Method for calibrating an inertial measurement unit
The method calibrates IMUs during vehicle operation using secondary sensors and cloud data to simplify and enhance the calibration process, reducing costs and improving accuracy and safety.
Patent Information
- Application Number
- PCT/EP2026/063421
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-09
- Filing Date
- 2026-05-08
- Publication Date
- 2026-11-12
AI Technical Summary
Existing IMU calibration methods are complex and resource-intensive, requiring dedicated factory-based calibration procedures that are time-consuming and costly, and do not account for manufacturing dependencies and environmental factors.
A method for calibrating IMUs during vehicle operation using secondary measurement data from independent sensors, combined with cloud-based data management to generate and apply calibration parameters, allowing for continuous optimization and compensation for manufacturing and aging effects.
Simplifies the calibration process, reduces costs, and enhances accuracy by leveraging existing vehicle sensors and cloud data to provide real-time calibration adjustments, improving safety and performance over the IMU's lifetime.
Smart Images

Figure EP2026063421_12112026_PF_FP_ABST
Abstract
Description
[0001] R.418346
[0002] - 1 -
[0003] Description
[0004] title
[0005] Method for calibrating an inertial measuring unit
[0006] The invention relates to a method for calibrating an inertial measuring unit in a vehicle, in particular in a motor vehicle, and an arrangement for carrying out the method.
[0007] State of the art
[0008] An inertial measurement unit (IMU) is a spatial combination of several inertial sensors, such as...
[0009] To understand acceleration sensors and gyroscopes. An IMU is typically used in an inertial navigation system as a sensory measurement unit. IMUs are used in particular to detect the six possible kinematic degrees of freedom.
[0010] In metrology, calibration refers to a measurement process in which the accuracy of a measuring instrument is verified. The measured value of the instrument is compared to a known, precise reference value. In the procedure presented here, calibration serves to determine calibration parameters for an IMU, which in turn are used to adjust the IMU.
[0011] In known methods, the IMU is calibrated during the production process in dedicated test chambers that perform defined movements at defined temperatures. Calibration parameters are determined by comparing the calibrated reference movement with the movement measured by the IMU itself. These parameters are used to correct the measured values during the IMU's operational phase. R.418346
[0012] - 2 -
[0013] The more precisely the measurement accuracies of the IMU are specified in this recognized and common procedure, the more complex these calibration procedures are.
[0014] Disclosure of the invention
[0015] Against this background, a method according to claim 1 and an arrangement according to claim 9 are presented. Embodiments are derived from the dependent claims and from the description.
[0016] A method for calibrating an inertial measurement unit (IMU) in a vehicle is presented, which is performed during vehicle operation, typically during normal vehicle use. During vehicle operation, the IMU provides initial measurement data. Secondary measurement data from at least one measuring device in the vehicle are used to determine calibration parameters for the IMU. This measuring device operates independently of the IMU.
[0017] Independent means that the at least one measuring device may acquire the same measured quantities as the inertial measuring unit, but obtains them using a separate or independent measurement procedure. Typically, the at least one measuring device is also spatially separated from the inertial measuring unit.
[0018] This presents a procedure that makes it possible to avoid calibration at the factory.
[0019] It should be noted that when used in a vehicle with SAE Level 3 functionality (SAE: Society of Automotive Engineers), various sensors are present. These measure the environment and develop an environmental model, which in turn enables the use of automated functions. R.418346
[0020] - 3 -
[0021] An example of this is the steering angle sensor and the wheel speed sensor. These detect the driver's input and determine the rolling motion of the wheels. Since the tires slip in certain situations, meaning that the rolling motion doesn't always correspond to the vehicle's movement, this information cannot be used directly. This is where the driving situation classifiers, which recognize specific driving situations, are important. They identify situations in which the initial measurements provided by the inertial measurement unit can be compared with the signals from the other sensors, which are used as the second set of measurements. It is important to have as complete a picture as possible of all situations, which then calibrates all sensor characteristics. This results in a correction value that can be used for future applications. To achieve even better alignment and, if necessary,To leverage manufacturing dependencies of wafers, a connection to cloud storage is helpful.
[0022] Calibration data is managed in the cloud storage. The algorithms verify the correction values calculated in the vehicles and compare them with near-production parameters to identify similarities. These parameters are typically acquired during the manufacturing process and often have very similar properties, for example, due to their position on a common wafer. This means that if one element of the wafer is known, typically all elements are known. Consequently, issues identified in individual sensors—especially sensor properties useful for improved calibration—can be applied to a large number of sensors on the same wafer. This can refer to a single wafer or, in specific configurations, can be applied to manufacturing parameters in general. This could, for example, mean that a manufacturing layer has identical parameters.This, particularly through such similarity analyses, allows calibration parameters to be completed, for example, by reference measurements of individual IMUs and partial measurements of the vehicles during operation. This is especially important for completing those calibration parameters that are not yet known from a partial measurement, particularly from the initial measured values. This increases safety and simplifies the process of optimizing calibration.
[0023] Calibration. R.418346
[0024] - 4 -
[0025] The presented method simplifies the manufacturing process, resulting in significant cost and resource savings. Furthermore, it eliminates the time-consuming calibration process at the factory, shifting this step to the operational phase. The method utilizes sensor data already present in the vehicle to identify specific driving situations, which can then be used to generate calibration data. A minor time constraint arises because vehicles calibrated at the factory meet full specifications immediately, from the very first kilometer. With this approach, significant tolerance issues are likely to occur initially, but these can be addressed in consultation with the vehicle manufacturer. The full implications of this are not yet entirely clear.It might be necessary to drive a specific distance, similar to how one used to move a mobile phone in a figure-eight motion to calibrate the sensors. This could mean, for example, that a vehicle manufacturer drives a specific course with the car and then adjusts the values.
[0026] In addition, the cloud database offers a way to create classes of IMUs with very similar characteristics. Specifically, calibration parameters for these classes are stored in the cloud database. These classes then allow the experience gained from partial measurements to be extrapolated to the rest of the class. The data for the entire class can be generated, for example, using reference samples in manufacturing. It is therefore also possible to age such reference samples to document changes in their properties. This database enables the use of optimized calibration data in the field at any time, particularly providing calibration parameters for the inertial measurement unit. Furthermore, it should be possible to detect significant deviations of individual sensors in the field in order to take targeted measures if necessary. Aging effects can also be monitored, and field behavior can be observed.
[0027] Preferably, data from a cloud is used, enabling even more precise calibration. This data can, for example, provide reference calibration parameters for entire classes of inertial measurement units (R.418346).
[0028] - 5 -
[0029] This includes categories formed based on manufacturing characteristics. The process can then determine the class of the vehicle's own measuring unit and access detailed class data from the cloud. Furthermore, the cloud can provide a reference experience space that can also include data on artificially aged reference samples. By comparing the properties measured during operation with this reference experience space, aging effects can be compensated for and potential outliers can be reliably identified.
[0030] The presented arrangement is designed to carry out the procedure described herein and includes an evaluation unit for this purpose. The arrangement can be implemented in hardware and / or software and integrated into a vehicle control unit or implemented as such a control unit.
[0031] Further advantages and embodiments of the invention will become apparent from the description and the accompanying drawings.
[0032] It is understood that the features mentioned above and those to be explained below can be used not only in the combinations specified, but also in other combinations or on their own, without leaving the scope of the present invention.
[0033] Brief description of the drawings
[0034] Figure 1 shows a schematic representation of an environment for carrying out an embodiment of the presented method.
[0035] Figure 2 shows a schematic representation of a vehicle to illustrate the presented method.
[0036] Figure 3 shows a schematic representation of the agreement of current IMU classification results. R.418346
[0037] - 6 -
[0038] Figure 4 shows a possible sequence of the described procedure in a flowchart.
[0039] Figure 5 shows, in a highly simplified, purely schematic representation, a vehicle with an arrangement for carrying out an embodiment of the described method.
[0040] Embodiments of the invention
[0041] The invention is schematically illustrated with reference to embodiments in the drawings and is described in detail below with reference to the drawings.
[0042] Figure 1 shows an environment in which an embodiment of the described method is carried out. The illustration shows a vehicle 10 whose movements 12, in particular rotational movements about three axes 14, are to be measured or determined here with an inertial measuring unit 16 and can, in particular, serve as initial measured values. Furthermore, a cloud 20 is provided, which serves as a backend and database for calibration data. Regular data analysis and pattern recognition 22 can be performed with this, for which typical correction values 24 are exchanged. Sensor information 26, such as...
[0043] Wheel speed, steering angle, camera information, and radar information are used as the basis for the second set of measurements. A calibration update (28) is also performed, whereby the calibration parameters can be determined, in particular, from the second set of measurement data and correction data from the cloud. Further components include a classifier (30) for driving situations and an IMU error calculation (32). Typical reference data values (34) are also transmitted.
[0044] Reference data values are values determined from the measured sensor information (the second measurements (362) or their precursors) and are related to the first measurements of the IMU. This means that these values can be used directly for the so-called calibration process. The sensor data, especially the second measurements, are typically R.418346
[0045] - 7 -
[0046] Other information, such as the rolling motion of the wheels, must first be converted into a quantity understandable to the IMU, for example, a velocity. Then, by integrating the acceleration, a velocity can be determined, which must then match the wheel speed.
[0047] In addition to the properties of the individual sensor modules, their arrangement within the vehicle must also be corrected. Figure 2 illustrates this in the vehicle coordinate system. The technical term for this is misalignment, and it is described as a rotation matrix.
[0048] Figure 2 shows a vehicle, which is identified by the reference numeral 100. This vehicle is equipped with a sensor device 102 comprising sensor modules 104. A first coordinate system 110 and a second coordinate system 112 are also shown. An arrow 114 indicates the size RGNR.
[0049] The sensors are designed to measure motion parameters such as acceleration and yaw rate during vehicle movement. The sensor is typically mounted on a circuit board integrated into a control unit located within the vehicle. Installation tolerances exist for each mounting location, resulting in the measuring sensor axis having a rotational orientation relative to the vehicle's axis. This static rotation, which is vehicle-specific, needs to be determined.
[0050] The following matrix illustrates the rate part of the IMU model for ACC and ACCredundant-
[0051] (W x \ fSw x (T) 0 0 \ (Mw xx Mw xy Mw xz \ / W xstim \ / OW X (T) / / Vw x \ I ™ y ] = 0 0 j I Mw yx Mw yy Mw yz I w ystlm l + OWyCT) + NWy
[0052]
[0053] \ W z / \ 0 0 Sw z (T)J \Mw zx Mw zyMw zz y \W zs tim / \0w z (T) / \NW Z J Output Sensitivity Orthogonal Stimulus Offset Noise Signal matrix Liability matrix
[0054] The stimulus is the vehicle acceleration w in the three spatial directions. This is transformed into the sensor output signal by the orthogonality matrix and the sensitivity matrix. Additionally, offset and R.418346 are used.
[0055] - 8 -
[0056] Add noise. The RGNR value is the product of the matrix multiplication (orthogonality matrix * sensitivity matrix).
[0057] This rotation is only measurable while the vehicle is in use. A simple rotation around the z-axis therefore causes a component in both the x- and y-sensing directions when driving perfectly straight. The angle can then be determined using the tangent function (tan).
[0058] It is taken into account that sensors from the same wafer have very similar properties. This can also apply to neighboring wafers.
[0059] If such similarities or classes exist, then the expected properties from the IMU classifier measurement should lie within this empirical space. This allows conclusions to be drawn about the entire parameter space based on a partial measurement, and the calibration parameters obtained through the reference measurement can be directly applied to the individual IMU.
[0060] Figure 3 shows a reference measurement experience space 200 containing an IMU classifier spectrum 302.
[0061] Figure 3 illustrates the agreement between the current IMU classifier results and the reference results stored in the cloud. If this is the case, then the experience gained can be applied to the entire class in the IMU module.
[0062] It should be noted that aging plays a role when using classes. Aging processes can be triggered using reference patterns, and the resulting new properties can then be measured. This adds a new dimension to the reference measurement experience space. This can also be applied to real-world devices in the field. This allows the devices to maintain optimal performance while simultaneously providing insights into aging and potential outliers. R.418346
[0063] - 9 -
[0064] Online monitoring allows these specific outliers to be exchanged in workshops and used for further analysis.
[0065] In a particularly advantageous embodiment, data 20 are stored in the cloud, enabling the classification of inertial measurement units 16. Manufacturing-related parameters, such as wafer identification, the position of the measurement unit on a wafer, or a production batch, can be used as classification criteria. It is assumed that measurement units of the same class, for example, from the same wafer, exhibit very similar physical behavior, particularly with regard to temperature dependencies and zero-point drift. Reference calibration parameters or characteristic property profiles are stored in the cloud for such classes. These can be obtained, for example, by precisely measuring a few reference samples of a class during production.The procedure is designed to determine or verify the vehicle's IMU's classification within one of these classes by ascertaining the IMU's characteristics in suitable driving situations, particularly through an initial measurement or a partial measurement in a specific driving situation. If the classification is positive, the detailed reference calibration parameters for the entire class can be transferred from the cloud to the vehicle's IMU and used for adjustment. This enables highly accurate calibration, even if only a limited amount of data is available for calibration within the vehicle itself.
[0066] Furthermore, the data provided by Cloud 20 can include a reference measurement experience space 200. This experience space is generated from measurements of reference patterns under defined conditions and represents the permissible parameter space for a functioning IMU of a specific class. The current property spectrum 302 of the IMU, determined in the vehicle using the classifier 30 from the initial measurement values 354, is compared with this reference experience space 200 in the cloud. To monitor and compensate for aging effects of the IMU in the field, the physical reference patterns, whose data are stored in the cloud, can be subjected to targeted, artificial aging processes. [The following appears to be unrelated and possibly a separate text fragment: R.418346
[0067] - 10 -
[0068] Measured property changes are also documented in the cloud, dynamically expanding or modifying the reference experience space. By comparing field-measured data with this aging data, the method can estimate the aging state of the IMU in the vehicle, adjust the calibration parameters accordingly, and ensure optimal performance over its lifetime.
[0069] Furthermore, the use of cloud data enables reliable detection of outliers. If the characteristic spectrum 302 determined in the vehicle of an IMU deviates significantly from the reference experience space 200 stored in the cloud for the corresponding class, the IMU is classified as mature or potentially defective. In this case, the process can generate a warning, send a message to a manufacturer's backend server, or prompt the driver to visit a workshop. This increases the safety of the overall system, as malfunctions are detected early, before they have a critical impact.
[0070] Figure 4 shows a flowchart illustrating a possible sequence of the presented procedure. In a first step 300, the vehicle begins operation. In a second step 302, initial measured values are obtained from an inertial measuring unit. In a second step 304, which can occur simultaneously or at a later time, second measured values are obtained from at least one independent measuring device, e.g., a sensor. In a third step 306, an evaluation is then performed, and calibration parameters 308 are obtained.
[0071] Figure 5 shows a schematic, highly simplified representation of a vehicle 350 with an inertial measuring unit 352, which provides first measured values 354, and an arrangement 356 for carrying out the described procedure. An evaluation unit 358 is provided in this arrangement 354. A measuring device 360, which provides second measured values 362, is also shown. The arrangement 356 compares the first and second measured values 354, 362 with each other and determines calibration parameters 364 for the inertial measuring unit 352.
Claims
R.418346 - 12 - Claims 1. Method for calibrating an inertial measuring unit (16, 352) in a vehicle (10, 100, 350), wherein during the operation of the vehicle (10, 100, 350) the inertial measuring unit (16, 352) provides first measured values (354), and second measured values (362) from at least one measuring device (360) in the vehicle (10, 100, 350) are used to determine calibration parameters (364) for the inertial measuring unit (16, 352), wherein which measures at least one measuring device (360) independently of the inertial measuring unit (16, 352).
2. Method according to claim 1, wherein an evaluation of the second measured values (362) of the at least one measuring device (360) is carried out by comparing these second measured values (362) with the first measured values (354) of the inertial measuring unit (16, 352).
3. Method according to claim 1, wherein the calibration parameters (364) are used to adjust the inertial measuring unit (16, 352).
4. Method according to one of claims 1 to 3, wherein measured values from a steering angle sensor are used as second measured values (362).
5. Method according to one of claims 1 to 4, wherein measured values from a wheel speed sensor are used as second measured values (362).
6. Method according to any one of claims 1 to 4, wherein a classifier is used to classify (30) different driving situations, in order to R.418346 - 13 - to determine suitable driving situations for determining the calibration parameters (364).
7. Method according to any one of claims 1 to 6, wherein data from a cloud (20) is used.
8. The method of claim 7, wherein the data used from the cloud (20) comprise reference calibration parameters for a class of inertial measurement units, the method further comprising the steps of: a. Determining the class of the inertial measurement unit (16, 352) installed in the vehicle based on its properties determined during operation or its manufacturing parameters; and b. Transferring the reference calibration parameters of the determined class from the cloud (20) for use in determining the calibration parameters (364) 9. Method according to claim 7 or 8, wherein the data used from the cloud (20) define a reference experience space (200) derived from measurements of reference patterns, and wherein the method comprises a comparison of properties (302) of the inertial measuring unit (16, 352) determined in operation with the reference experience space (200).
10. Method according to claim 9, wherein the data of the reference experience space (200) include data of artificially aged reference patterns and the comparison is used to determine or compensate for aging effects of the inertial measuring unit (16, 352).
11. Method according to claim 9 or 10, wherein, in the event of a significant deviation of the determined properties (302) from the reference experience space (200), the inertial measuring unit (16, 352) is identified as an outlier and a corresponding measure is initiated, in particular a warning is generated, a message is sent to a workshop or manufacturer server and / or an entry is made in a fault memory of the vehicle. R.418346 - 14 - 12. Method according to any one of claims 1 to 11, wherein the calibration parameters (364) are determined continuously and / or at time intervals.
13. Arrangement for calibrating an inertial measuring unit (16, 352) which has an evaluation unit (358) and is set up to carry out a method according to one of claims 1 to 12.
14. Arrangement according to claim 13, which is configured to exchange data with a cloud (20), wherein the evaluation unit (358) is configured to receive reference calibration parameters for a class of inertial measurement units from the cloud (20) and to determine the class membership of the inertial measurement unit (16, 352) in order to use the received reference calibration parameters to determine the calibration parameters (364).
15. Arrangement according to claim 14, wherein the evaluation unit (358) and / or the cloud (20) is configured to compare properties (302) of the inertial measuring unit (16, 352) determined during operation with a reference experience space (200) stored in the cloud (20) in order to detect aging effects or outliers.
