DE102019000804B4 - Process and device for precise position determination and creation of up-to-date maps with sensor fusion - Google Patents

Abstract

This invention includes a method and an apparatus for precise position determination and the creation of highly topical maps. The method merges the distance measurements from a lidar, the images from cameras, the pseudorange, carrier phase and Doppler measurements from receivers of satellite navigation systems, the wheel speeds and steering angles from odometry, and the acceleration and yaw rate measurements from inertial sensors. Initially, an artificial intelligence model (neural network) is used to estimate the movement with the camera and lidar data and the measurements of the wheel odometry and inertial sensors. The resulting motion estimation is then fused with the pseudorange, carrier phase and Doppler measurements from receivers of satellite navigation systems in a Kalman filter. The unknown integer and temporally constant ambiguities of the carrier phase measurements of the receivers of satellite navigation systems are also estimated as state parameters in the Kalman filter and, after a certain convergence time, are fixed to integer values. The method also includes functional monitoring of the sensor data and the estimated position parameters and their statistics. Function monitoring is carried out using an artificial intelligence model (neural network).

Description

State of the art with references

Blösch et al. [ 1 ] have developed a visual-inertial odometry in which feature areas in camera images are reliably tracked and the position and location of the camera are estimated from the camera images and measurements of an inertial sensor using an extended Kalman filter. In addition to the position and location, the distances and direction vectors from the camera to the feature areas are estimated as their own status parameters. The photometric error is evaluated directly for the status update.

Hess et al. [2] have developed a method that recognizes a location at which distance measurements are carried out with a lidar sensor and where distance measurements have already been carried out with a lidar sensor, and the position accuracy of the simultaneous localization and mapping (SLAM) with lidar measurements through the recognition of locations.

Henkel et al. [ 3 ] have investigated a sensor fusion of a local ultra-broadband-based positioning system with wheel odometry and inertial sensors and a sensor fusion of camera, wheel odometry and inertial sensors. A Kalman filter and the Robust Visual-Inertial Odometry (ROVIO) method from [ 1 ] is used. A position solution that is precise to the centimeter was achieved.

[ 1 ] M. Blösch, M. Burri, S. Omary, M. Hutter and R. Siegwart: IEKF-based Visual-Inertial Odometry using Direct Photometry Feedback, J. of Robotics Research, vol. 36, no. 10, pp. 1053-1072, 2017.

[2] W. Hess, D. Kohler, H. Rapp and D. Andor: Real-Time Loop Closure in 2D Lidar SLAM, IEEE Proc. of ICRA, Stockholm, Sweden, pp. 1271 - 1278, May 2016 .

[3] P. Henkel, A. Sperl, U. Mittmann, P. Färber and C. Günther: Precise Positioning of Robots with Fusion of GNSS, INS, Odometry, Barometer, Local Positioning System and Visualization, Proc. of the 31st Intern. Techn. Meeting of the Institute of Navigation (ION GNSS +), Miami, USA, pp. 3078-3087, Sep. 2018 .

Task and objective

The invention specified in claim 1 is based on the problem of determining the position of a vehicle or robot very precisely and reliably with inexpensive sensors. High availability and reliable knowledge of the current position accuracy are also of great importance.

the solution of the problem

• - dass Messungen (11) von Empfängern (9) von Satellitennavigationssystemen, Messungen (6) der Odometrie (5), Messungen (8) von Inertialsensoren (7), Kameradaten (4) von einer Kamera (3) und Lidardaten (2) von einem Lidar (1) oder eine Teilmenge dieser Messungen und Daten zur Positionsbestimmung verwendet werden, und
• - dass eine Karte (24) der Umgebung mit den Kameradaten (4) und Lidardaten (2) und den Messungen (6) der Odometrie (5) und Inertialsensoren (7) oder einer Teilmenge dieser Daten und Messungen, und einem Modell zur künstlichen Intelligenz (15) erstellt wird, und
• - dass eine Bewegungsschätzung und Lokalisierung (16) in einer Karte (24) mit den Kameradaten (4) und Lidardaten (2) und den Messungen (6) der Odometrie (5) und Inertialsensoren (7) oder einer Teilmenge dieser Daten und Messungen, und einem Modell zur künstlichen Intelligenz (15) durchgeführt wird, und
• - dass die Positionsparameter (25) und deren Statistiken mit einem Kalman Filter (19, 20) bestimmt werden, das die Messungen (11, 12) von Empfängern (9, 10) von Satellitennavigationssystemen, die Messungen (6) der Odometrie (5), die Messungen (8) der Inertialsensoren (7) und die Bewegungsschätzung und Lokalisierung (18) in einer Karte (24) aus den Kameradaten (4) und Lidardaten (2) oder einer Teilmenge der Messungen, der Bewegungsschätzung und der Lokalisierung verwendet, und
• - dass die Sensordaten (6, 8, 14), die Positionsparameter (25) und deren Statistiken oder eine Teilmenge dieser Parameter und Statistiken mit einem Modell der künstlichen Intelligenz (21) überwacht (22) werden.

This problem is solved by the features listed in claim 1. The features include

• - that measurements ( 11 ) of recipients ( 9 ) of satellite navigation systems, measurements ( 6th ) odometry ( 5 ), Measurements ( 8th ) of inertial sensors ( 7th ), Camera data ( 4th ) from a camera ( 3 ) and lidar data ( 2 ) from a lidar ( 1 ) or a subset of these measurements and data are used to determine the position, and
• - that a card ( 24 ) the environment with the camera data ( 4th ) and lidar data ( 2 ) and the measurements ( 6th ) odometry ( 5 ) and inertial sensors ( 7th ) or a subset of these data and measurements, and an artificial intelligence model ( 15th ) is created, and
• - that a motion estimation and localization ( 16 ) in a card ( 24 ) with the camera data ( 4th ) and lidar data ( 2 ) and the measurements ( 6th ) odometry ( 5 ) and inertial sensors ( 7th ) or a subset of these data and measurements, and an artificial intelligence model ( 15th ) is performed, and
• - that the position parameters ( 25th ) and their statistics with a Kalman filter ( 19th , 20th ) can be determined that the measurements ( 11 , 12th ) of recipients ( 9 , 10 ) of satellite navigation systems that take measurements ( 6th ) odometry ( 5 ), the measurements ( 8th ) of the inertial sensors ( 7th ) and the motion estimation and localization ( 18th ) in a card ( 24 ) from the camera data ( 4th ) and lidar data ( 2 ) or a subset of the measurements, motion estimation and localization are used, and
• - that the sensor data ( 6th , 8th , 14th ), the positional parameters ( 25th ) and their statistics or a subset of these parameters and statistics with a model of artificial intelligence ( 21st ) supervised ( 22nd ) will.

Advantages of the solution

• - dass eine absolute, zentimeter-genaue Positionslösung durch die Verwendung von Trägerphasen-Messungen von mindestens einem Satellitennavigationsempfänger und einer weiteren Referenzstation, und die Auflösung der Mehrdeutigkeiten der Trägerphasen-Messungen erreicht wird, und
• - dass eine höhere Robustheit und Verfügbarkeit der Positionslösung durch die Fusionierung von mehreren Sensoren mit komplementären Eigenschaften erzielt werden kann als dies mit einem einzelnen Sensor möglich ist, und
• - dass eine hochgenaue und aktuelle Karte aus den Kamerabildern und der hochgenauen Positionslösung bestimmt wird, und
• - dass die Sensordaten als auch die geschätzten Positionsparameter und deren Statistiken mit einem Modell der künstlichen Intelligenz überwacht werden, und
• - dass ein sehr effizienter Ansatz mit einem Kalman Filter zur Sensorfusion und einem neuronalen Netz zur Kamera- und Lidar-basierten Bewegungsschätzung und Lokalisierung in einer Karte realisiert wird.

The advantages achieved with the invention are

• - that an absolute, centimeter-accurate position solution is achieved through the use of carrier phase measurements from at least one satellite navigation receiver and a further reference station, and the resolution of the ambiguities of the carrier phase measurements, and
• - that greater robustness and availability of the position solution can be achieved by merging several sensors with complementary properties than is possible with a single sensor, and
• - that a highly accurate and current map is determined from the camera images and the highly accurate position solution, and
• - that the sensor data as well as the estimated position parameters and their statistics are monitored with an artificial intelligence model, and
• - that a very efficient approach with a Kalman filter for sensor fusion and a neural network for camera and lidar-based motion estimation and localization will be implemented in a map.

Detailed explanation of the claims / technical description of the invention

An embodiment of the invention is in 1 and is described in more detail below.

1 shows an example of a flowchart for the method for determining the position of an object with multiple sensors. The distance measurements ( 2 ) a lidar ( 1 ), the pictures ( 4th ) one or more mono or stereo cameras ( 3 ), the wheel speeds and steering angles ( 6 ) the wheel odometry ( 5 ) and the acceleration, yaw rate and magnetic field measurements ( 8th ) an inertial sensor ( 7th ) or a subset of these measurements and data used to match an artificial intelligence model ( 15th ) a lidar / visual based odometry solution ( 16 ) to be determined. This includes a motion estimation ( 18th ), ie the change in position and location. In addition, a 3D map is created and a scene interpretation ( 17th ) so that a card ( 24 ) is available with semantic information. The absolute position and attitude determination is carried out with a Kalman filter, which is based on a state prediction ( 19th ) and a status update ( 20th ) consists. The state parameters of the Kalman filter include the position and location ( 25th ) of the object and optionally also the speed, acceleration and rotation rates of the object. For the status update, the motion estimation ( 18th ), the wheel speeds and steering angles ( 6 ) the wheel odometry ( 5 ), the acceleration, yaw rate and magnetic field measurements ( 8th ) an inertial sensor ( 7th ) and the corrected measurements ( 14th ) from at least one recipient ( 9 ) of satellite navigation systems and the corrected measurements ( 14th ) from another recipient ( 10 ) used by satellite navigation systems or a subset of these measurements and data. The last-mentioned receiver of satellite navigation systems serves as a reference station for correcting common or similar errors such as orbit errors, clock errors or biases of the satellites. The Kalman filter is a recursive filter, i.e. the state parameters ( 25th ) will be after the update ( 20th ) traced back to state prediction ( 19th ). The status parameters of the position and location as well as their statistics are also one of the output variables ( 23 ) this procedure.
The last step of this procedure involves function monitoring ( 22nd ) with a model of artificial intelligence ( 21st ). The motion estimation ( 18th ), the corrected measurements ( 14th ) of satellite navigation systems, the wheel speeds and steering angles ( 6 ) the wheel odometry ( 5 ) and the acceleration, yaw rate and magnetic field measurements ( 8th ) an inertial sensor ( 7th ) and the position parameters and their statistics ( 25th ) of the Kalman filter or a subset of these measurements, data, parameters and statistics. The result of the function monitoring is status information ( 26th ) for the quality of the relevant measurements, data, parameters and statistics.

An advantageous embodiment of the invention is specified in claim 2. The development according to claim 2 enables an absolute position and speed to be determined. For this purpose the pseudorange, carrier phase and Doppler measurements ( 11 ) on one or more frequencies or a linear combination of measurements on the different frequencies from at least one receiver ( 9 ) used by satellite navigation systems. The pseudorange and carrier phase measurements ( 11 ) depend not only on the position and the clock error of the receiver, but also on the positions, clock errors and biases of the satellites, and atmospheric delays. An estimate of the positions and the clock errors of the satellites is transmitted by them directly in the navigation message from the satellites to the receiver, so that the pseudorange and carrier phase measurements can be corrected in order to eliminate the influence of the positions and the clock errors of the satellites on the measurements. Correction to the pseudorange and carrier phase measurements is determined by projecting the satellite position in the direction from the satellite to the receiver and adding the satellite's clock error estimate ( 13th ). The corrected pseudorange and carrier phase measurements ( 14th ) are then used to update the status ( 20th ) used in the Kalman filter.

In addition to the pseudorange and carrier phase measurements, the Doppler measurements are also corrected. The Doppler measurements are of the speed and clock drift of the receiver and the satellite dependent. The speed and the clock drift of the satellites are transmitted from the satellites to the receiver in the navigation message of the satellites, so that the Doppler measurements can be corrected. The correction of the Doppler measurements is determined in a similar way to the pseudorange and carrier phase corrections, ie the satellite speeds are first projected onto the direction between the satellites and the receiver and then the clock drift of the satellites is added to the projected satellite speed.

The pseudorange and carrier phase measurements ( 12th ) of the reference station ( 10 ) are corrected in the same way as the pseudorange and carrier phase measurements ( 11 ) recipient ( 9 ) of satellite navigation systems.

Another advantageous embodiment of the invention is specified in claim 3. The further development according to claim 3 enables an absolute, centimeter-accurate position solution to be achieved. This is achieved by initially positioning the receivers with centimeter accuracy ( 9 ) of satellite navigation systems and the reference station ( 10 ) is determined, and then the relative position between the receivers ( 9 ) of satellite navigation systems to the very precise, measured, absolute position of the reference station ( 10 ) is added. A centimeter-accurate relative position is determined by first taking the pseudorange, carrier phase and Doppler measurements between the receivers ( 9 ) of satellite navigation systems and the reference station ( 10 ) are subtracted in order to eliminate orbital errors, clock errors and biases of the satellites and atmospheric delays, and then from these differential measurements with a Kalman filter ( 19th , 20th ) the relative position and the ambiguities of the differential carrier phase measurements are determined. The ambiguities of the differential carrier phase measurements are integers. This means that the real-valued estimates of the ambiguities determined by the Kalman filter can be fixed to integer values after a certain convergence time. The estimated integer ambiguities can then be subtracted from the carrier phase measurements so that the ambiguities no longer appear as state parameters in the Kalman filter ( 19th , 20th ) must be determined. The accuracy of the resulting positional solution with fixed ambiguities is on the order of the accuracy of the carrier phase measurements. The noise of the carrier phases is in the order of magnitude of approx. 5 mm, whereby in addition to the noise also the multipath errors of the carrier phase measurements limit the accuracy to a few centimeters.

Another advantageous embodiment of the invention is specified in claim 4. The development according to claim 4 enables the raw sensor data ( 6th ) the wheel odometry ( 5 ) for motion estimation and sensor fusion. The integration of the raw sensor data instead of position or speed information has the advantage that a close coupling between the various sensors is possible and position information can therefore also be determined if individual measurement data fail. The sensor raw data ( 6th ) the wheel odometry ( 5 ) contain the measurements of the individual wheel speeds or rotation rates and the steering angle or a subset of the measurements of the individual wheel speeds or rotation rates and the steering angle.

Another advantageous embodiment of the invention is specified in claim 5. The further development according to claim 5 enables the raw sensor data ( 8th ) an inertial sensor ( 7th ) can be used for motion estimation and for sensor fusion, thus enabling close coupling. The sensor raw data ( 8th ) of the inertial sensor ( 7th ) include the measurements of the acceleration in three orthogonal directions, the rotation rates around three orthogonal axes, and the magnetic field in three orthogonal directions, or a subset of these measurements. The measurements ( 8th ) of the inertial sensor ( 7th ) are, however, affected by systematic errors (biases), which have a different temporal stability depending on the quality and price of the respective model. In addition to the acceleration of the object, the acceleration measurements are also dependent on gravity. The sensor raw data ( 8th ) of the inertial sensor ( 7th ) are therefore before further processing for motion estimation in ( 16 , 17th ) and before the Kalman filter based sensor fusion ( 19th , 20th ) corrected for the influence of biases and gravity.

Another advantageous embodiment of the invention is specified in claim 6. The development according to claim 6 makes it possible to determine the absolute or relative three-dimensional position and the three-dimensional orientation or a subset of these position parameters. The absolute, three-dimensional orientation of an object can with 3 permanently installed receivers ( 9 ) can be determined by satellite navigation systems. First, the relative positions between the recipients ( 9 ) is determined in a local coordinate system with the differential pseudorange and carrier phase measurements, with the axes of the local coordinate system oriented either to east, north and above or to north, east and below. In addition, the relative positions of the recipients ( 9 ) of satellite navigation systems on the object measured in a local, object-fixed coordinate system after assembly. The relative orientation of the two coordinate systems contains information about the orientation of the object. This orientation can be described with the three angles of position (roll, pitch and yaw angles) from which a rotation matrix can be formed. The rotation matrix is determined in such a way that the quadratic deviation between the estimate of the relative positions in the local coordinate system oriented to the east, north and above or to the north, east and below, and that after the rotation matrix has been applied to the estimate of the relative positions in the local object-fixed coordinate system is minimized becomes. This is the classic Wahba's problem, which can be solved with a singular value decomposition. The differential carrier phase measurements are ambiguous, so that in addition to the relative positions between the receivers ( 9 ) of satellite navigation systems, the integer ambiguities of the differential carrier phase measurements must also be determined.

As soon as these ambiguities are fixed on integer values and the Wahba's problem is solved, a highly accurate absolute orientation is available. This is then predicted with the Kalman filter ( 19th ) and with the camera / lidar-based motion estimation ( 18th ), the wheel speeds and steering angles ( 6th ) the wheel odometry ( 5 ), and the yaw rate measurements ( 8th ) of the inertial sensor ( 7th ) updated. This makes it possible to have a highly precise orientation even if the reception of signals from satellite navigation systems is severely restricted.

Another advantageous embodiment of the invention is specified in claim 7. The further development according to claim 7 makes it possible to determine the three-dimensional speed, the three-dimensional acceleration and the three-dimensional rotation rates or a subset of these parameters in addition to the position and location. Linear information about the three-dimensional speed of the object is contained in the wheel speeds ( 6 ) the wheel odometry ( 5 ) and in the corrected Doppler measurements ( 14th ) recipient ( 9 ) from satellite navigation systems. In addition, the camera / lidar-based motion estimation ( 18th ) taking into account a model ( 15th ) speed information for artificial intelligence. The acceleration sensor of the inertial sensor provides linear information about the three-dimensional acceleration ( 7th ), and linear information about the three-dimensional rotation rates is provided by the gyroscope / rotation rate sensor of the inertial sensor ( 7th ). The three-dimensional speed, the three-dimensional acceleration and the three-dimensional yaw rates or a subset of these parameters are therefore used as state parameters in the Kalman filter ( 19th , 20th ) estimated. According to the state prediction ( 19th ) the state is updated ( 20th ), where the wheel speeds ( 6 ) the wheel odometry ( 5 ), the corrected Doppler measurements ( 14th ) recipient ( 9 ) of satellite navigation systems that use camera / lidar-based motion estimation ( 18th ) and the acceleration and yaw rate measurements corrected for the influence of biases and gravity ( 8th ) of the inertial sensor ( 7th ) be used.

Another advantageous embodiment of the invention is specified in claim 8. The further development according to claim 8 enables a robust camera / lidar-based motion estimation ( 18th ) with an artificial intelligence model ( 15th ) to obtain. The artificial intelligence model corresponds to a neural network. The advantage of the neural network / model for artificial intelligence is that it does not require any modeling of the measurements and their statistics, which is often error-prone. Errors in preprocessing, for example when correcting phase jumps and / or gravitation, cannot be completely avoided and recognized. In addition, unknown temporal correlations due to multipath errors in the Kalman filter are not taken into account and often lead to an underestimation of the actual uncertainty of the state parameters. The artificial intelligence model ( 15th ) consists of several consecutive layers. The input variables of the first layer are the camera ( 4th ) and lidar data ( 2 ), the measurements ( 6th ) odometry ( 5 ) and the measurements ( 8th ) of the inertial sensors ( 7th ) or a subset of these data and measurements. The input variables of the other layers are identical to the output variables of the previous layer. An output variable of a layer is determined by forming a linear combination of all input variables of this layer, adding a bias and mapping the intermediate result obtained in this way to the output value with a function. For example, the sigmoid function can be used with the function, which has a range of values between 0 and 1 has a linear range around 0 and goes into saturation with larger arguments. The output variables of the last layer correspond to the motion estimation ( 18th ) of the object, ie the change in position and location and the speed and rotation rates.

Another advantageous embodiment of the invention is specified in claim 9. The further development according to claim 9 enables the parameters of the model for artificial intelligence ( 15th ) to be determined. The parameters include the coefficients / weights of the linear combinations and the biases for all layers, and are compared with a reference solution Position parameters determined. A typical criterion for determining the parameters of the artificial intelligence model ( 15th ) is the minimization of the mean square deviation between the position parameters of the reference solution and those of the artificial intelligence model ( 15th ) calculated position parameters.

Another advantageous embodiment of the invention is specified in claim 10. The development according to claim 10 enables the parameters of the model for artificial intelligence ( 15th ) efficiently. The coefficients / weights of the linear combinations and the biases are determined for each layer using an iterative Gauss-Newton method and the steepest descent method: For this purpose, the gradients are calculated from the mean square deviation between the position parameters of the reference solution and those calculated by the model of artificial intelligence Position parameters determined according to the coefficients / weights of the linear combinations and the biases. The coefficients / weights of the linear combinations and the biases result in an iteration step from the coefficients / weights and biases of the previous iteration step, in which these are corrected by the scaled gradients. The iterations are ended as soon as the difference in the coefficients / weights and biases of two successive iterations is small enough.

Another advantageous embodiment of the invention is specified in claim 11. The development according to claim 11 makes it possible to find a good compromise between the effort required to train the neural network / model of the artificial intelligence ( 15th ) and the performance of the neural network ( 15th ) to find. The number of layers of the neural network ( 15th ) is an important parameter because the number of coefficients / weights and biases increases linearly with the number of layers. On the other hand, a higher number of layers also enables better mapping of complex models. The optimal number of layers is therefore determined with a certain database of raw sensor data and a reference solution of the position parameters.

Another advantageous embodiment of the invention is specified in claim 12. The further development according to claim 12 enables the performance of the neural network / model of artificial intelligence ( 15th ) through the integration of satellite navigation-based position information. The position parameters ( 25th ) and their statistics of the Kalman filter after the status update ( 20th ) into the neural network / model of artificial intelligence ( 16 ). The corrected measurements ( 14th ) recipient ( 9 ) position parameters derived from satellite navigation systems motion estimation and localization ( 16 ) in a card ( 24 ). The differential, carrier phase based position solution with receivers ( 9 ) of satellite navigation systems has centimeter accuracy with good satellite visibility and is therefore a valuable support for the neural network / model of artificial intelligence ( 15th ).

Another advantageous embodiment of the invention is specified in claim 13. The further development according to claim 13 makes it possible to build a device that contains a processor platform on which the method for determining position with sensor fusion described in one of claims 1 to 12 is implemented, and in which the sensors are connected to the processor via interfaces. Platform can be connected.

List of reference symbols

1

Lidar

2

Lidar data

3

Camera (s)

4th

Camera data

5

Wheel odometry

6th

Rad odometry measurements

7th

Inertial sensors

8th

Measurements of the inertial sensors

9

Receiver of satellite navigation systems

10

Receiver of satellite navigation systems (reference station)

11, 12

Measurements of a receiver of satellite navigation systems

13

Correction of the measurements of a receiver of satellite navigation systems

14th

corrected measurements of a receiver of satellite navigation systems

15th

Artificial intelligence model

16

Motion estimation and localization

17th

3D map creation and scene interpretation

18th

Motion estimation and localization measurements

19th

State prediction Kalman filter

20th

Status update Kalman filter

21st

Artificial intelligence model

22nd

Function monitoring

23

Output variables Kalman filter

24

Map of the area

25th

State parameters Kalman filter

26th

Function monitoring result

Claims (13)

Method for the precise determination of the position of an object and for the creation of highly up-to-date maps with several sensors, characterized in that - measurements (11, 12) from receivers (9, 10) of satellite navigation systems, measurements (6) of the odometry (5), measurements ( 8) of inertial sensors (7), camera and lidar data (4, 2) or a subset of these measurements and data are used to determine the position, and - that a map of the surroundings (24) with the camera and lidar data (4, 2) and the measurements (6, 8) of the odometry (5) and inertial sensors (7) or a subset of these data and measurements, and an artificial intelligence model (15) is created, and - that a motion estimation and localization (16) in one Map with the camera and lidar data (4, 2) and the measurements (6, 8) of the odometry (5) and inertial sensors (7) or a subset of these data and measurements, and a model for artificial intelligence (15) is carried out, and - that the Position parameters (25) and their statistics are determined with a Kalman filter, the measurements (11, 12) of receivers (9, 10) of satellite navigation systems, the measurements (6) of the odometry (5), the measurements (8) of the inertial sensors (7) and the motion estimation and localization (18) in a map from the camera and lidar data (4, 2) or a subset of the measurements, the motion estimation and the localization (16), and - that the sensor data (6, 8 , 14, 18), the position parameters (25) and their statistics or a subset of these parameters and statistics with an artificial intelligence model (21) are monitored (22). Procedure according to Claim 1 - characterized in - that the measurements (11, 12) of receivers (9, 10) of satellite navigation systems contain the pseudorange, carrier phase and Doppler measurements of the satellites on one frequency or several frequencies or a linear combination of the measurements on the different frequencies, and - That the pseudorange and carrier phase measurements are corrected before the position determination, the correction (13) including an estimate of the position, the clock error and the bias of a satellite, and the Doppler measurements are corrected, the correction (13) of the Doppler measurements being an estimate of the Speed of the satellites and the clock drift of the satellites. Procedure according to the Claims 1 to 2 - characterized in - that the measurements (11, 12) or corrected measurements (14) from receivers (9, 10) of satellite navigation systems from the measurements (11, 12) or corrected measurements (14) from the measurements (11, 12) of a further receiver are subtracted from satellite navigation systems, and that the ambiguities of the corrected and differentiated carrier phase measurements are determined and subtracted from the corrected carrier phase measurements. Procedure according to Claim 1 , characterized in that the measurements (6) of the odometry (5) contain the measurements of the wheel speeds and the steering angle or a subset of the measurements of the wheel speeds and the steering angle. Procedure according to Claim 1 , characterized in - that the measurements (8) of the inertial sensor (7) include measurements of the acceleration in three orthogonal directions, the rotation rates around three orthogonal axes, and the magnetic field in three orthogonal directions, or a subset of these measurements, and - that the Measurements (8) can be corrected in order to reduce the influence of gravity and biases on the measurements. Procedure according to Claim 1 , characterized in that the position parameters (25) contain the absolute or relative three-dimensional position and the three-dimensional orientation or a subset of these position parameters. Procedure according to Claim 1 , characterized in that in addition to the position parameters (25) also the three-dimensional speed, the three-dimensional acceleration and the three-dimensional rotation rates or a subset of these parameters are determined. Method according to one of the Claims 1 to 7th , characterized - that the model of the artificial intelligence (15) consists of several layers, and - that the input variable of the first slice is the camera and lidar data (4, 2) and the measurements (6, 8) of the odometry (5) and inertial sensors (7) or a subset of these data and measurements, and that the input variables of the others Layers are equal to the output variables of the previous layer, and - that an output variable of a layer is formed in which a linear combination of all input variables of this layer is formed, a bias is added and the intermediate result thus obtained is mapped to the output value with a function, that the output variables of the last slice correspond to the motion estimation (18) of the object. Method according to one of the Claims 1 to 8th , characterized in that the coefficients of the linear combinations and the biases of the layers of the model for artificial intelligence (15) are determined with sensor data (2, 4, 6, 8) and a reference solution of the position parameters (25). Method according to one of the Claims 1 to 9 , characterized in that the iterative Gauss-Newton method with the steepest descent method is used to determine the coefficients of the linear combinations and the biases. Method according to one of the Claims 1 to 10 , characterized in that the optimal number of layers of the model for artificial intelligence (15) is determined with sensor data (2, 4, 6, 8) and a reference solution of the position parameters (25). Method according to one of the Claims 1 to 11 , characterized in - that the position parameters (25) and their statistics of the Kalman filter are fed back into the motion estimation and localization (16) in a map (24), and - that the fed back position parameters (25) for motion estimation and localization (16) can be used in a card (24) with the model of artificial intelligence (15). Device for precise position determination and creation of highly accurate maps with sensor fusion, characterized in that - that the device contains a processor platform on which the in one of the Claims 1 to 12th The described method for high-precision position determination and the creation of up-to-date maps is implemented with sensor fusion, and that the sensors (1, 3, 5, 7, 9, 10) are connected to the processor platform via interfaces.

Images