METHOD AND SYSTEM FOR INITIALIZING A SENSOR FUSION SYSTEM

Description

[0001] The invention relates to a method for initializing a sensor fusion system according to claim 1 and a system for initializing a sensor fusion system according to claim 5.

[0002] All measurement data is inherently subject to error, and in many cases, consistent availability of the measurement data is not guaranteed. In addition to the dependence of the measurement data on sensor-inherent properties, the measurement data is also often dependent on environmental conditions. Sensor errors or measurement errors can be divided into quasi-stationary components that are constant across multiple measurements, such as a so-called offset, and statistical components that are random from measurement to measurement, such as noise. While the random components cannot, in principle, be deterministically corrected, quasi-stationary errors can generally be corrected given a certain level of observability. Significant errors that cannot be corrected can usually at least be avoided given a certain level of detectability. Furthermore, errors in the measurement data can often only be identified after a series of comparative measurements have been recorded and evaluated.

[0003] In this context, sensor fusion methods are already known in the state of the art, which are usually also suitable for correcting or filtering measurement data from different sensors or sensor systems. In the automotive sector in particular, special requirements must be taken into account, as a large number of different sensors record a common environmental situation or vehicle condition using different measurement principles and describe this environmental situation or vehicle condition using a large number of different measurement data. For sensor fusion applicable in the automotive sector, the greatest possible robustness against random disturbances as well as the detection and compensation of systematic errors are required. Likewise, temporal influences on the measurement data must be corrected and temporary failures or the unavailability of sensors must be bridged.

[0004] DE 10 2012 216 211 A1 describes a method for selecting a satellite, wherein the satellite is a satellite of a global navigation system. Before such a satellite is used to determine the position of a vehicle, the received GNSS signals are checked for plausibility in various ways. Different redundancies or known relationships are exploited for this verification. For example, DE 10 2012 216 211 A1 discloses determining both the distance of the vehicle from the satellite and the relative speed of the vehicle to the satellite from the satellite. The distance can be determined using the signal's propagation time, while the relative speed can be determined using a phase measurement of the signal. Since the distance and the relative speed are interdependent, they can be verified against each other.Furthermore, the values ​​determined from the signal can be verified against known boundary conditions, since a vehicle typically moves within a specific speed range. It is also described that when multiple signals are received from different satellites, the distances to multiple satellites are determined, and these distances are simultaneously verified against each other using trigonometric relationships and the known distance between the satellites. Finally, the distance determined from the signal or the speed determined from the signal can also be verified using other sensors that also allow position or speed determination. If the signals from a satellite cannot be verified, that satellite is not used for position or speed determination.

[0005] DE 10 2010 063 984 A1 discloses a sensor system comprising several sensor elements. The sensor elements are designed to detect at least partially different primary measured variables and to use at least partially different measuring principles. Further measured variables are then derived, at least in part, from the primary measured variable of the sensor elements. Furthermore, the sensor system comprises a signal processing device, an interface device, and several functional devices. The sensor elements and all functional devices are connected to the signal processing device. The primary measured variables thus provide redundant information, which can be compared with each other in the signal processing device or can support each other. By comparing the observables calculated using different methods, conclusions can be drawn about the reliability and accuracy of the observables.The signal processing device qualifies the accuracy of the observables and makes the observables, together with an accuracy specification, available to various functional devices via an interface device.

[0006] DE 10 2012 219 478 A1 discloses a sensor system for independently evaluating the integrity of its data. The sensor system is preferably used in motor vehicles and comprises a plurality of sensor elements configured to at least partially detect different primary measurement variables or to at least partially utilize different measurement principles. The sensor system further comprises a signal processing device that at least partially evaluates the sensor signals jointly and simultaneously evaluates the information quality of the sensor signals. The signal processing device also provides information about the consistency of at least one datum of a physical quantity, wherein the datum of the physical quantity is calculated based on the sensor signals of sensor elements that either directly detect the physical quantity or from whose sensor signals the physical quantity can be calculated.The information about the consistency of the date is now calculated on the basis of directly or indirectly redundant sensor information.

[0007] WO 94 / 28435 A1 relates to a navigation system for terrestrial vehicles with raw measurement data and / or position and speed solutions from a Global Position System (GPS) receiver mixed with odometer and heading information from a dead reckoning instrument, comprising: input interface means for accepting a plurality of raw GPS pseudo-range (PR) and pseudo-range rate (PRR) measurement data, GPS position, and GPS speed solutions from a GPS receiver; first computing means for modeling an error effect on the raw measurement data and / or position and speed solutions due to selective availability (SA); second computing means for modeling a scale factor error effect on an accumulated heading error of a dead reckoning yaw rate sensor; third computing means for modeling a scale factor error effect of a dead reckoning odometer sensor;and Kalman filter means for mixing a plurality of outputs from the first to third computing means such that the heading error and odometer scale factor error outputs are estimated from at least one of the GPS PRR raw measurements and the GPS velocity solutions.;

[0008] EP 1 870 669 A2 relates to a method for checking an inertial measuring unit of vehicles, in particular of aircraft, in the stationary state, wherein the inertial measuring unit has at least one acceleration sensor which supplies a translational acceleration signal, and at least one yaw rate sensor which supplies a yaw rate signal.

[0009] US 2013 / 138264 A1 relates to an integrated INS / GPS navigation system for a vehicle for tracking a position of the vehicle, comprising: an inertial measurement unit (IMU) with microelectromechanical system (MEM) sensors for measuring the acceleration and angular velocity of the vehicle; an inertial navigation system (INS) for generating vehicle state estimates based on the acceleration and angular velocity measurement from the IMU; an auxiliary measurement (Aux) unit for generating auxiliary measurement data including an analytical state derived from the mechanical state of a vehicle; a global positioning system (GPS) unit for receiving GPS satellite signals from multiple GPS satellites via a GPS antenna to generate GPS measurement outputs indicating an absolute position and velocity of the vehicle; and a Kalman filter that combines the state estimates of the INS, the auxiliary measurement data of the Aux unit, and the GPS measurement outputs of the GPS unit and performs Kalman filter processing thereon.wherein the analytical condition included in the Aux unit is a relationship between a lateral directional velocity of a vehicle, a distance of the sensor position with respect to a rear wheel axle, and an angular velocity with respect to the z-axis of a vehicle. ;

[0010] However, the generic methods and sensor systems known in the prior art are disadvantageous in that they do not adequately consider problems that arise during the initialization of these sensor systems. During initialization, many physical quantities are initially unknown because no measurement data for these quantities can be acquired directly. At the same time, the potential errors in these physical quantities can be so large that the linearization conditions assumed for a generic sensor data fusion become invalid in the fusion filter, rendering the fusion filter and the error estimation it performs unstable. This leads to the failure of the sensor fusion.

[0011] It is therefore an object of the invention to propose an improved method for initializing a sensor fusion system.

[0012] This object is achieved according to the invention by the method for initializing a sensor fusion system according to claim 1.

[0013] The invention relates to a method for initializing a sensor fusion system when booting or starting the sensor fusion system, wherein the sensor fusion system comprises a base system and at least one correction system, wherein measured values ​​are each recorded by the base system and by the at least one correction system, wherein the measured values ​​directly or indirectly describe physical quantities, wherein the measured values ​​each contain error values, wherein the error values ​​describe deviations of the measured values ​​from the described physical quantities, wherein at least the physical quantities indirectly described in the measured values ​​of the base system and their error values ​​cannot be determined during the initialization of the sensor fusion system, wherein the measured values ​​are continuously fused to form a fusion data set after completion of the initialization, and wherein start values ​​are determined from the measured values ​​of the at least one correction system during the initialization for at least one of the physical quantities indirectly described in the measured values ​​of the base system and their error values.

[0014] This results in the advantage that when the sensor fusion system is booted up or started, measured values ​​for all the physical variables required by the sensor fusion system are available right from the start. Thus, the sensor fusion system is essentially fully functional right from the start.

[0015] Known sensor fusion systems typically include a so-called fusion filter, which operates as a recursive filter. This means that the fusion filter infers the present from past measured values. However, when initializing the sensor fusion system, no past measured values ​​are yet available, which is why the method according to the invention uses initial values ​​from the measured values ​​of at least one correction system to provide the fusion filter with the measured values ​​as initial values.

[0016] The invention utilizes the fact that the at least one correction system can usually record and provide measured values ​​essentially immediately upon initialization.

[0017] During operation, the sensor fusion system typically fuses the measurement data of the base system and the correction system into a common fusion data set, which is more reliable and precise than the individual measurement data and, in particular, allows an assessment of the accuracy or reliability of the fused measurement data by means of an error estimation.

[0018] The term "during initialization of the sensor fusion system" is used in the context of the invention synonymously with "during the startup of the sensor fusion system" or "during the start-up of the sensor fusion system" and refers to the time frame of the initialization process of the sensor fusion system. According to the invention, initialization is considered complete when the fusion filter receives all the information required for its correct functioning from the base system, i.e., both the physical quantities directly and indirectly described in the measured values ​​of the base system, and only requires the measured values ​​of the correction systems to determine the error values ​​of the base system.

[0019] Preferably, the method according to the invention is carried out not only during the first initialization of the sensor fusion system but also during some or all subsequent reinitializations.

[0020] Indirectly described physical quantities, and in particular their error values, are preferably determined from directly measured physical quantities using physical or mathematical relationships. An example of an indirectly described physical quantity is a velocity, which is indirectly described by a change in position over time. Position as a physical quantity is thus measured directly, and the velocity is then determined from the change in position over time using differentiation.

[0021] Preferably, only measured values ​​with an identical timestamp are used to create the joint fusion dataset. The measured values ​​thus describe the same physical quantity at the same time. Thus, the measured values ​​used are ideally identical, or ideally, must be. This results in the advantage that error values ​​in the measured values ​​can be identified more easily and reliably. This also improves the accuracy and reliability of the fusion dataset.

[0022] Preferably, it is provided that the common fusion data set comprises at least the error values ​​of the base system, wherein the error values ​​of the base system are determined by comparison with the measured values ​​of the at least one correction system and, if appropriate, by weighting the measured values ​​of the base system against the measured values ​​of the at least one correction system.

[0023] Furthermore, it is preferred that the at least one correction system determines the starting values ​​and the correction values ​​by means of independent measured value processing. For this purpose, the at least one correction system can, for example, comprise independent preprocessing of the measured values ​​that is largely independent of the sensor fusion system. In particular, within the framework of tight coupling, which is also known as tight coupling, the at least one correction system first independently determines the values ​​of those physical quantities that are described by the measured values ​​of the at least one correction system. Within the framework of this preprocessing, an initial correction of the measured values ​​of the at least one correction system is then already possible.

[0024] According to the invention, the base system is an inertial navigation system and the at least one correction system is a global satellite navigation system and / or an odometry navigation system. The present invention is therefore particularly suitable for navigation purposes and for navigation systems, preferably in motor vehicles. The sensor fusion system thus determines the position, in particular the position of a motor vehicle, from the measured values. The global satellite navigation system can be, for example, a so-called GPS navigation system. The odometry navigation system first determines the speed, e.g., using the known rolling circumference of the motor vehicle tires, and thus enables position determination taking into account the steering angle within the framework of dead reckoning. It is particularly expedient for the satellite navigation system to comprise at least two satellite signal receivers.This improves the quality of the detected satellite signals and thus the reliability and accuracy of the satellite navigation system.

[0025] According to the invention, the starting values ​​for the error values ​​are determined as variances based on the measurement inaccuracies of the at least one correction system. This first of all takes into account the fact that the sensor fusion system or the fusion filter generally also determines error values ​​and, for this purpose, relies on the availability of measured values. By using the variances of the measured values ​​of the at least one correction system as starting values ​​for the error values, this quantity can also be made available to the sensor fusion system or the fusion filter in a comparatively simple manner.

[0026] In particular, it is preferred that the measurement inaccuracies be taken from the sensor specifications or a data sheet of the sensors of the base system or the at least one correction system, or be determined by means of error propagation. The basis for the determination by means of error propagation is the self-assessment of the measurement accuracy by the base system or the at least one correction system and the respective preprocessing. Preprocessing is typically used for known sensor fusion systems anyway and can therefore also be used for the method according to the invention without additional effort.

[0027] According to the invention, the physical quantities indirectly described in the measured values ​​of the base system include a speed and / or a position and / or a receiver clock error of the global satellite navigation system and / or a receiver clock error drift of the global satellite navigation system. These quantities can be determined directly from the correction systems, particularly when a global satellite navigation system and an odometry navigation system are involved.

[0028] Furthermore, it is preferred that the starting values ​​be determined step by step from the measured values ​​of the at least one correction system. This offers the advantage that the initialization speeds and, if applicable, other properties of the different correction systems can be taken into account, thus enabling the measured values ​​to be provided as starting values ​​at a time that is individually adapted to the correction systems. The starting values ​​are thus provided in the chronological order in which they are output by the correction systems.

[0029] Furthermore, it is particularly preferred that initial values ​​are first determined from the measured values ​​of the odometry navigation system. The odometry navigation system has the advantage that it can detect wheel rotation signals immediately upon vehicle start-up and thus determine the speed immediately upon vehicle start-up. Furthermore, the wheel rotation signals are of particular importance for vehicle safety, as they are used not only for determining speed but also for various driver assistance systems, such as ABS or ESC.

[0030] According to the invention, the error values ​​are detected and corrected using an error-state-space filter, in particular using an error-state-space Kalman filter. The error-state-space filter represents a fusion filter for fusing the measured values, in particular for fusing normally distributed measured values. At the same time, the error-state-space filter preferably estimates or determines the error values ​​of at least the base system. The base system is preferably an inertial navigation system of a motor vehicle. Using the correction systems, the error values ​​and, if applicable, also unknown variables of the inertial navigation system can then be estimated or determined. A special feature of the error-state-space filter is that, instead of the sensor signals or measured values, only error values ​​are incrementally estimated or determined and subsequently corrected.The error values ​​have a significantly lower temporal dynamics than the measured values ​​themselves, which results in a large-scale decoupling of the dynamics of the error state space filter from the properties of the base system or of the at least one correction system, which results in a large-scale decoupling of the dynamics of the error state space filter from the sensor properties.

[0031] Another special feature of the Error-State-Space Kalman filter is that by applying a correction, the estimated or determined error values ​​are zero after each duty cycle of the Error-State-Space Kalman filter, which eliminates an otherwise usual prediction step for predicting the error values ​​in the following duty cycle, thus reducing the computational effort for the Error-State-Space Kalman filter.

[0032] The invention further relates to a system for initializing a sensor fusion system when booting up or starting the sensor fusion system, wherein the sensor fusion system comprises a base system and at least one correction system, wherein the base system and the at least one correction system are each designed to record measured values, wherein the measured values ​​directly or indirectly describe physical quantities, wherein the measured values ​​are each subject to error values, wherein the error values ​​describe deviations of the measured values ​​from the described physical quantities, wherein the sensor fusion system is designed such that at least the physical quantities indirectly described in the measured values ​​of the base system as well as their error values ​​cannot be determined during the initialization of the sensor fusion system, wherein the sensor fusion system is designed toto continuously fuse the measured values ​​into a fusion data set after completion of the initialization, and wherein the sensor fusion system is designed to determine starting values ​​for at least one of the physical quantities indirectly described in the measured values ​​of the base system, as well as their error values, from the measured values ​​of the at least one correction system during the initialization. The system according to the invention thus comprises all devices necessary for carrying out the method according to the invention.

[0033] It is preferably provided that the system is designed to also carry out preferred embodiments of the method according to the invention.

[0034] Further preferred embodiments emerge from the subclaims and the following description of an embodiment with reference to figures.

[0035] It shows Fig. 1 shows an example of a possible embodiment of a system according to the invention, which is designed for position determination, in a motor vehicle and Fig. 2 shows an example of a further possible embodiment of a system according to the invention, which is also designed for position determination, in a motor vehicle.

[0036] Fig. 1 shows a schematic representation of an embodiment of the system according to the invention, which is intended for installation and use in a motor vehicle (not shown). The sensor fusion system shown is designed, for example, for determining the position of the motor vehicle. All elements or components comprised by the system, i.e., the basic system or the correction systems, are illustrated as functional blocks, and their interaction with one another is shown.

[0037] The sensor fusion system comprises an inertial navigation system 101, which is configured to detect at least the accelerations along a first, a second, and a third axis, as well as at least the angular rates around the first, the second, and the third axis. The first axis corresponds, for example, to the longitudinal axis of the motor vehicle, the second axis corresponds to the transverse axis of the motor vehicle, and the third axis corresponds to the vertical axis of the motor vehicle. These three axes form a Cartesian coordinate system, the so-called motor vehicle coordinate system.

[0038] Inertial navigation system 101, for example, forms the so-called base system, whose measured values ​​are corrected using the so-called correction systems described below. The correction systems are odometry navigation system 103 and satellite navigation system 104.

[0039] The sensor fusion system further includes a strapdown algorithm unit 102, which executes a strapdown algorithm to convert the measured values ​​from the inertial sensor navigation system 101 into position data, among other things. For this purpose, the measured values ​​from the inertial sensor navigation system 101, which naturally describe accelerations, are integrated twice over time. The orientation of the motor vehicle is also determined by integrating the corresponding measured values ​​from the inertial sensor navigation system 101 twice over time. The orientation and speed of the motor vehicle are also determined by a single integration over time. Furthermore, the strapdown algorithm unit 102 compensates for a Coriolis force acting on the inertial sensor navigation system 101.

[0040] The output data from strapdown algorithm unit 102 therefore includes the following physical quantities: the speed, acceleration, and yaw rate of the motor vehicle, for example, relative to the aforementioned three axes of the motor vehicle coordinate system and, for example, additionally relative to a world coordinate system suitable for describing the orientation or dynamic quantities of the motor vehicle in the world. For example, the aforementioned world coordinate system is a GPS coordinate system. Furthermore, the output data from strapdown algorithm unit 102 includes the position relative to the motor vehicle coordinate system and the orientation relative to the world coordinate system. In addition, the output data from strapdown algorithm unit 102 includes the variances as information about the data quality of the aforementioned physical quantities.These variances are not calculated in the strapdown algorithm unit 102, for example, but are simply used and forwarded by it. The aforementioned physical quantities calculated by the strapdown algorithm unit 102 are output via output module 112 and made available to other motor vehicle systems.

[0041] The sensor fusion system also includes an odometry navigation system 103 in the form of wheel speed sensors for each wheel of the motor vehicle. For example, this is a four-wheeled motor vehicle with four wheel speed sensors, each of which detects the speed of the wheel assigned to it and its direction of rotation. Furthermore, the odometry navigation system 103 includes a steering angle sensor element that detects the steering angle of the motor vehicle. In addition, the odometry navigation system 103 provides, for example, a starting speed or starting speed information via a fusion filter 105, at least upon starting or switching on or initializing the sensor fusion system.

[0042] Furthermore, the sensor fusion system illustrated by way of example comprises satellite navigation system 104, which is configured to determine the distance between an associated satellite and the motor vehicle, as well as the speed between the associated satellite and the motor vehicle. In addition, satellite navigation system 104 provides a starting position or starting position information, for example, to fusion filter 105, at least upon starting or switching on or initializing the sensor fusion system.

[0043] The sensor fusion system also includes fusion filter 105. Fusion filter 105 provides a fusion data set 106 during the joint evaluation of the measurement data from odometry navigation system 103, satellite navigation system 104, and inertial sensor navigation system 101. Fusion data set 106 contains the acquired measurement data from the various sensor systems, with fusion data set 106 additionally including, for example, error values ​​and variances associated with the error values, which describe the data quality.

[0044] The measurement data or values ​​from the inertial sensor navigation system 101 are stored for a predetermined period of time during the operation of the motor vehicle in a dedicated electronic data storage device 113 of the fusion filter 105. The inertial navigation system 101 represents the so-called base system, while the odometry navigation system 103 and the satellite navigation system 104 represent the so-called correction systems, whose measured values ​​are used to correct the measured values ​​of the base system. This ensures that values ​​that were at least apparently recorded at an identical time can always be compared.

[0045] Fusion data set 106 provided by fusion filter 105 includes, for example, the quantitative errors of the base system determined by means of the plausibility-checked measured values ​​of the correction systems.

[0046] Strapdown algorithm unit 102 now corrects the measured values ​​of the base system using fusion data set 106.

[0047] Fusion data set 106 is calculated by fusion filter 105 from the measured values ​​of odometry navigation system 103, satellite navigation system 104 and inertial navigation system 101.

[0048] Fusion filter 105 is designed, for example, as an Error-State-Space-Kalman filter, i.e. as a Kalman filter, which in particular carries out a linearization of the measured values ​​and in which the quantitative error values ​​of the measured values ​​are calculated or estimated and which works sequentially and thereby corrects the measured values ​​available in the respective functional step of the sequence.

[0049] Fusion filter 105 is configured to always asynchronously capture the most current measurement values ​​available from inertial navigation system 101, odometry navigation system 103, and satellite navigation system 104. For example, the measurement values ​​are routed via vehicle model unit 107 and orientation model unit 109.

[0050] Motor vehicle model unit 107 is designed to calculate at least the speed along a first axis, the speed along a second axis and the rotation rate about a third axis from the measured values ​​of odometry navigation system 103 and to provide these to fusion filter 105.

[0051] The exemplary sensor fusion system also includes a tire parameter estimation unit 110, which is configured to calculate at least the radius, for example, the dynamic radius, of all wheels and additionally calculates the cornering stiffness and the slip stiffness of all wheels, and provides these to the motor vehicle model unit 107 as additional input variables. Tire parameter estimation unit 110 is further configured to use a substantially linear tire model to calculate the tire sizes.

[0052] The exemplary input variables of tire parameter estimation unit 110 are the measured values ​​describing the wheel speeds and the steering angle, at least partially the output values ​​of strapdown algorithm unit 102 and the variances determined by fusion filter 105.

[0053] The exemplary sensor fusion system also includes GPS error detection and plausibility check unit 111, which is configured to receive, for example, the measured values ​​from satellite navigation system 104 and at least some of the measured values ​​from strapdown algorithm unit 102 as input data and to incorporate them into its calculations. GPS error detection and plausibility check unit 111 checks the measured values ​​against a stochastic model adapted to satellite navigation system 104. If the measured values ​​correspond to the model within a tolerance that takes noise into account, they are checked for plausibility.

[0054] GPS error detection and plausibility check unit 111 is additionally connected to fusion filter 105 at the data level and transmits the plausibility-checked measured values ​​to fusion filter 105.

[0055] GPS error detection and plausibility unit 111 is designed, for example, to carry out a method for selecting a satellite, inter alia, by means of the following method steps: Measuring position data of the motor vehicle relative to the satellite based on the sensor signals from the satellite navigation system 104, determining redundant reference position data of the motor vehicle to the position data determined based on the sensor signals from the satellite navigation system 104, Selecting the satellite if a comparison of the position data and the reference position data satisfies a predetermined condition, wherein for the comparison of the position data and the reference position data a difference is formed between the position data and the reference position data, wherein the predetermined condition is a maximum permissible deviation of the position data from the reference position data, wherein the maximum permissible deviation is dependent on a standard deviation which is calculated based on a sum of a reference variance for the reference position data and a measurement variance for the position data and wherein the maximum permissible deviation corresponds to a multiple of the standard deviation such that a probability that the position data fall into a scatter interval dependent on the standard deviation falls below a predetermined threshold.

[0056] The exemplary sensor fusion system also includes a standstill detection unit 108, which is configured to detect a standstill of the motor vehicle and, if a standstill of the motor vehicle is detected, provides at least fusion filter 105 with information from a standstill model. The information from a standstill model describes that the rotation rates around all three axes have the value zero and the speeds along all three axes have the value zero. Standstill detection unit 108 is configured, for example, to use the measured values ​​of the wheel speed sensors of odometry navigation system 103 and the measured values ​​of inertial navigation system 101 as input data.

[0057] The sensor fusion system uses, for example, a first group of measured values ​​related to a vehicle coordinate system and, in addition, a second group of measured values ​​related to a world coordinate system. The world coordinate system is used to describe the orientation and dynamic variables of the vehicle. An orientation angle between the vehicle coordinate system and the world coordinate system is determined using the orientation model unit 109.

[0058] The orientation angle between the vehicle coordinate system and the world coordinate system determined by the orientation model unit 109 is determined on the basis of the following physical quantities: the vectorial velocity with respect to the world coordinate system, the vectorial velocity with respect to the vehicle coordinate system, the steering angle and the respective quantitative errors of the measurement data or values ​​describing the aforementioned quantities.

[0059] Alignment model unit 109 accesses all of the output data from strapdown algorithm unit 102.

[0060] Alignment model unit 109 is designed, for example, such that, in addition to the alignment angle, it also calculates information about the data quality of the alignment angle in the form of a variance and provides fusion filter 105.

[0061] Fusion filter 105 uses the orientation angle and the variance of the orientation angle in its calculations, the results of which it forwards to strapdown algorithm unit 102 via fusion data set 106.

[0062] Fusion filter 105 thus records the measurement data from inertial navigation system 101, the base system, as well as from odometry navigation system 103 and satellite navigation system 104, the correction systems.

[0063] During initialization of the exemplary sensor fusion system, all sensor systems comprised by the sensor fusion system are initialized, i.e., base system 101 and the two aforementioned correction systems 103 and 104, as well as fusion filter 105. For correct functioning, even immediately after initialization, fusion filter 105 relies on the physical quantities directly and indirectly described in the measured values ​​of base system 101, which are corrected using measured values ​​from correction systems 103 and 104, or whose error values ​​are determined using measured values ​​from correction systems 103 and 104. However, since the physical quantities indirectly described in the measured values ​​of base system 101, such as the position and speed of the vehicle, cannot yet be determined, starting values ​​are determined from the measured values ​​of correction systems 103 and 104.Odometry navigation system 103, for example, determines a vehicle's speed and provides this to fusion filter 105 as a starting value for the speed determination. Likewise, a measurement inaccuracy inherent in odometry navigation system 103 is passed to fusion filter 105 as a starting value for the speed error value. As soon as satellite navigation system 104 has made an initial determination of the position, the position determined in this way is also passed to fusion filter 105 as a starting value for the position. At the same time, a variance based on the measurement inaccuracy of the determined position is output to fusion filter 105 as an error value of the determined position. The position determined as the starting value and the measurement inaccuracy of the position are determined as part of a preprocessing of the measured values ​​by satellite navigation system 104.Likewise, the speed determined as the starting value is determined during preprocessing by the odometry navigation system 103, but not the measurement inaccuracy used as the starting value for the error value. This originates, for example, from the sensor specifications of the individual wheel speed sensors. Thus, the fusion filter 105 already has all the information required for correct functioning during initialization.

[0064] Fig. 2 shows, by way of example, another possible embodiment of a sensor fusion system, which is also designed for position determination, in a motor vehicle (not shown). The sensor fusion system comprises, for example, inertial navigation system 201, satellite navigation system 204, and odometry navigation system 203 as different sensor systems. Inertial navigation system 201, satellite navigation system 204, and odometry navigation system 203 output measured values ​​that directly or indirectly describe physical quantities, namely a position, a speed, an acceleration, an orientation, a yaw rate, or a yaw acceleration, to fusion filter 205. The measured values ​​are output via a vehicle data bus, for example via a so-called CAN bus. For example, satellite navigation system 204 outputs its measured data in raw data form.

[0065] The central element in determining the position of the motor vehicle is the inertial navigation system 201, which is a so-called MEMS-IMU (Micro-Electro-Mechanical-System-Inertial Measurement Unit) It is used in combination with the strapdown algorithm unit 207, as this is assumed to be error-free. This means that the values ​​from the inertial navigation system 201 are always assumed to correspond to their stochastic model, that they only exhibit noise influences, and are thus free from external or random errors or disturbances. The noise as well as remaining, unmodeled errors from the inertial navigation system 201, such as nonlinearity, are assumed to be mean-free, stationary, and normally distributed (so-called Gaussian white noise) across the measurement range.

[0066] Inertial navigation system 201 comprises three rotation rate sensors that are orthogonal to one another and three acceleration sensors that are orthogonal to one another.

[0067] Satellite navigation system 204 comprises a GPS receiver, which first measures the distance to the receivable GPS satellites based on the satellite signal propagation time and then determines the distance traveled by the motor vehicle based on the change in the satellite signal propagation time and the change in the number of wavelengths of the satellite signals. Odometry navigation system 203 comprises a wheel speed sensor on each wheel of the motor vehicle and a steering angle sensor. The wheel speed sensors each determine the wheel rotational speed of the wheel assigned to them, and the steering angle sensor determines the steering angle.

[0068] Inertial navigation system 201 outputs its measured values ​​to preprocessing unit 206 of inertial sensor navigation system 201. Preprocessing unit 206 then corrects the measured values ​​or the physical quantities described therein using correction values ​​that preprocessing unit 206 receives from fusion filter 205. The thus corrected measured values ​​or the physical quantities described therein are passed on to strapdown algorithm unit 207.

[0069] Strapdown algorithm unit 207 then determines the position based on the corrected measurement data or values ​​from preprocessing unit 206. This position determination is a so-called dead reckoning based on inertial navigation system 201. For this purpose, the corrected measurement values ​​output by preprocessing unit 206 or the physical quantities described therein are continuously integrated or summed over time. Strapdown algorithm unit 207 further compensates for a Coriolis force acting on inertial sensor navigation system 201, which can affect the measurement data from inertial navigation system 201. To determine the position, strapdown algorithm unit 207 performs a double integration of the measurement values ​​recorded by inertial navigation system 201, which describe accelerations, over time.This enables an update of a previously known position as well as an update of a previously known orientation of the motor vehicle. To determine a speed or a rotation rate of the motor vehicle, strapdown algorithm unit 207 performs a simple integration of the measurement data or values ​​acquired by inertial sensor navigation system 201 over time. Furthermore, strapdown algorithm unit 207 also corrects the determined position using corresponding correction values ​​from fusion filter 205. In this example, fusion filter 205 therefore only performs the correction indirectly via strapdown algorithm unit 207. The measured values ​​or physical quantities determined and corrected by strapdown algorithm unit 207, i.e., the position, speed, acceleration, orientation, rotation rate, and angular acceleration of the motor vehicle, are then passed to output module 212 and to fusion filter 205.

[0070] The so-called strapdown algorithm executed by strapdown algorithm unit 207 is computationally relatively simple and can therefore be implemented as a real-time capable basic system. It represents a process flow for integrating the measured values ​​from inertial navigation system 201 regarding speed, orientation, and position and does not involve any filtering, resulting in an approximately constant latency and group delay.

[0071] The term "base system" describes the sensor system whose measurement data or values ​​are corrected using the measurement data or values ​​of the other sensor systems, the so-called correction systems. For example, as already explained, the correction systems are odometry navigation system 203 and satellite navigation system 204.

[0072] Inertial navigation system 201, preprocessing unit 206 of inertial sensor navigation system 201 and strapdown algorithm unit 207 together form, for example, the so-called basic system, to which fusion filter 205 is also counted proportionally.

[0073] Output module 212 forwards the physical quantities determined and corrected by strapdown algorithm unit 207 to any other systems of the motor vehicle.

[0074] The measured values ​​acquired by satellite navigation system 204 are initially forwarded, for example, in the form of sensor signals via a so-called UART data connection to preprocessing unit 208 of satellite navigation system 204. Preprocessing unit 208 then determines a position and a speed of the motor vehicle in the GPS coordinate system from the measured values ​​output by satellite navigation system 204, which represent GPS raw data and also include a description of the orbit of the respective GPS satellite transmitting the GPS signals. Satellite navigation system 204 also determines a relative speed of the motor vehicle to the GPS satellites from which GPS signals are received.Furthermore, preprocessing unit 208 corrects a time error contained in the measured values ​​of a receiver clock of satellite navigation system 204, which error is caused by a drift in the receiver clock, as well as, using a correction model, the changes in the signal propagation time and signal path caused by atmospheric influences on the GPS signals transmitted by the GPS satellites. The time error and the atmospheric influences are corrected using correction values ​​received from fusion filters 205 via the CAN bus.

[0075] Satellite navigation system 204 is also assigned to plausibility module 209, which checks the measured values ​​of the physical variables output by preprocessing unit 208, i.e., the physical variables of the position and speed of the motor vehicle. The measured data or values ​​checked by plausibility module 209 are then output to fusion filter 205.

[0076] The system further comprises preprocessing unit 210 of odometry navigation system 203, which receives the measured values ​​recorded by odometry navigation system 203 via the CAN bus. In this case, the recorded measured values ​​are the measured values ​​of the individual wheel speed sensors and the measured values ​​of the steering angle sensor. Preprocessing unit 210 then determines the position and orientation of the motor vehicle in the motor vehicle coordinate system from the measured values ​​output by odometry navigation system 203 according to a so-called dead reckoning method. Furthermore, the speed, acceleration, yaw rate, and angular acceleration of the motor vehicle are also determined, likewise in the motor vehicle coordinate system. Furthermore, preprocessing unit 210 corrects the measured data or values ​​received from odometry navigation system 203 using correction values ​​obtained from fusion filter 205.

[0077] Odometry navigation system 203 is also assigned to plausibility module 211, which checks the plausibility of the measured values ​​output by preprocessing unit 210, i.e. the position, orientation, speed, acceleration, yaw rate and angular acceleration of the motor vehicle. Since the disturbances in the measured values ​​of odometry navigation system 203 are often random, environmentally caused disturbances that do not correspond to white noise, e.g. in the case of comparatively large wheel slip, the measured values ​​determined by means of inertial navigation system 201 and by means of satellite navigation system 204 are used to check the plausibility of the measured values ​​of odometry navigation system 203. First, however, the measured values ​​are compared against an assigned, sensor-specific model, which takes measurement uncertainties such as noise influences into account. If the measured values ​​correspond to the model within the given limit values ​​orIf the measured values ​​do not correspond to the tolerance ranges, an initial plausibility check is performed here, and the thus validated values ​​are further processed. The validated values ​​are then passed on to fusion filter 205. If a plausibility check for these measured values ​​cannot be performed, the corresponding measured values ​​are discarded and not further processed.

[0078] The individual modules plausibility module 209 and plausibility module 211 together form a plausibility module within the meaning of the invention with the proportional inclusion of fusion filter 205.

[0079] Fusion filter 205 is configured, for example, as an error-state space Kalman filter. The main task of fusion filter 205 is, for example, to correct the measured values ​​of the base system, i.e., the inertial navigation system 201, using measured values ​​from the odometry navigation system 203 and the satellite navigation system 204, which represent the correction systems, or to output corresponding correction values ​​to the strapdown algorithm unit 207. Since the inertial navigation system 201 is assumed, for example, to be free of random errors and external disturbances, the measured values ​​of the inertial navigation system 201 are subject exclusively to white noise. Due to the different signal output delays of the inertial navigation system 201, the odometry navigation system 203, and the satellite navigation system 204, the measured values ​​of the inertial navigation system 201 are stored over a period of 25 measurement epochs in an electronic data storage device (not shown).This ensures that measured values ​​from the inertial navigation system 201 are available for plausibility checks, corrections, or fusion for both the measured values ​​from the odometry navigation system 203 and the satellite navigation system 204, which were recorded at an identical time.

[0080] Since fusion filter 205 is a so-called error-state-space Kalman filter, only the quantitative error values ​​of the measured values ​​are determined and corresponding corrections are performed. This simplifies and accelerates the fusion of the measured values ​​from inertial navigation system 201, odometry navigation system 203, and satellite navigation system 204 into a common fusion data set performed by fusion filter 205. This enables real-time position determination and correction of the position determination.

[0081] During initialization of the described sensor system, starting values ​​for those physical quantities that are only indirectly described in the measured values ​​of inertial navigation system 201, namely the speed, the position, the receiver clock error of satellite navigation system 204 and the receiver clock error drift of satellite navigation system 204, are determined from the measured values ​​of odometry navigation system 203 and satellite navigation system 204. Likewise, the starting values ​​for the error values ​​are also determined as variances based on the measurement inaccuracies of satellite navigation system 204 and odometry navigation system 203.

[0082] The Fig. 2The system shown represents a so-called virtual sensor, although the inertial navigation system 201, odometry navigation system 203, and satellite navigation system 204 are not components of the virtual sensor. A virtual sensor is a system that always generates the same output data or outputs, regardless of the type of integrated sensor systems—here, inertial navigation system 201, odometry navigation system 203, and satellite navigation system 204. The output data or outputs do not reveal which sensor systems are integrated into the system.

Claims

1. Method for initializing a sensor fusion system when booting up or starting the sensor fusion system, wherein the sensor fusion system comprises a base system (101, 201), which is in the form of an inertia navigation system (101, 201), and at least one correction system (103, 104, 203, 204), which is in the form of a global satellite navigation system (104, 204) and / or an odometry navigation system (103, 203), wherein respective measured values are recorded by the base system (101, 201) and by the at least one correction system (103, 104, 203, 204), wherein the measured values directly or indirectly describe physical variables, wherein the measured values are each afflicted by error values, wherein the error values describe deviations of the measured values from the described physical variables, wherein at least the physical variables indirectly described in the measured values of the base system (101, 201) include a speed and / or a position and / or a receiver clock error of the global satellite navigation system (104, 204) and / or a receiver clock error drift of the global satellite navigation system (104, 204) and said variables and the error values thereof cannot be determined during the initialization of the sensor fusion system and wherein the measured values are continuously fused to form a fusion data set after the initialization has concluded, characterized in that, for at least one of the physical variables indirectly described in the measured values of the base system (101, 201) and the error values thereof, starting values are determined from the measured values of the at least one correction system (103, 104, 203, 204) during initialization, wherein the starting values for the error values are determined as variances based on the measurement inaccuracies of the at least one correction system (103, 104, 203, 204), wherein the error values are identified and corrected by means of an error state space filter (105, 205), in particular by means of an error state space Kalman filter (105, 205).

2. Method according to Claim 1, characterized in that the at least one correction system (103, 104, 203, 204) determines the starting values by means of respective independent measured value processing.

3. Method according to either of Claims 1 and 2, characterized in that the starting values are determined from the measured values of the at least one correction system (103, 104, 203, 204) in stages.

4. Method according to Claim 3, characterized in that starting values are first determined from the measured values of the odometry navigation system (103, 203).

5. System for initializing a sensor fusion system when booting up or starting the sensor fusion system, wherein the sensor fusion system comprises a base system (101, 201), which is in the form of an inertia navigation system (101, 201), and at least one correction system (103, 104, 203, 204), which is in the form of a global satellite navigation system (104, 204) and / or an odometry navigation system (103, 203), wherein the base system (101, 201) and the at least one correction system (103, 104, 203, 204) are each designed to record measured values, wherein the measured values directly or indirectly describe physical variables, wherein the measured values are each afflicted by error values, wherein the error values describe deviations of the measured values from the described physical variables, wherein the sensor fusion system is designed such that at least the physical variables indirectly described in the measured values of the base system (101, 201) include a speed and / or a position and / or a receiver clock error of the global satellite navigation system (104, 204) and / or a receiver clock error drift of the global satellite navigation system (104, 204) and said variables and the error values thereof cannot be determined during the initialization of the sensor fusion system and wherein the sensor fusion system is designed to fuse the measured values continuously to form a fusion data set after the initialization has concluded, characterized in that the sensor fusion system is designed, for at least one of the physical variables indirectly described in the measured values of the base system and the error values thereof, to determine starting values from the measured values of the at least one correction system (103, 104, 203, 204) during initialization, wherein the starting values for the error values are determined as variances based on the measurement inaccuracies of the at least one correction system (103, 104, 203, 204), wherein the error values are identified and corrected by means of an error state space filter (105, 205), in particular by means of an error state space Kalman filter (105, 205).

6. System according to Claim 5, characterized in that the system is designed to carry out a method according to at least one of Claims 2 to 4.