Monitoring and detection of sensor errors in inertial measurement systems
Through the combination of multiple hypothesis method and Kalman filter, the relative signal difference between the two inertial measurement units and the correction of the object tracking unit is solved, and the high sensitivity identification of micro faults is achieved and switch to redundant sensors to ensure the accuracy and safety of the vehicle's movement status.
Patent Information
- Application Number
- CN202380033590.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-04-12
- Filing Date
- 2023-02-27
- Publication Date
- 2024-11-22
AI Technical Summary
The existing inertial measurement unit fault detection methods have low sensitivity in identifying small faults and rotational degree of freedom deviations, making it difficult to meet the high requirements of autonomous driving and driving assistance systems for safety and accuracy.
Using multiple assumptions, using the relative sensor signal difference between the two inertial measurement units and the error detection of the Kalman filter, the vehicle's self-motion state is detected and corrected through the object tracking unit, and identified and switched to the redundant sensor to isolate it. Fault.
It realizes high sensitivity detection and identification of small amplitude faults in the inertial measurement unit, improves the accuracy and safety of fault diagnosis, and ensures accurate calculation of vehicle motion status.
Smart Images

Figure CN119013533A_ABST
Abstract
Description
[0001] The invention relates to a method for detecting faults in an inertial measurement unit according to the preamble of claim 1 .
[0002] An inertial measurement unit (IMU) consists of an accelerometer and a gyroscope and is used to measure the specific forces (acceleration) and rotational velocity of a rigid body in space. The IMU plays a decisive role in estimating the vehicle's motion state. This motion state is required in vehicle control systems, driver assistance systems and autonomous driving, for example, to accurately interpret information from environmental sensors (cameras, lidar, radar). Devices and methods for calculating the vehicle's egomotion based on the data from the IMU in combination with additional vehicle sensors have established the concept of a "vehicle motion observer (VMO)" [5]. Figure 1 Such a device is shown in . Figure 1 An example diagram of using an inertial measurement unit 1 for motion estimation and navigation according to the prior art is shown. The inertial measurement unit 1 sends a rotation speed signal via a gyroscope and a specific force signal via an accelerometer [9]. In a sensor fusion unit 2, motion parameters (such as azimuth α, velocity v and / or position P) can be calculated by integrating these parameters and corrected with the help of other sensors or sensor units 3 (such as odometer, magnetometer, barometric altimeter, steering angle, vehicle level and / or GNSS receiver). The result of the above data fusion is referred to as "estimated motion state" or simply "estimated motion" in the following text. In cases where there is a possibility of confusing vehicle motion with object motion in the vehicle environment, expressions such as "self-motion state" or "estimated self-motion" should be used in relation to the vehicle.
[0003] An IMU sensor is considered to be faulty if the sensor loses a specific (i.e., predetermined by design) association between the physical measurement variable and the output sensor signal, which results in an unspecific measurement behavior. This makes the associated IMU sensor unusable. If a fault in the sensor is propagated undetected into the motion calculation, it can lead to serious errors or large inaccurate values in the useful signal and thus to safety risks in safety-related applications.
[0004] Vehicle control systems, driver assistance systems and autonomous driving are all applications with extremely high safety requirements. Therefore, the system must automatically detect and locate IMU sensor failures and exclude the erroneous sensor from motion calculations within the system-specified time (FTTI, fault-tolerant time interval), such as by switching to a redundant sensor (backup).
[0005] The class of automotive IMU systems considered here (automotive-grade IMUs) already has basic monitoring functions, especially for electrical faults. This type of monitoring is referred to as "unilateral monitoring" in the following text, because the monitoring is based on information from only one sensor at a time. The low sensitivity of the unilateral approach means that only relatively large faults can be detected, which means that the monitoring thresholds are often too high for driver assistance systems and automated driving. This diagnostic coverage is too low for even minor faults that can also pose a safety risk. The diagnostic coverage can be greatly increased by introducing redundant sensor elements.
[0006] The prior art for arranging sensors in a simple redundant manner is summarized in [4]. Figure 2 A solution to this problem is shown in [4]. Here, the two redundant inertial measurement units 4 (IMU A) and 5 (IMU B) are first calibrated with respect to their system errors (bias, sensitivity, misalignment, etc.), and then the corresponding sensor values calibrated in this way are compared. If the difference between the measured values exceeds a threshold value, a fault may have occurred in the corresponding sensor pair. However, this fault cannot be attributed to either of the two sensors, because an additional reference parameter is required here. The solution to this problem in [4] is to monitor the changes of the corresponding Kalman filters used to estimate the self-motion. Since the motion state predicted by the IMU is fused with the remaining vehicle sensors (wheel speed, steering angle) in the two Kalman filters, and the vehicle sensors represent an independent third-party source of information, IMU errors can be located by evaluating the changes (i.e., the weighted difference between the predicted value and the measured value). The disadvantage of this method is that the measurement variables obtained from the vehicle sensors are basically limited to the translational velocity, so the monitoring method described in [4] is not equally applicable to all degrees of freedom of motion of the IMU. For example, the sensitivity to faults that cause deviations in the rotational degrees of freedom is low. In addition, the detection of small faults is limited by the resolution of the wheel speed sensor. Especially in object tracking, when fusing the estimated self-motion state and environmental information, the requirements for identifying IMU faults are somewhat higher, the detection threshold is smaller, and the fault identification time is shorter, which may not be met by the above methods.
[0007] Although the general requirement that faults in the self-motion estimate can be identified when fusing environmental sensors is presented in [7] and [6], no specific implementation is provided. Moreover, this requirement is mainly aimed at faults that often lead to physically unreasonable information about the vehicle's motion state. The motion state in such faults is physically possible but incorrect, but no explanation is given for this. Furthermore, the solution proposed here does not involve redundant inertial measurement units and self-motion calculations. However, due to the low fault threshold, redundancy of inertial sensors is necessary to shorten the fault identification time and increase the sensitivity of fault diagnosis.
[0008] The object of the present invention is to provide a new method for detecting faults in an inertial measurement unit.
[0009] This object is achieved according to the invention by a method having the features of claim 1 .
[0010] Advantageous embodiments of the invention are the subject matter of the dependent claims.
[0011] The method according to the invention for detecting faults in an inertial measurement unit, which is used in a vehicle for measuring angular velocity and specific force, uses a multiple hypothesis method, wherein two inertial measurement units are used, each of which has a plurality of sensors including accelerometers and gyroscopes.
[0012] According to the present invention, a first inertial measurement unit is used as a primary inertial measurement unit, and a second inertial measurement unit whose performance may be lower than that of the first inertial measurement unit is used as a secondary inertial measurement unit; wherein the measurement value of the primary inertial measurement unit is used as a reference value to compensate the measurement value of the secondary inertial measurement unit by estimating the error model parameters relative to the primary inertial measurement unit, so as to identify a fault by comparison based on two corresponding sensor signals of each of the two inertial measurement units; the following one of the two inertial measurement units is detected as having a fault: the motion estimate calculated by the one inertial measurement unit in the corresponding downstream Kalman filter and in the object tracking unit leads to an erroneous prediction of the calculated position of an object in the vehicle environment.
[0013] In one embodiment, sensor values of the inertial measurement unit are input to a monitoring unit and error states are transmitted from the monitoring unit to the object tracking unit so that in the presence of a fault multiple hypotheses of motion states can be checked, while only one motion state is used in normal operation.
[0014] In one embodiment, the vehicle's ego-motion is described by a vehicle pose and the vehicle pose is transmitted to an object tracking unit.
[0015] In one embodiment, the object tracking unit predicts a new object position of a single object or new object positions of multiple objects in the vehicle body coordinate system from the calculated vehicle pose hypothesis and a previously calculated object position of a single object or previously calculated object positions of multiple objects.
[0016] In one embodiment, the object positions of one or more objects measured by a vehicle environment sensor system are compared with the object positions predicted via a self-motion hypothesis of the vehicle, which is determined with the aid of a Kalman filter, and the following vehicle self-motion hypothesis is detected as erroneous: in which the corresponding deviation between the predicted object position and the measured object position exceeds a threshold value.
[0017] In one embodiment, in the event of a fault in one of the inertial measurement units, a switch is made to the other inertial measurement unit if the fault is identified by comparison with a redundant sensor of the other inertial measurement unit and the self-motion hypothesis calculated using the erroneous sensor values is rejected as an error.
[0018] In one embodiment, the monitoring results of both inertial measurement units are incorporated into the error detection of the object tracking unit, in which case a fault is only identified if the deviation of the object position predicted by one of the hypotheses exceeds a threshold while the deviation of the object position predicted by the other hypotheses does not exceed the threshold and an error is simultaneously identified by sensor comparison of the IMU error detection unit.
[0019] In one embodiment, only one object or multiple objects identified as stationary are taken into account in the error detection, and a fault is detected for a hypothesis in which the predicted object position changes significantly while for other hypotheses the position remains constant.
[0020] The present invention exploits the higher sensitivity of object tracking methods to errors in ego-motion estimation and is specifically designed for error detection, thereby also being able to detect IMU faults of very small magnitude.
[0021] Embodiments of the present invention are explained in more detail below with reference to the accompanying drawings.
[0022] in:
[0023] Figure 1 A schematic diagram showing a typical example of estimating vehicle motion parameters using an inertial measurement unit (IMU),
[0024] Figure 2 shows a schematic diagram of monitoring IMU channels in a simple redundant manner,
[0025] Figure 3A schematic diagram of a solution according to the invention is shown, comprising a solution for evaluating self-motion and error detection in object tracking,
[0026] Figure 4A and Figure 4B A schematic diagram illustrating the basic concept of the present invention in two-dimensional space is shown.
[0027] In all the figures, parts which correspond to one another are provided with the same reference numerals.
[0028] like Figure 2 As shown, in a method for detecting a fault F in an inertial measurement unit in a simple redundant manner, two inertial measurement units 4 and 5 are used, each of which has a plurality of accelerometers and gyroscope sensors.
[0029] In the figure, the two inertial measurement units 4 and 5 are also labeled as IMU A and IMU B, respectively. Each inertial measurement unit provides a vector with three angular velocity signals and a vector with three specific force signals To simplify the representation of these signals, superscripts and subscripts are omitted. Then we can refer to the sensor signal as and
[0030] exist Figure 2 In the embodiment, the first inertial measurement unit 4 is used as the main inertial measurement unit, and the second inertial measurement unit 5, which may have lower performance than the first inertial measurement unit 4, is used as the secondary inertial measurement unit or backup inertial measurement unit; wherein the measurement value of the main inertial measurement unit 4 is used in the monitoring unit 12 to estimate the system error parameters in the secondary inertial measurement unit 5 relative to the main inertial measurement unit 4 in the model, and to detect the signal deviation between the signals compensated in this way; wherein the excessive signal deviation may be caused by a potential fault F in one of the two inertial measurement units 4 or 5. However, it is not possible to identify which of the two inertial measurement units 4 or 5 is affected by the fault F based on the detection of the monitoring unit 12 alone. However, the specific sensor pair affected can be identified.
[0031] In order to calculate the vehicle's motion state, the angular velocity signals of the two inertial measurement units 4 and 5 are and ratio signal As input variables, they are provided to Kalman filters 13 and 14 (vehicle motion observer, abbreviated as VMO) and are fused with the measured variables of other vehicle sensors (such as wheel speed and steering angle) in the Kalman filters 13 and 14, wherein both Kalman filters 13 and 14 use the same measured variables. Kalman filters 13 and 14 determine the motion states BZ A and BZ B for the corresponding inertial measurement units 4 and 5, respectively. Figure 1 As can be seen from FIG. 1 , data fusion of the signal of the inertial measurement unit 1 with the remaining vehicle sensors as described above is a standard procedure for determining vehicle motion parameters.
[0032] like Figure 2 As shown, the solution proposed in document [4] as a prior art provides that changes of the two Kalman filters 13 and 14 are monitored in the monitoring unit 15 and used together with the information from the monitoring unit 12 for error detection IF in the error detection unit 16 .
[0033] The following formula symbols are used in the description below:
[0034] t 0 Time
[0035] u, u′ Vehicle pose estimated from IMU and vehicle sensors
[0036] Vehicle heading increment
[0037] Vehicle position increment
[0038] The rotation matrix calculated from the azimuth increment
[0039] H(u) is the homogeneous transformation from the ground-fixed coordinate system to the vehicle body coordinate system.
[0040] x object position
[0041] z Object position measurements obtained using environmental sensors
[0042] K Kalman filter amplifier
[0043] v Changes
[0044] S measurement error covariance matrix
[0045] P - The state error covariance matrix before the correction measurement (a priori)
[0046] Covariance matrix of the R measurements
[0047] κ threshold
[0048] Ψ The yaw angle of the vehicle relative to the ground-fixed coordinate system
[0049] ΔΨ Yaw angle increment
[0050] ω z Yaw rate
[0051] δω z Yaw rate error
[0052] v x Vehicle longitudinal speed
[0053] v y Vehicle lateral speed
[0054] Δt The time interval between two measurements
[0055] l r The distance between the body coordinate system and the rear axle of the vehicle.
[0056] and Figure 2 Different from the prior art in the present invention, according to the present invention, Figure 3 As shown in the figure, the calculation results of the vehicle motion state (correct vehicle posture u and incorrect vehicle posture u‘ ) and the information from the monitoring unit 12 are input to the object tracking unit 17. In the object tracking unit 17, the motion state is used to predict the position of the object in the vehicle environment in the prediction module 18. x - Then, the predicted position is corrected in the object position correction unit 19. x - Measure position of objects from environmental sensors z Compare and make corresponding corrections (corrected positions of objects in the vehicle environment x + ). If one of the inertial measurement units 4 and 5 fails F, the predicted position x - and measure or correct position x + There will be a large deviation between them. Then, combined with the independent IMU error detection in the monitoring unit 12, it can be used to identify and locate the IMU fault F (assigned to the relevant IMU) and, if necessary, isolate the IMU error by switching to the redundant solution (i.e. the corresponding error-free inertial measurement unit 4 or 5). If there is no fault F in both inertial measurement units 4 and 5, then the predicted position x - and measure or correct position x + There is a very good fit between them.
[0057] The detection problem can be solved by transmitting two redundant ego-motion estimates further to the object tracking unit 17. They can serve here as hypotheses for the ego-motion of the vehicle.
[0058] The object tracking unit 17 typically receives data about objects in the vehicle environment from various sensor sources, such as cameras, radars, lidars, and ultrasound. Typically, this data is fused and tracked over a long period of time to extract object information, track the temporal motion of these objects, and predict their future motion. For example, dynamic objects such as vehicles and pedestrians can be tracked. Another task of the object tracking unit 17 is to identify static objects, such as small obstacles on the roadway.
[0059] When fusing objects in the vehicle environment, the vehicle's self-motion must be compensated so that the correct position and speed of these objects can be calculated. To this end, accurate and reliable motion estimates from VMO13 or 14 are input into the object tracking unit 17. In the object tracking unit 17, the self-motion state is used to predict the position of surrounding objects between observations by environmental sensors. Incorrect self-motion states can lead to errors in the prediction of the positions of these objects. For example, this may cause a static object to change its position at different observation time points. All tracked static objects are affected by this effect at the same time. This leads to the basic idea of the present invention: if an object in the vehicle environment that was previously classified as stationary (such as a wall) is suddenly predicted to move, then there is a high probability that there will be interference in the motion calculation.
[0060] The solution according to the invention consists in calculating two ego-motion estimates for the two redundant inertial measurement units 4 and 5 as assumptions in the corresponding Kalman filters 13, 14 or VMO. At the same time, the sensor values of the inertial measurement units 4, 5 are transmitted to the monitoring unit 12. If both inertial measurement units 4 and 5 are functioning properly, they provide similar sensor signals and their difference is below the threshold of the monitoring unit 12, whereby the ego-motion states calculated are also identical and consistent with the objects in the vehicle environment observed by the object tracking unit 17. If a fault F occurs in one of the IMU sensors 4, 5, this event is detected by the monitoring unit 12. In other words, the assumptions about the ego-motion state of the vehicle are different at this time and the wrong solution leads to inconsistent predicted and measured object positions. On the other hand, the normal motion state calculated by the normal IMU is consistent with the observed object. Combining these two methods, it is possible to clearly locate the faulty IMU 4, 5 and the sensor with the fault F and switch the system to the backup solution.
[0061] In one embodiment, the movement states BZ A, BZ B calculated by the two redundant inertial measurement units 4 , 5 can be supplied to an object tracking unit 17 for further processing.
[0062] In one embodiment, the error state of the IMU monitoring unit 12 determined by comparing the corresponding IMU sensors of the two inertial measurement units 4 and 5 can be input into the object tracking unit 17, so that only in the event of a fault F do the two hypotheses of the motion state need to be checked.
[0063] In one embodiment, the self-motion of the vehicle 22 is determined using the vehicle pose
[0064]
[0065] To describe, is the vehicle 22 relative to x n and n is the position increment (angle increment) of the local earth-fixed (tangential) coordinate system 20 of the axis, is the position increment of the vehicle 22 in the ground-fixed coordinate system 20. represents the rotation matrix that can be calculated from the position increment. Then the overall transformation (rotation and displacement) of the object from ground-fixed coordinates to body coordinates can be expressed as the homogeneous matrix (homogeneous transformation from ground-fixed coordinates to body coordinates) well known in practice and in the literature [8, 10]:
[0066]
[0067] to describe.
[0068] In one embodiment, the predicted value of the object position in the vehicle environment can be calculated based on the two states of vehicle self-motion and previous object position. represents the position of an object in ground-fixed coordinates, then its position in vehicle body coordinates It can be derived from the following formula:
[0069]
[0070] here, Refers to the position of an object that can be measured relative to the vehicle 22 by onboard environmental sensors (camera, lidar, radar).
[0071] In one embodiment, the position of the object in Earth-fixed coordinates
[0072]
[0073] represents the object state for object tracking, and
[0074]
[0075] is the measurement variable provided by the environmental sensor (the object position measurement obtained by the environmental sensor), which must be fused in the Kalman filter 13, 14 in order to locate the object. For simplicity, it is assumed here that the object has been classified as static. In this case, x must be a constant. The fusion task in Kalman filters 13 and 14 is to use the measured value z Estimated object status (the estimated position of the object) is corrected (updated). If now z k represents a series of measurements of an object at discrete time points k = 0, 1, 2, ..., u k represents the relevant body posture, represents the object position estimate before correction (a priori), and then the corrected (a posteriori) position estimate Corrected by the following formula:
[0076]
[0077] The difference between the measured value and the predicted measured variable determined based on the vehicle's self-motion and a priori position estimate is called the change v k , References [2,3,1]:
[0078]
[0079] According to Kalman filter theory, the statistical distribution of the changes can be calculated through its covariance matrix:
[0080]
[0081] in is the covariance matrix of the prior state estimation error, R k is the covariance matrix of the measurement errors. The latter contains basic statistical assumptions about the accuracy of the measurement errors and the ego-motion estimates (error distribution). Then, for the change vector v k Each component v ik Get an estimate
[0082]
[0083] This means that the difference between the measured and predicted values should be less than a multiple of their standard deviation, which is measured by the diagonal elements of the covariance s. iik(If κ=3, then the statistical mean of the changes should be 99.73% within the range described by equation (9). If κ=4, then 99.994%.) If this is not the case, then either the measurement is wrong (outliers) or the prediction is wrong (e.g. due to an error in the calculation of the self-motion). Now and As two self-motion assumptions based on inertial measurement units A and B, 4 and 5, as the a priori estimated positions of objects in the vehicle environment classified as static. If IMUA, the fault F in 4 and the resulting The error in the result causes the corresponding changed element to
[0084]
[0085] This corresponds to the assumption In terms of the elements of change,
[0086]
[0087] It can then be clearly detected that the assumption A is wrong and that there is an error in the ego-motion A. In this way, it is possible to clearly find out to which of the two IMUs 4 , 5 the error previously identified by the IMU monitor of the monitoring unit 12 belongs.
[0088] Now, as shown in Figure 4, a simple two-dimensional space example is used to illustrate this process. Figure 4a shows the relationship between the ground-fixed coordinate system 20 of the vehicle 22 and the vehicle body coordinate system 21. The vehicle coordinate system 21 has an axis x in the two-dimensional space. b and b . Where the yaw angle Ψ represents the rotation between the two coordinate systems. The corresponding rotation matrix is
[0089]
[0090] In the example shown in FIG. 4 b , the vehicle 22 is traveling at a constant speed v x , heading straight towards an object position 25 of an object located directly in front of the vehicle 22 , which is described by the following vector:
[0091]
[0092] At the time point t corresponding to the discrete time point k = 0 0 , the vehicle 22 has a yaw angle ΔΨ 0 =0, position increment
[0093]
[0094] The motion of the vehicle 22 is described by the following three equations:
[0095]
[0096] In this case, the rotation speed ω z Also known as yaw rate, it is measured by the inertial measurement units 4 and 5. Velocity v x is the longitudinal velocity of the vehicle 22, which can be obtained by fusing the wheel speed sensor and IMU data in VMO 13 and 14. A common simplified calculation method in this example is to calculate the lateral velocity of the vehicle 22 according to the following formula:
[0097] v y ≈l r ω z (18)
[0098] Among them, l r is the distance from the origin 21 of the vehicle body coordinate system to the rear axle 22 of the vehicle. This is called pure kinematic driving. When the vehicle 22 is moving straight, ω z = 0. It can be seen that ΔΨ = 0 until the time point t corresponding to the discrete time point k = 1 1 =t 0 +ΔtIntegrating the motion formula, we can get the new position 23 of the vehicle 22 u 1
[0099]
[0100] Therefore, in body coordinates, the measured position of the object is
[0101]
[0102] At this time, if the yaw angular velocity sensor of the IMU is at time point t 0 A constant bias error δω appears z , then the sensor cannot measure the true yaw angular velocity ω z = 0, but will measure the wrong yaw rate
[0103] ω z =0+δω z (twenty one)
[0104] Integrate formula (15) at time point t 1 The wrong yaw angle
[0105] ΔΨ 1 =δω z ·Δt (22)
[0106] This error will propagate to the position calculation link, resulting in an error in the calculation of the position and posture 24 of the vehicle 22. For the sake of demonstration, we assume that the yaw angle error is very small and is expressed by trigonometric functions, and its angle is approximately
[0107] sinΔΨ≈ΔΨ,cosΔΨ≈1 (23)
[0108] Using this approximation, the error position caused by the yaw rate disturbance can be analyzed and calculated by integrating equations (16) and (17).
[0109]
[0110] Then, the rotation matrix obtained from the small angle approximation is
[0111]
[0112] The matrix H( u 1 ).
[0113]
[0114] The final result of the predicted measurements based on the vehicle's self-motion is
[0115]
[0116] This result is best illustrated by a numerical example:
[0117]
[0118] l r =1.5m (29)
[0119]
[0120] x o =80m (31)
[0121] Δt=0.3s (32)
[0122] In the absence of fault F, the yaw rate is the accurate measured prediction value
[0123]
[0124] In the presence of fault F, the resulting yaw rate is
[0125]
[0126] This means that in the vehicle body coordinates, the position of the object 25 will be subject to an erroneous self-motion and is expected to shift to the right by more than 0.4 m in the y component, while in the case of fault F the measured object position 26 is still directly in front of the vehicle 22. For example, if the relevant standard deviation is The corresponding assumption of κ=3 may be discarded due to error.
[0127] In one embodiment, two hypotheses about the predicted position of the object can be compared with the object position 25, 26 measured by the environmental sensors. If the deviation of the measured object position 25, 26 from the predicted object position 25, 26 exceeds a threshold for one of the hypotheses but does not exceed the threshold for the other hypothesis, it can be concluded that the relevant inertial measurement unit 4, 5 is faulty.
[0128] In one embodiment, the monitoring results of the two IMU sensors 4, 5 can be incorporated into the error detection of the object tracking unit 17. In this case, a fault F is only identified if the deviation of the predicted object position 25, 26 for one of the hypotheses exceeds a threshold value, while for the other hypothesis the deviation does not exceed the threshold value, and at the same time an error is detected by the sensor comparison of the IMU error detection unit 16.
[0129] In one embodiment, false detection may include only one object identified as stationary or multiple objects identified as stationary, and a fault may be identified for an assumption in which the predicted object position 25, 26 changes significantly while the predicted object position 25, 26 remains constant for other assumptions.
[0130] The invention enables a simple redundant method for fault-tolerant determination of the motion variables of the vehicle 22 in the event of an IMU failure F (failure to operate). No special requirements are imposed on the geometry and sensor quality. On the contrary, the invention is also applicable to low-cost, heterogeneous IMU hardware that is already partially available in the vehicle 22.
[0131] The method according to the present invention can mask sensor errors for the signals of main inertial measurement unit 4 as a measure for ensuring the functional safety (fail-operational) of the vehicle self-motion estimation in view of the high safety requirements of autonomous driving.
[0132] The key point here is that heterogeneous (different types) inertial measurement units (IMUs) 4 , 5 installed at different locations on the vehicle 22 are used.
[0133] The solution according to the invention enables a fail operational in the event of a fault in the calculation of the vehicle motion variables. Here too, use is made of systems and methods for object detection and fusion that are already used in driver assistance systems and autonomous vehicles.
[0134] References
[0135] [1]Berman,Z.:Outliers Rejection in Kalman Filtering–Some NewObservations.:IEEE / ION Position, Location and Navigation Symposium-PLANS2014,S.1008–1013 9
[0136] [2]Grewal, MS;Adrews, AP: Kalman Filtering. New York: John Wiley, 20019
[0137] [3] Groves, PD: Principles of GNSS, Inertial and Multisensor integrated Navigation systems. Artech House, 2013 9
[0138] [4]Kalkkuhl,J.;Bergmann,M.;Engelhardt,T.;Bleimund,F.:DE 102021 004103A1:Verfahren und Anordnung zur und Detektion vonSensorfehlern in Inertial-Mess-Systemen 3,17
[0139] [5]Klier, W.; Reim, A.; Stapel, D.: Robust Estimation of Vehicle SideslipAngle–An Approach w / o Vehicle and Tire Models. From: SAE Technical Paper Series(2008),Nr.2008-01-0582 1
[0140] [6]Mercep, L.;Pollach, M.: US10553044B2: Self-diagnosis of faults with a secondary system in an autonomous driving system. 4
[0141] [7]Mercep, L.;Pollach, M.: US11145146B2: Self-Diagnosis of Faults in an Autonomous Driving System 4
[0142] [8]Rabe, C.: Detection of Moving Objects by Spatio-Temporal Motion Analysis, Christian-Albrechts- Kiel, Diss., 2011 7
[0143] [9]Wendel, J.: Integrierte Navigationssysteme. Oldenbourg, 2007
[0144]
[10] Woo, A.;Fidan, B.;Melek, W.W.: Localization for Autonomous Driving. From: Zekavat, S.A. (publisher); Buehrer, R.M. (publisher): Handbook of Position Localization. IEEE Press Wiley, Chapter 29 7。
Claims
1. A method for identifying a fault (F) in an inertial measurement unit (4, 5) for measuring angular velocity in a vehicle (22) Hebi in, using two inertial measurement units (4, 5), each having a plurality of sensors including an accelerometer and a gyroscope, The invention is characterized in that a first inertial measurement unit (4) is used as a main inertial measurement unit, and a second inertial measurement unit (5) whose performance can be lower than that of the first inertial measurement unit (4) is used as a secondary inertial measurement unit; the measurement value of the main inertial measurement unit (4) is used as a reference value to compensate the measurement value of the secondary inertial measurement unit (5) by estimating the error model parameters relative to the main inertial measurement unit (4), so as to identify a fault (F) by comparing the two corresponding sensor signals of the two inertial measurement units (4, 5); and one of the two inertial measurement units (4, 5) is detected as having a fault: the motion estimate of the one inertial measurement unit calculated in the corresponding downstream Kalman filter (13, 14) and in the object tracking unit (17) leads to an erroneous prediction of the calculated position of an object in the vehicle (22) environment.
2. The method according to claim 1, It is characterized in that The sensor values of the inertial measurement units (4, 5) are input to a monitoring unit (12), from which error states are transmitted to an object tracking unit (17), so that in the event of a fault (F) multiple hypotheses for the movement states (BZ A, BZ B) can be checked, whereas in normal operation only one movement state (BZ A, BZ B) is used.
3. The method according to claim 1 or 2, It is characterized in that The ego-motion of the vehicle (22) is described by the vehicle pose (u, u') and transmitted to the object tracking unit (17).
4. The method according to claim 3, It is characterized in that The object tracking unit (17) predicts a new object position (25, 26) of a single object in the vehicle body coordinate system or new object positions (25, 26) of multiple objects in the vehicle body coordinate system based on the calculated vehicle pose (u, u') hypothesis and the previously calculated object position (25, 26) of a single object or the previously calculated object positions (25, 26) of multiple objects.
5. The method according to any one of claims 1 to 4, It is characterized in that The object positions (25, 26) of one or more objects measured by an environmental sensor system of a vehicle (22) are compared with the object positions (25, 26) predicted by a self-motion hypothesis of the vehicle (22), the self-motion hypothesis of the vehicle being determined by means of a Kalman filter (13, 14), wherein a self-motion hypothesis is detected as erroneous in which a respective deviation between the predicted object position and the measured object position (25, 26) exceeds a threshold value.
6. The method according to any one of claims 1 to 5, It is characterized in that In the event of a fault (F) in one of the inertial measurement units (4, 5), a switch is made to the other inertial measurement unit (4, 5) if the fault (F) is identified by comparison with a redundant sensor of the other inertial measurement unit (4, 5) and a self-motion hypothesis calculated using erroneous sensor values is rejected as an error.
7. The method according to any one of claims 1 to 6, It is characterized in that The monitoring results of the two inertial measurement units (4, 5) are incorporated into the error detection of the object tracking unit (17), in which case a fault (F) is only detected when the deviation of the object position (25, 26) predicted by one of the hypotheses exceeds a threshold value, while the deviation of the object position predicted by the other hypotheses does not exceed the threshold value and an error is detected by the sensor comparison of the IMU error detection unit (16).
8. The method according to any one of claims 1 to 7, It is characterized in that In the error detection, only one object or multiple objects identified as stationary are taken into account, and an error is detected for a hypothesis in which the predicted object position (25, 26) changes significantly while for other hypotheses the predicted object position remains constant.
Citation Information
Cited By
IMU system fault identification method and device based on MEMS, and terminal
CN119573776A
MEMS-based IMU System Fault Identification Method, Device and Terminal
CN119573776B
