Auto-labeling sensor data for machine learning

1. A method of auto-labeling sensor data for machine learning (ML), the method comprising:

determining, for a platform, platform ownship data;

recording, from an electro-optical sensor aboard the platform, sensor data comprising a plurality of sensor images;

receiving position data for at least one intruder object;

based at least on the position data for the at least one intruder object and the platform ownship data, determining a relative position and a relative velocity of the at least one intruder object relative to the platform;

for each respective sensor image of the plurality of sensor images:

based at least on the relative position and the relative velocity of the at least one intruder object and a field of view of the sensor, determining an expected position of the at least one intruder object in the respective sensor image for a time interval corresponding to the respective sensor image; and

labeling the respective sensor image, wherein the labeling comprises annotating the respective sensor image with a region of interest and an object identification, wherein the region of interest is a region in the sensor image likely to contain a view of the at least one intruder object and is determined based on the expected position of the at least one intruder object; and

training an artificial intelligence (AI) model using the labeled sensor images to produce a trained AI model for collision avoidance.

2. The method of claim 1 , wherein the sensor comprises a sensor type selected from a list consisting of: an optical camera, a light detection and ranging (LIDAR) sensor, an infrared sensor, and a radio detection and ranging (radar) sensor.

3. The method of claim 1 , further comprising:

recording sensor data from a plurality of electro-optical sensors aboard the platform; and

based at least on the relative position and relative velocity of the at least one intruder object, determining which sensor of the plurality of sensors has a view of the at least one intruder object in the time interval.

4. The method of claim 1 , wherein the position data for the at least one intruder object comprises automatic dependent surveillance-broadcast (ADS-B) data.

5. The method of claim 1 , wherein determining the region of interest comprises, based at least on a distance from the platform to the at least one intruder object, determining a bounding box within the sensor image.

6. The method of claim 1 , wherein the object identification includes a vehicle class.

7. The method of claim 1 , wherein determining the region of interest comprises, based at least on accuracy of the ownship data and accuracy of the position data or velocity data for the at least one intruder object, determining a size of the region of interest within the sensor image.

8. The method of claim 1 , further comprising:

autonomously navigating an unmanned aerial vehicle (UAV) using the trained AI model in combination with a second sensor aboard the UAV, wherein the autonomously navigating comprises collision avoidance.

9. A system for auto-labeling sensor data for machine learning (ML), the system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:

determining, for a platform, platform ownship data;

recording, from an electro-optical sensor aboard the platform, sensor data comprising a plurality of sensor images;

receiving position data for at least one intruder object;

based at least on the position data for the at least one intruder object and the platform ownship data, determining a relative position and a relative velocity of the at least one intruder object relative to the platform;

for each respective sensor image of the plurality of sensor images:

based at least on the relative position and the relative velocity of the at least one intruder object and a field of view of the sensor, determining an expected position of the at least one intruder object in the respective sensor image for a time interval corresponding to the respective sensor image; and

labeling the respective sensor image, wherein the labeling comprises annotating the respective sensor image with a region of interest and an object identification, wherein the region of interest is a region in the sensor image likely to contain a view of the at least one intruder object and is determined based on the expected position of the at least one intruder object; and

training an artificial intelligence (AI) model using the labeled sensor images to produce a trained AI model for collision avoidance.

10. The system of claim 9 , wherein the sensor comprises a sensor type selected from a list consisting of: an optical camera, a light detection and ranging (LIDAR) sensor, an infrared sensor, and a radio detection and ranging (radar) sensor.

11. The system of claim 9 , wherein the operations further comprise:

recording sensor data from a plurality of electro-optical sensors aboard the platform; and

based at least on the relative position and relative velocity of the at least one intruder object, determining which sensor of the plurality of sensors has a view of the at least one intruder object in the time interval.

12. The system of claim 9 , wherein the position data for the at least one intruder object comprises automatic dependent surveillance-broadcast (ADS-B) data.

13. The system of claim 9 , wherein determining the region of interest comprises, based at least on a distance from the platform to the at least one intruder object, determining a bounding box within the sensor image.

14. The system of claim 9 , wherein the object identification includes a vehicle class.

15. The system of claim 9 , wherein determining the region of interest comprises, based at least on accuracy of the ownship data and accuracy of the position data or velocity data for the at least one intruder object, determining a size of the region of interest within the sensor image.

16. The system of claim 9 , wherein the operations further comprise:

autonomously navigating an unmanned aerial vehicle (UAV) using the trained AI model in combination with a second sensor aboard the UAV, wherein the autonomously navigating comprises collision avoidance.

17. A computer program product, comprising a computer usable medium having a computer readable program code embodied therein, the computer readable program code adapted to be executed to implement a method of auto-labeling sensor data for machine learning (ML), the method comprising:

determining, for a platform, platform ownship data;

recording, from an electro-optical sensor aboard the platform, sensor data comprising a plurality of sensor images;

receiving position data for at least one intruder object;

based at least on the position data for the at least one intruder object and the platform ownship data, determining a relative position and a relative velocity of the at least one intruder object relative to the platform;

for each respective sensor image of the plurality of sensor images:

based at least on the relative position and the relative velocity of the at least one intruder object and a field of view of the sensor, determining an expected position of the at least one intruder object in the respective sensor image for a time interval corresponding to the respective sensor image; and

labeling the respective sensor image, wherein the labeling comprises annotating the respective sensor image with a region of interest and an object identification, wherein the region of interest is a region in the sensor image likely to contain a view of the at least one intruder object and is determined based on the expected position of the at least one intruder object; and

training an artificial intelligence (AI) model using the labeled sensor images to produce a trained AI model for collision avoidance.

18. The computer program product of claim 17 , wherein the sensor comprises a sensor type selected from a list consisting of: an optical camera, a light detection and ranging (LIDAR) sensor, an infrared sensor, and a radio detection and ranging (radar) sensor.

19. The computer program product of claim 17 , wherein the method further comprises:

recording sensor data from a plurality of electro-optical sensors aboard the platform; and

based at least on the relative position and relative velocity of the at least one intruder object, determining which sensor of the plurality of sensors has a view of the at least one intruder object in the time interval.

20. The computer program product of claim 17 , wherein the position data for the at least one intruder object comprises automatic dependent surveillance-broadcast (ADS-B) data.

21. The computer program product of claim 17 , wherein determining the region of interest comprises, based at least on a distance from the platform to the at least one intruder object, determining a bounding box within the sensor image.

22. The computer program product of claim 17 , wherein determining the region of interest comprises, based at least on accuracy of the ownship data and accuracy of the position data or velocity data for the at least one intruder object, determining a size of the region of interest within the sensor image.