Machine-learned tillage plug detection in an autonomous farming vehicle

1. A method of detecting plugs between tilling shanks by an autonomous farming vehicle, comprising:

accessing, in real-time, image data corresponding to a tilling pattern produced by the autonomous farming vehicle as the autonomous farming vehicle pulls a set of tilling shanks through earth while performing an original autonomous routine;

applying a machine-learned model to the image data, the machine-learned model trained on training image data representative of tilling patterns produced by sets of plugged tilling shanks and tilling patterns produced by sets of unplugged tilling shanks, the machine-learned model configured to detect a plug or an absence of a plug based on the image data corresponding to the tilling pattern produced by the autonomous farming vehicle;

in response to detecting a plug, performing a modified autonomous routine of the autonomous farming vehicle to remove the plug from the set of tilling shanks;

accessing additional image data as the autonomous farming vehicle pulls the set of tilling shanks through earth while performing the modified autonomous routine; and

verifying, using the additional image data, the plug is absent from the set of tilling shanks.

2. The method of claim 1 , further comprising:

accessing, in real-time, through-beam sensor data from a through-beam sensor comprising an emitter and a receiver coupled to a first and a second tilling shank, respectively, of the set of tilling shanks; and

determining a measure of a signal transmitted by the emitter and received at the receiver is below a threshold level indicative of the plug in between the first and second tilling shanks.

3. The method of claim 2 , further comprising:

determining a location of the plug among the set of tilling shanks based on an identifier associated with the through-beam sensor of a plurality of through-beam sensors coupled to the set of tilling shanks.

4. The method of claim 1 , further comprising:

accessing, in real-time, ultrasonic sensor data from an ultrasonic sensor coupled to a tilling shank of the set of tilling shanks, the ultrasonic sensor configured to indicate a distance between the tilling shank and another tilling shank of the set of tilling shanks; and

determining a measure of a signal received by the ultrasonic sensor is below a threshold level indicative of a plug in between the tilling shank and the other tilling shank.

5. The method of claim 4 , further comprising:

determining a location of the plug among the set of tilling shanks based on an identifier associated with the ultrasonic sensor of a plurality of ultrasonic sensors coupled to the set of tilling shanks.

6. The method of claim 1 , further comprising:

creating a training set comprising training tilling pattern images, each training tilling pattern image associated with a plug or associated with no plug; and

training the machine-learned model using the training set.

7. The method of claim 6 , wherein each image of the training tilling pattern images is associated with one or more plugs within the set of tilling shanks and a location of each plug within the set of tilling shanks.

8. The method of claim 6 , wherein the training tilling pattern images are captured by a plurality of autonomous farming vehicles operating in a plurality of environmental conditions.

9. The method of claim 1 , wherein modifying the autonomous routine of the autonomous farming vehicle comprises pausing the autonomous routine.

10. The method of claim 1 , further comprising:

accessing load cell data from a load cell sensor coupled to the set of tilling shanks, the load cell data associated with a plurality of draft forces associated with pulling the set of tilling shanks through the earth; and

in response to determining the plurality of draft forces exceeds a baseline draft force associated with pulling the set of tilling shanks free of plugs, determining that a plug is present.

11. The method of claim 10 , further comprising:

determining a weighted score based on the load cell data and a classification by the machine-learned model; and

detecting the plug based on the weighted score.

12. The method of claim 1 , further comprising:

accessing environmental data representative of a plurality of conditions under which the autonomous farming vehicle performs the autonomous routine;

applying a statistical model to the environmental data, the statistical model configured to estimate a likelihood of detecting the plug based on the environmental data; and

detecting the plug further based on the application of the statistical model.

13. The method of claim 12 , wherein the environmental data comprises a time, a temperature, or soil moisture level.

14. The method of claim 1 , further comprising:

determining a geographic location where the plug is detected;

determining whether a threshold number of plugs have been detected at the geographic location; and

in response to determining that the threshold number of detections has occurred at the geographic location:

generating a notification identifying the geographic location, and

transmitting the notification to a remote monitoring system.

15. The method of claim 1 , wherein verifying the plug is removed from the set of tilling shanks further comprises:

transmitting the additional image data indicating the plug is removed from the set of tilling shanks to a user of the autonomous farming vehicle; and

receiving instructions for the autonomous farming vehicle to perform the original autonomous routine.

16. The method of claim 1 , wherein verifying the plug is removed from the set of tilling shanks further comprises:

applying the machine-learned model to the additional image data, the machine-learned model additionally configured to determine the plug is absent from the set of tilling shanks based on the image data corresponding to tilling patterns produced by the autonomous farming vehicle; and

in response to determining the plug is removed from the set of tilling shanks, performing the original autonomous routine of the autonomous farming vehicle.

17. A method of detecting plugs between tilling shanks by an autonomous farming vehicle, comprising:

accessing, in real-time, image data corresponding to a tilling pattern produced by the autonomous farming vehicle as the autonomous farming vehicle pulls a set of tilling shanks through earth while performing an autonomous routine;

applying a machine-learned model to the image data, the machine-learned model trained on training image data representative of tilling patterns produced by sets of plugged tilling shanks and tilling patterns produced by sets of unplugged tilling shanks, the machine-learned model configured to detect a plug based on the image data corresponding to the tilling pattern produced by the autonomous farming vehicle wherein applying the machine-learned model to the image data comprises:

detecting a number of tilling streaks in the tilling pattern identifiable in the image data,

comparing the number of the detected tilling streaks to an expected number of tilling streaks, and

detecting the plug based on the comparison; and

in response to detecting a plug, modifying the autonomous routine of the autonomous farming vehicle.

18. A method of detecting plugs between tilling shanks by an autonomous farming vehicle, comprising:

accessing, in real-time, image data corresponding to a tilling pattern produced by the autonomous farming vehicle as the autonomous farming vehicle pulls a set of tilling shanks through earth while performing an autonomous routine;

accessing equipment data representative of an arrangement of the set of tilling shanks, wherein the arrangement of the set of tilling shanks comprises a minimum spacing between each tilling shank of the set of tilling shanks;

applying a statistical model the equipment data, the statistical model configured to estimate a likelihood of detecting the plug based on the equipment data;

applying a machine-learned model to the image data, the machine-learned model trained on training image data representative of tilling patterns produced by sets of plugged tilling shanks and tilling patterns produced by sets of unplugged tilling shanks, the machine-learned model configured to detect a plug based on the image data corresponding to the tilling pattern produced by the autonomous farming vehicle; and

in response to detecting a plug based on an output of the machine-learned model and the likelihood determined by the statistical model, modifying the autonomous routine of the autonomous farming vehicle.

19. The method of claim 18 , further comprising:

determining a first location corresponding to where the plug is first detected;

determining a second location corresponding to where the plug is determined to be absent; and

modifying the autonomous routine to include tilling a path defined in the autonomous routine, the path between the first location and the second location.

20. A detection system for an autonomous farming vehicle configured to perform steps comprising:

accessing, in real-time, image data corresponding to a tilling pattern produced by the autonomous farming vehicle as the autonomous farming vehicle pulls a set of tilling shanks through earth while performing an original autonomous routine;

applying a machine-learned model to the image data, the machine-learned model trained on training image data representative of tilling patterns produced by sets of plugged tilling shanks and tilling patterns produced by sets of unplugged tilling shanks, the machine-learned model configured to detect a plug based on the image data corresponding to the tilling pattern produced by the autonomous farming vehicle;

in response to detecting a plug, performing a modified autonomous routine of the autonomous farming vehicle to remove the plug from the set of tilling shanks;

accessing additional image data as the autonomous farming vehicle pulls the set of tilling shanks through earth while performing the modified autonomous routine; and

verifying, using the additional image data, the plug is absent from the set of tilling shanks.

21. A non-transitory computer readable storage medium storing executable instructions that, when executed by one or more processors, cause the one or more processors to perform steps comprising:

accessing, in real-time, image data corresponding to a tilling pattern produced by an autonomous farming vehicle as the autonomous farming vehicle pulls a set of tilling shanks through earth while performing an original autonomous routine;

applying a machine-learned model to the image data, the machine-learned model trained on training image data representative of tilling patterns produced by sets of plugged tilling shanks and tilling patterns produced by sets of unplugged tilling shanks, the machine-learned model configured to detect a plug based on the image data corresponding to the tilling pattern produced by the autonomous farming vehicle;

in response to detecting a plug, performing a modified autonomous routine of the autonomous farming vehicle to remove the plug from the set of tilling shanks;

accessing additional image data as the autonomous farming vehicle pulls the set of tilling shanks through earth while performing the modified autonomous routine; and

verifying, using the additional image data, the plug is absent from the set of tilling shanks.