as a stillness indicator;

[0041] - Time-based data, where, for example, a specific time can be evaluated as an indication of stillness;

[0042] - When the vehicle is in the workshop, identify when the workshop is stationary; and / or

[0043] - Data related to the status of the vehicle, such as internal system indicators indicating that the ignition is off.

[0044] In principle, this list is not definitive. There are multiple possibilities for detecting a stationary vehicle. What is important for the described method is to combine or correlate the different data considered to enable the identification of stationary vehicles with high confidence.

[0045] Furthermore, this method is advantageous if, for the examination of the corrected parameters checked in step c), at least the following parameters are checked:

[0046] - The covariance matrix of the Kalman filter;

[0047] - The estimated state vector of the Kalman filter;

[0048] - The parameters and / or structure of the system model of the Kalman filter; and

[0049] - A noise vector describing the system noise of the Kalman filter.

[0050] Furthermore, this method is advantageous if other data sources are strapdown filters set in the filtering network in addition to the Kalman filter, wherein the strapdown filter uses data from at least one inertial sensor to generate comparison parameters for checking the calibrated parameters in step c).

[0051] A strapdown filter is a filter present in the filtering network other than a Kalman filter, and it preferably performs the processing of the second set of sensor data in parallel with the Kalman filter.

[0052] Furthermore, preferably, the strapdown filter also obtains data from the Kalman filter as input data. Particularly preferably, the strapdown filter obtains the output data obtained by the Kalman filter at a previous (previous) time point as input data.

[0053] The core of this invention is a static verification scheme that uses parallel strapdown filters to verify filter differences. Its advantage lies in its ability to detect and react to sensor malfunctions or deviations in filter modeling (e.g., due to sensor aging).

[0054] The strapdown filter preferably processes only data from the inertial sensor system. The strapdown filter preferably has a system model that corresponds in its structure to the system model of the Kalman filter, is periodically adapted to the system model in the Kalman filter, and is updated or propagated from a previous time point using only the second set of data (preferably only data from the inertial sensor). The term "propagation" here means that the system model is updated from a previous time point using only the data from the inertial sensor system. Therefore, the term "propagation" here specifically means, in stillness, that is, using only inertial sensor data, checking how the state vector changes when it is updated using the strapdown filter. Here, the change can only occur within the expected range. In particular, a simultaneous increase in the difference and its respective associated parameters can be expected. In principle, the difference should always contain the associated parameters. Preferably, the check of the corrected parameters includes a comparison of the parameters and their differences. An error or escape of a parameter from the difference will trigger the error function described in step d). The system models in the strapdown filter and the Kalman filter respectively model the escape of errors in the absence of new input data. During the stationary phase, this escape can be measured because it occurs here, while no additional positional change flows into the parameters. Using the described method, according to step c), it is determined in principle whether the actual escape of the error in station (which inevitably always occurs due to the inaccuracies of the inertial sensor system and its data processing) fits into or is correctly set within the system model. Therefore, it can be determined whether the system model is suitable for correctly considering the behavior of the inertial sensor system during operation.

[0055] In a preferred embodiment of the method, the strapdown filter operates continuously and independently of static conditions. However, the check described in step c) is performed only under static conditions, which are preferably identified independently of the strapdown filter.

[0056] Strapdown is the standard term for algorithms associated with fixed acceleration and yaw rate sensors.

[0057] Furthermore, it is advantageous that the state vector parallel to the Kalman filter utilizes the propagation path of the strapdown filter, and at least one covariance matrix of the Kalman filter is propagated using the strapdown filter as the propagated covariance matrix.

[0058] Preferably, the inertial sensor signal (if present) is corrected using an estimated sensor fault from the state vector x of the Kalman filter via a strapdown filter. The corrected inertial sensor values ​​are used in the strapdown filter to propagate the initial position, velocity, and orientation over time. In parallel, the covariance matrix is ​​propagated using the system model, inertial sensor data, and system noise.

[0059] Then, the propagated covariance matrix is ​​compared with the propagation status of position, velocity, and orientation.

[0060] The initialization error and measurement error (inaccuracy) of an inertial sensor system ensure that the propagation error of the state increases. In a stationary state, the propagation error can be calculated by subtracting the initial value from the current propagation state, assuming constant position and orientation and zero velocity. In an ideally designed Kalman filter, three times the standard deviation (from the filter's covariance matrix) should contain the estimation error. This cannot be checked in a conventional system without a reference system.

[0061] This check can be achieved by comparing the propagation error and propagation difference (also described above). If this condition is violated over a longer period (tuning factor), it is assumed that the system model no longer accurately describes the sensor system. A response can be made based on the magnitude of the deviation, either by adjusting the system model parameters during runtime or by outputting an error signal for larger deviations.

[0062] Furthermore, it is advantageous to perform a comparison between the covariance matrix obtained using the Kalman filter and the state vector obtained using the Kalman filter, and the covariance matrix and state vector propagated using the strapdown filter, in order to check the corrected parameters according to step c).

[0063] The comparison of the covariance matrix and the state vector does not need to be complete; that is, for this method, it is not necessary to compare every element of the matrix and vector. Individual values ​​can also be compared, for example, only the diagonal values ​​of the matrix. Differences are typically contained within the diagonal values ​​of the covariance matrix. By comparing these differences, a check can be performed according to step c).

[0064] A control device including a processor should also be described herein, which is adapted / configured to perform the described methods.

[0065] Preferably, the described sensor data source (GNSS sensor, inertial sensor, wheel sensor, etc.) is connected to the control device.

[0066] Preferably, the control device relates to generating high-precision positioning data from sensor data and periodically executing the described method to ensure particularly high quality of the positioning data and to execute the error function described according to step d) when the quality of the positioning data cannot be ensured.

[0067] In addition, a computer program product should be described, which includes instructions that, when executed by a computer, cause the computer to perform the described method.

[0068] The computer program product is preferably mounted on the described control device.

[0069] A computer-readable storage medium is further described, comprising instructions that, when executed by a computer, cause the computer to perform the described method or steps of the described method. Attached Figure Description

[0070] The present invention and its technical environment will then be explained in detail with the aid of the accompanying drawings. The drawings illustrate preferred embodiments, but the invention is not limited to these embodiments. In particular, it should be noted that the drawings and, especially, the size ratios shown in the drawings are merely schematic. Wherein:

[0071] Figure 1 A filtering network for performing the method is shown;

[0072] Figure 2 A flowchart of the method is shown;

[0073] Figure 3 An example is shown where a deviation occurs between the state vector of the Kalman filter and the state vector propagated using a strapdown filter; and

[0074] Figure 4 Another example is shown where a discrepancy occurs between the state vector of the Kalman filter and the state vector propagated using a strapdown filter. Detailed Implementation

[0075] Figure 1 A filtering network 2 with a Kalman filter 1 is shown, which processes a first set of input data 3 to determine positioning data. The first set of input data 3 includes at least data from the GNSS sensor 8, data from the inertial sensor 9, and possibly additionally data from the rotary sensor 11.

[0076] Kalman filter 1 has an internal system model with multiple parameters, which are typically in the form of a state vector X, a covariance matrix 14, and a noise vector 15, and may exist as additional parameters if necessary. The system model of Kalman filter 1 is used to generate accurate positioning data from input data. The Kalman filter continuously performs a calibration function to correct the parameters of the system model, enabling the generation of accurate positioning data even with system biases in the input data. Kalman filter 1 preferably generates continuously corrected parameters 5 for the system model, which are specifically located in the state vector X, the covariance matrix 14, and the noise vector 15.

[0077] In addition to the Kalman filter 1, the filter network 2 also has a strapdown filter 10, which operates in parallel with the Kalman filter 1 and processes the second set of input data 4. The second set of input data 4 includes data from the inertial sensor 9, and in particular excludes data from the GNSS sensor 8.

[0078] The strapdown filter 10 propagates the parameters of the Kalman filter 1 based on the second set of input data 4. This means that the strapdown filter 10 propagates the parameters of the Kalman filter 1, particularly independently of the data from the GNSS sensor 8. Therefore, comparison parameters 6 are generated by the strapdown filter 10, namely, the propagated state vector X, the propagated covariance matrix 16, and the propagated noise vector 17. Preferably, the comparison parameters 6 occur continuously and in parallel with the operation of the Kalman filter 1 using the propagation of the strapdown filter 10.

[0079] Filtering network 2 also has a stillness recognition 12, which is based on (here in Figure 1 The system identifies stationary conditions (without detailed limitations) and outputs a stationary signal 13. Then, if a stationary condition exists and a stationary signal 13 is output, a check function 20 is activated, which performs a comparison between the corrected parameter 5 and the comparison parameter 6. In a stationary condition, only the expected deviation should occur between the corrected parameter 5 and the comparison parameter 6, and in particular, deviations exceeding limits should not occur. If no deviation is set, the check function 20 yields a negative result and executes the error function 7. If necessary, the error 19 estimated by the Kalman filter 1 can be transmitted to the strapdown filter 10, and the strapdown filter 10 can be used to generate corrections for further processing of the inertial sensor signal 18.

[0080] Figure 2 A flowchart of the method is shown. Step a) is identifiable, used to identify a stationary condition. If a stationary condition is identified, the result of step b) (the corrected parameters of Kalman filter 1) is then subjected to a check according to step c). Step c) is preferably divided into sub-steps c1), c2), and c3). Steps b) and c1) are also performed continuously, independently of step a), if necessary, wherein if step a) indicates the existence of a stationary condition, then preferably only the results of these steps (i.e., the corrected parameters 5 of Kalman filter 1 as the result of step b) and the corresponding comparison parameters 6 of strapdown filter 10 as the result of step c1) for steps c2) and c3).

[0081] In step c2), it is preferable to compare the corrected parameter 5 with the comparison parameter 6. Preferably, the corrected parameter 5 and the comparison parameter 6 each comprise a plurality of individual parameters, wherein the comparison specifically includes comparing the respective individual parameters composed of the corrected parameter 5 or the comparison parameter 6.

[0082] In step c3), preferably, the differences in the parameters are compared. If steps c2) and c3) indicate that the comparison is unsuccessful, then in step d), preferably, error function 7 is executed.

[0083] Figure 3 and Figure 4 The diagrams illustrate how the deviation between the corrected parameter 5 and the comparison parameter 6 can form during a static condition, and which deviation might lead to an identification error or a negative check result when checking the corrected parameter 5 with the help of the comparison parameter 6.

[0084] Difference 23 and parameter 24 are recorded on parameter axis 22 via time axis 21, with the parameter being the monitored error. For simplified understanding: parameter 24 is, for example, a position estimate. Difference 23 describes the uncertainty of this position estimate 24. It can be seen that parameter 24 is based on... Figure 3 The expected difference 23 corresponds to and remains below that difference. This indicates that parameter 24 can be estimated here using the difference 23. Correct or permissible behavior can be determined. The check according to step c) of the method described above shows that the check yields a positive result. Figure 4 A deviation of 25 indicates a problem or systemic error in the filter model, or may mean a negative result when checking the corrected parameter 5 using comparison parameter 6. Here, the increase in parameter 24 is stronger than the difference 23. The check according to step c) of the method described above indicates that the check yields a negative result.

Claims

1. A method for checking the correction of a system model in a Kalman filter (1), said Kalman filter being part of a filtering network (2) for determining positioning data in a motor vehicle and using a first set of sensor data (3) to determine the positioning data, wherein, The method includes at least the following steps: a) During operation, the stationary status of the motor vehicle is identified by stationary identification (12), thereby generating a stationary signal (13). b) In the Kalman filter (1) during the resting state, a calibration function for the parameters of the system model is performed, wherein the corrected parameters (5) are obtained. c) When a stationary signal (13) is present, the corrected parameter (5) is checked by comparing it with the comparison parameter (6), wherein the comparison parameter (6) is obtained using another data source (10), wherein the other data source (10) creates the comparison parameter (6) using a second set of sensor data (4), wherein the second set of sensor data (4) is reduced relative to the first set of sensor data (3). d) If the check performed in step c) yields a negative result, execute the error function (7).

2. The method according to claim 1, wherein, The first set of sensor data (3) used by the Kalman filter (1) includes at least GNSS signals from at least one GNSS sensor (8) and inertial sensor signals from an inertial sensor (9).

3. The method according to any one of the preceding claims, wherein, Changes in the aging determination of the GNSS sensor (8) and / or the inertial sensor (9) are identified by checking the calibrated parameter (5) as described in step c).

4. The method according to any one of the preceding claims, wherein, For identifying the stationary state in step a), data from the GNSS sensor (8) is not considered.

5. The method according to any one of the preceding claims, wherein, For identifying the stationary state in step a), sensor data from at least one inertial sensor (9) and / or sensor data from a rotary sensor (11) are used.

6. The method according to any one of the preceding claims, wherein, For the inspection of the corrected parameter (5) checked in step c), at least the following parameters shall be checked: - The covariance matrix (14) of the Kalman filter (1). - The estimated state vector (x) of the Kalman filter (1); - The parameters of the system model of the Kalman filter (1); and - Noise vector (15) describing the system noise of the Kalman filter (1).

7. The method according to claim 6, wherein, The other data source (10) is a strapdown filter (10) set in the filtering network (2) in addition to the Kalman filter (1), wherein the strapdown filter (10) uses data from at least one inertial sensor (9) to generate a comparison parameter (6) for checking the corrected parameter (5) in step c).

8. The method according to claim 7, wherein, The comparison parameter (6) generated by the strapdown filter (10) includes at least one propagated state vector (y), which corresponds to the state vector (x) estimated by the Kalman filter (1) and propagates the state vector (x).

9. The method according to any one of claims 6 to 8, wherein, The state vector (x) parallel to the Kalman filter (1) is propagated by the strapdown filter (10), and at least one covariance matrix (14) of the Kalman filter (1) is propagated by the strapdown filter (10) as the propagated covariance matrix (14).

10. The method according to any one of claims 6 to 9, wherein, In order to check the corrected parameter (5) according to step c), a comparison is performed between the covariance matrix (14) obtained by the Kalman filter (1) and the state vector (x) obtained by the Kalman filter (5) and the covariance matrix (14) and the state vector (y) propagated by the strapdown filter (10).

11. A control device comprising a processor adapted / configured such that the processor performs the method according to any one of claims 1 to 10.

12. A computer program product comprising instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 10.

13. A computer-readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform the steps of the method according to any one of claims 1 to 10.

Images