Localization via machine learning based on perceived channel characteristics and inertial measurement unit monitoring

Abstract

Certain aspects of this disclosure provide techniques and apparatus for improved machine learning. A sequence of data records is accessed, each data record containing wireless channel measurements and inertial measurement unit (IMU) data. Known position information corresponding to at least a first data record is accessed. A first sequence of position is determined by processing a set of IMU data and the known position information using forward calculations. A second sequence of position is determined by processing a set of IMU data and the known position information using backward calculations. IMU tuning parameters are generated using the first and second sequences of position. Pseudolabels are generated for the second data record using the IMU tuning parameters and the set of IMU data. A machine learning model is trained to predict position using one or more wireless channel measurements, using the second data record and the pseudolabels.

Claims

1. A method that is executed by a processor, Accessing a sequence of data records, wherein each individual data record in the sequence of data records includes an individual set of one or more wireless channel measurements and an individual set of inertial measurement unit (IMU) data. Accessing known location information corresponding to at least the first data record of the sequence of data records, Determining a first sequence of positions based on processing the set of IMU data and at least a subset of the known position information using a forward double integral operation, Determining a second sequence of positions based on processing the set of IMU data and at least a subset of the known position information using backward double integral operations, Based on the first and second sequences of the aforementioned positions, IMU adjustment parameters are generated, Based on the IMU adjustment parameters and the set of IMU data, a pseudo-label is generated for the second data record in the sequence of data records. To predict location based on one or more wireless channel measurements, a machine learning model is trained based on the second data record and the pseudo-label. A method of execution by the processor, including the method itself.

2. The first sequence of the position is determined using the forward double integral operation. Determining the initial velocity, initial acceleration, and initial position indicated by the first data record in the sequence of data records corresponding to the first control point, Based on the initial velocity and the initial acceleration, a first sequence of velocity is generated that starts at the initial velocity and ends at the predicted final velocity. Based on the first sequence of the initial position and the velocity, a first sequence of position is generated that starts at the initial position and ends at the predicted final position. A method performed by the processor according to claim 1, including the method described in claim 1.

3. Using the aforementioned backward double integral operation to determine the second sequence of the position, To determine the actual final velocity, actual final acceleration, and actual final position reflected by the third data record in the sequence of data records corresponding to the second control point, Based on the actual final velocity and the actual final acceleration, a second velocity sequence is generated that starts at the actual final velocity and ends at the predicted initial velocity. Based on the actual final position and the second sequence of velocity, a second sequence of position is generated, which starts at the actual final position and ends at the predicted initial position. A method performed by the processor according to claim 2, including the method described in claim 2.

4. A method performed by the processor according to claim 3, wherein the IMU adjustment parameters are generated based on a first loss defined on a first and second sequence of position and a second loss defined on a first and second sequence of velocity.

5. A method performed by the processor according to claim 4, wherein the IMU adjustment parameters are further generated based on the regularization loss.

6. Based on processing wireless channel measurements from the second data record using the machine learning model, an updated location prediction is generated. Based at least partially on the updated location prediction, an updated pseudo-label is generated. To predict location based on one or more wireless channel measurements, a second machine learning model is trained based on the updated pseudo-labels, A method performed by the processor according to claim 1, further comprising:

7. The generation of the updated pseudo-label is Based on the updated position prediction and the first position prediction generated for the second data record during the forward double integral calculation, a first position loss is generated. A second position loss is generated based on the updated position prediction and the second position prediction generated for the second data record during the backward double integral calculation. Based on the first and second position losses, the IMU adjustment parameters are updated, Based on the updated IMU adjustment parameters, the updated pseudo-label is generated, A method performed by the processor according to claim 6, including the method described in claim 6.

8. The generation of the aforementioned pseudo-labels is Determining the initial velocity and initial position indicated by the first data record corresponding to the control point, Based on the initial speed and the IMU adjustment parameters, a first speed sequence is generated that starts at the initial speed and ends at the predicted speed of the second data record. Based on the first sequence of the initial position and the velocity, a third sequence of position is generated, which starts at the initial position and ends at the predicted position of the second data record. A method performed by the processor according to claim 1, including the method described in claim 1.

9. A processing system, One or more memory locations containing computer executable instructions, One or more processors configured to execute the aforementioned computer executable instructions and to cause the processing system to perform operations, The above operation is provided, Accessing a sequence of data records, wherein each individual data record in the sequence of data records includes an individual set of one or more wireless channel measurements and an individual set of inertial measurement unit (IMU) data. Accessing known location information corresponding to at least the first data record of the sequence of data records, Determining a first sequence of positions based on processing the set of IMU data and at least a subset of the known position information using a forward double integral operation, Determining a second sequence of positions based on processing the set of IMU data and at least a subset of the known position information using backward double integral operations, Based on the first and second sequences of the aforementioned positions, IMU adjustment parameters are generated, Based on the IMU adjustment parameters and the set of IMU data, a pseudo-label is generated for the second data record in the sequence of data records. To predict location based on one or more wireless channel measurements, a machine learning model is trained based on the second data record and the pseudo-label. A processing system that includes this.

10. The first sequence of the position is determined using the forward double integral operation. Determining the initial velocity, initial acceleration, and initial position indicated by the first data record in the sequence of data records corresponding to the first control point, Based on the initial velocity and the initial acceleration, a first sequence of velocity is generated that starts at the initial velocity and ends at the predicted final velocity. Based on the first sequence of the initial position and the velocity, a first sequence of position is generated that starts at the initial position and ends at the predicted final position. The processing system according to claim 9, including the following:

11. Using the aforementioned backward double integral operation to determine the second sequence of the position, To determine the actual final velocity, actual final acceleration, and actual final position reflected by the third data record in the sequence of data records corresponding to the second control point, Based on the actual final velocity and the actual final acceleration, a second velocity sequence is generated that starts at the actual final velocity and ends at the predicted initial velocity. Based on the actual final position and the second sequence of velocity, a second sequence of position is generated, which starts at the actual final position and ends at the predicted initial position. The processing system according to claim 10, including the following:

12. The processing system according to claim 11, wherein the IMU adjustment parameters are generated based on a first loss defined on a first and second sequence of position and a second loss defined on a first and second sequence of velocity.

13. The processing system according to claim 12, wherein the IMU adjustment parameter is further generated based on the regularization loss.

14. The aforementioned operation, Based on processing wireless channel measurements from the second data record using the machine learning model, an updated location prediction is generated. Based at least partially on the updated location prediction, an updated pseudo-label is generated. To predict location based on one or more wireless channel measurements, a second machine learning model is trained based on the updated pseudo-labels, The processing system according to claim 9, further comprising:

15. The generation of the updated pseudo-label is Based on the updated position prediction and the first position prediction generated for the second data record during the forward double integral calculation, a first position loss is generated. A second position loss is generated based on the updated position prediction and the second position prediction generated for the second data record during the backward double integral calculation. Based on the first and second position losses, the IMU adjustment parameters are updated, Based on the updated IMU adjustment parameters, the updated pseudo-label is generated, The processing system according to claim 14, including the following:

16. The generation of the aforementioned pseudo-labels is Determining the initial velocity and initial position indicated by the first data record corresponding to the control point, Based on the initial speed and the IMU adjustment parameters, a first speed sequence is generated that starts at the initial speed and ends at the predicted speed of the second data record. Based on the first sequence of the initial position and the velocity, a third sequence of position is generated, which starts at the initial position and ends at the predicted position of the second data record. The processing system according to claim 9, including the following:

17. One or more non-temporary computer-readable media containing computer-executable instructions, wherein when the computer-executable instructions are executed by one or more processors of a processing system, the processing system... Accessing a sequence of data records, wherein each individual data record in the sequence of data records includes an individual set of one or more wireless channel measurements and an individual set of inertial measurement unit (IMU) data. Accessing known location information corresponding to at least the first data record of the sequence of data records, Determining a first sequence of positions based on processing the set of IMU data and at least a subset of the known position information using a forward double integral operation, Determining a second sequence of positions based on processing the set of IMU data and at least a subset of the known position information using backward double integral operations, Based on the first and second sequences of the aforementioned positions, IMU adjustment parameters are generated, Based on the IMU adjustment parameters and the set of IMU data, a pseudo-label is generated for the second data record in the sequence of data records. To predict location based on one or more wireless channel measurements, a machine learning model is trained based on the second data record and the pseudo-label. One or more non-temporary computer-readable media that cause an operation including the execution of such an operation.

18. The first sequence of the position is determined using the forward double integral operation. Determining the initial velocity, initial acceleration, and initial position indicated by the first data record in the sequence of data records corresponding to the first control point, Based on the initial velocity and the initial acceleration, a first sequence of velocity is generated that starts at the initial velocity and ends at the predicted final velocity. Based on the first sequence of the initial position and the velocity, a first sequence of position is generated that starts at the initial position and ends at the predicted final position. One or more non-temporary computer-readable media according to claim 17, including the following:

19. Using the aforementioned backward double integral operation to determine the second sequence of the position, To determine the actual final velocity, actual final acceleration, and actual final position reflected by the third data record in the sequence of data records corresponding to the second control point, Based on the actual final velocity and the actual final acceleration, a second velocity sequence is generated that starts at the actual final velocity and ends at the predicted initial velocity. Based on the actual final position and the second sequence of velocity, a second sequence of position is generated, which starts at the actual final position and ends at the predicted initial position. One or more non-temporary computer-readable media according to claim 18, including the following:

20. One or more non-temporary computer-readable media according to claim 19, wherein the IMU adjustment parameters are generated based on a first loss defined on a first and second sequence of position and a second loss defined on a first and second sequence of velocity.

21. The one or more non-temporary computer-readable media according to claim 20, wherein the IMU adjustment parameters are generated based on the regularization loss.

22. The aforementioned operation, Based on processing wireless channel measurements from the second data record using the machine learning model, an updated location prediction is generated. Based at least partially on the updated location prediction, an updated pseudo-label is generated. To predict location based on one or more wireless channel measurements, a second machine learning model is trained based on the updated pseudo-labels, One or more non-temporary computer-readable media according to claim 17, further comprising:

23. The generation of the updated pseudo-label is Based on the updated position prediction and the first position prediction generated for the second data record during the forward double integral calculation, a first position loss is generated. A second position loss is generated based on the updated position prediction and the second position prediction generated for the second data record during the backward double integral calculation. Based on the first and second position losses, the IMU adjustment parameters are updated, Based on the updated IMU adjustment parameters, the updated pseudo-label is generated, One or more non-temporary computer-readable media according to claim 22, including the following:

24. The generation of the aforementioned pseudo-labels is Determining the initial velocity and initial position indicated by the first data record corresponding to the control point, Based on the initial speed and the IMU adjustment parameters, a first speed sequence is generated that starts at the initial speed and ends at the predicted speed of the second data record. Based on the first sequence of the initial position and the velocity, a third sequence of position is generated, which starts at the initial position and ends at the predicted position of the second data record. One or more non-temporary computer-readable media according to claim 17, including the following:

25. A processing system, A sequence of data records, wherein each individual data record in the sequence of data records includes an individual set of one or more wireless channel measurements and an individual set of inertial measurement unit (IMU) data, and means for accessing the sequence of data records. Means for accessing known location information corresponding to at least one data record in the sequence of data records, A means for determining a first sequence of positions based on processing the set of IMU data and at least a subset of the known position information using a forward double integral operation, A means for determining a second sequence of positions based on processing the set of IMU data and at least a subset of the known position information using a backward double integral operation, Means for generating IMU adjustment parameters based on the first and second sequences of the aforementioned positions, Means for generating a pseudo-label for a second data record in the sequence of data records based on the IMU adjustment parameters and the set of IMU data, Means for training a machine learning model based on the second data record and the pseudo-label in order to predict location based on one or more wireless channel measurements, A processing system equipped with the following features.

26. The means for determining the first sequence of the position using the forward double integral operation, Means for determining the initial velocity, initial acceleration, and initial position indicated by the first data record in the sequence of data records corresponding to the first control point, A means for generating a first velocity sequence that starts at the initial velocity and ends at a predicted final velocity, based on the initial velocity and the initial acceleration, A means for generating a first sequence of positions, which starts at the initial position and ends at a predicted final position, based on the first sequence of the initial position and the velocity, The processing system according to claim 25, comprising:

27. The means for determining the second sequence of the position using the backward double integral operation, Means for determining the actual final velocity, actual final acceleration, and actual final position reflected by the third data record of the sequence of data records corresponding to the second control point, Means for generating a second velocity sequence, which starts at the actual final velocity and ends at the predicted initial velocity, based on the actual final velocity and the actual final acceleration, Means for generating a second sequence of positions, which starts at the actual final position and ends at the predicted initial position, based on the actual final position and the second sequence of speeds, The processing system according to claim 26, comprising:

28. means for generating updated location predictions based on processing wireless channel measurements from the second data record using the machine learning model, Means for generating updated pseudo-labels based at least partially on the updated location predictions, Means for training a second machine learning model based on the updated pseudo-labels in order to predict location based on one or more wireless channel measurements, The processing system according to claim 25, further comprising the following:

29. The means for generating the updated pseudo-labels, A means for generating a first position loss based on the updated position prediction and the first position prediction generated for the second data record during the forward double integral calculation, A means for generating a second position loss based on the updated position prediction and the second position prediction generated for the second data record during the backward double integral calculation, Means for updating the IMU adjustment parameters based on the first and second position losses, means for generating the updated pseudo-label based on the updated IMU adjustment parameters, The processing system according to claim 28, comprising:

30. The means for generating the pseudo-labels, Means for determining the initial velocity and initial position indicated by the first data record corresponding to the control point, A means for generating a first sequence of speeds, which starts at the initial speed and ends at the predicted speed of the second data record, based on the initial speed and the IMU adjustment parameters, A means for generating a third sequence of position, which starts at the initial position and ends at the predicted position of the second data record, based on the first sequence of the initial position and the velocity, The processing system according to claim 25, comprising:

Images