System and method for adjusting track using sensor data

1. A method comprising:

obtaining a first track associated with an object;

generating, based on the first track, a first set of parameters defining a first box, wherein the first box represents the first track;

obtaining measurement data from one or more sensors;

extracting a first set of features from the measurement data;

generating, based on the first set of parameters and the first set of features, a second set of parameters by a machine learning model, the second set of parameters defining a second box, wherein the second set of parameters comprise one or more parameters that are applied to the first set of parameters,

wherein the second set of parameters further comprise extents of the second box, the extents of the second box including a length and a width of the second box,

wherein generating the second set of parameters by the machine learning model comprises: (1) inputting the first set of features to the machine learning model; and (2) executing the machine learning model to output the second set of parameters, wherein the machine learning model has been trained by calculating an adjustment loss of using sample data representing a ground truth track, and updating the machine learning model based on the adjustment loss;

adjusting, based on the second set of parameters, the first track to generate a second track associated with the object; and

providing the second track to an autonomous vehicle control system for autonomous control of a vehicle.

2. The method of claim 1 , wherein

the first track is associated with a first time, and

the measurement data and the second track are associated with a second time that is later than the first time.

3. The method of claim 1 , wherein the second set of parameters define the second box to be consistent with the measurement data.

4. The method of claim 3 , wherein the second set of parameters comprise at least one of (1) an angle to rotate the first box such that a heading of the first box matches the second box or (2) a translation from the first box to the second box.

5. The method of claim 3 , wherein

the second set of parameters comprise a translation from the first box to the second box, and

the translation represents a distance between a particular portion of the first box and a particular portion of the second box.

6. The method of claim 1 , wherein the first set of features comprise at least one of (1) track or automotive vehicle (AV) metadata, (2) lidar points, (3) radar points, or (4) Hough points.

7. The method of claim 1 , wherein the adjustment loss comprises at least one of translation loss, angle loss, or extent loss.

8. A system comprising one or more processors and one or more memories operably coupled with the one or more processors, wherein the one or more memories store instructions that, in response to the execution of the instructions by the one or more processors, cause the one or more processors to perform at least the following operations of:

obtaining a first track associated with an object;

generating, based on the first track, a first set of parameters defining a first box, wherein the first box represents the first track;

obtaining measurement data from one or more sensors;

extracting a first set of features from the measurement data;

generating, based on the first set of parameters and the first set of features, a second set of parameters by a machine learning model, the second set of parameters defining a second box, wherein the second set of parameters comprise one or more parameters that are applied to the first set of parameters,

wherein the second set of parameters further comprise extents of the second box, the extents of the second box including a length and a width of the second box,

wherein generating the second set of parameters by the machine learning model comprises: (1) inputting the first set of features to the machine learning model; and (2) executing the machine learning model to output the second set of parameters, wherein the machine learning model has been trained by calculating an adjustment loss using sample data representing a ground truth track, and updating the machine learning model based on the adjustment loss;

adjusting, based on the second set of parameters, the first track to generate a second track associated with the object; and

providing the second track to an autonomous vehicle control system for autonomous control of a vehicle.

9. The system of claim 8 , wherein

the first track is associated with a first time, and

the measurement data and the second track are associated with a second time that is later than the first time.

10. The system of claim 9 , wherein the second set of parameters define the second box to be consistent with the measurement data associated with the second time.

11. The system of claim 10 , wherein the second set of parameters comprise at least one of (1) an angle to rotate the first box such that a heading of the first box matches the second box or (2) a translation from the first box to the second box.

12. The system of claim 10 , wherein

the second set of parameters comprise a translation from the first box to the second box, and

the translation represents a distance between a particular portion of the first box and a particular portion of the second box.

13. The system of claim 8 , wherein the first set of features comprise at least one of (1) track or automotive vehicle (AV) metadata, (2) lidar points, (3) radar points, or (4) Hough points.

14. The system of claim 8 , wherein the adjustment loss comprises at least one of translation loss, angle loss, or extent loss.

15. At least one non-transitory computer-readable medium comprising instructions that, in response to execution of the instructions by one or more processors, cause one or more processors to perform at least the following operations of:

obtaining a first track associated with an object;

generating, based on the first track, a first set of parameters defining a first box, wherein the first box represents the first track;

obtaining measurement data from one or more sensors;

extracting a first set of features from the measurement data;

generating, based on the first set of parameters and the first set of features, a second set of parameters by a machine learning model, the second set of parameters defining a second box, wherein the second set of parameters comprise one or more parameters that are applied to the first set of parameters,

wherein generating the second set of parameters by the machine learning model comprises: (1) inputting the first set of features to the machine learning model; and (2) executing the machine learning model to output the second set of parameters, wherein the machine learning model has been trained by (a) calculating an adjustment loss using sample data representing a ground truth track, the adjustment loss comprises a translation loss, an angle loss, and an extent loss, (b) summing the translation loss, the angle loss, and the extent loss using weights to calculate the adjustment loss, and (c) updating the machine learning model based on the adjustment loss;

adjusting, based on the second set of parameters, the first track to generate a second track associated with the object; and

providing the second track to an autonomous vehicle control system for autonomous control of a vehicle.

16. The at least one non-transitory computer-readable medium of claim 15 , wherein

the first track is associated with a first time, and

the measurement data and the second track are associated with a second time that is later than the first time.

17. The method of claim 1 , wherein the machine learning model has been trained by calculating a translation loss, an angle loss, and an extent loss using the sample data representing the ground truth track, and summing the translation loss, the angle loss, and the extent loss using weights to calculate the adjustment loss.

18. The system of claim 8 , wherein the machine learning model has been trained by calculating a translation loss, an angle loss, and an extent loss using the sample data representing the ground truth track, and summing the translation loss, the angle loss, and the extent loss using weights to calculate the adjustment loss.