Online machine learning for calibration of autonomous earth moving vehicles
In some implementations, the EMV uses a calibration to inform autonomous control over the EMV. To calibrate an EMV, the system first selects a calibration action comprising a control signal for actuating a control surface of the EMV. Then, using a calibration model comprising a machine learning model trained based on one or more previous calibration actions taken by the EMV, the system predicts a response of the control surface to the control signal of the calibration action. After the EMV executes the control signal to perform the calibration action, the EMV system monitors the actual response of the control signal and uses that to update the calibration model based on a comparison between the predicted and monitored states of the control surface.
1. A method comprising:
accessing, by a training system, information describing movement instructions provided to one or more earth moving vehicles (EMV);
determining, by the training system, measures of EMV movement corresponding to the movement instructions;
generating, by the training system, a training set of data correlating movement instructions to the measures of EMV movement;
training, by the training system, a machine-learned model based on the generated training set of data, the machine-learned model configured to predict an EMV movement resulting from a movement instruction;
comparing, by the training system and for target movement instructions, an EMV movement predicted by the machine-learned model to an actual EMV movement corresponding to the target movement instructions; and
in response to the actual EMV movement drifting out of line with the EMV movement predicted by the machine-learned model, updating the machine-learned model based on the comparison of the EMV movement predicted by the machine-learned model to the actual EMV movement until the machine-learned model predicts movement of the EMV to within a threshold level of precision.
2. The method of claim 1 , wherein the machine-learned model is further configured to generate a control signal for actuating a control surface of the EMV based on the predicted EMV movement resulting from the movement instruction.
3. The method of claim 2 , wherein the actual EMV movement is measured based on a state of the control surface at a time when the control signal is executed.
4. The method of claim 1 , wherein the comparison of EMV movements predicted by the machine-learned model to actual movements is performed periodically.
5. The method of claim 1 , wherein the machine-learned model is further trained based on baseline training data gathered from a set of additional EMVs similar to the EMV.
6. The method of claim 1 , wherein updating the machine-learned model comprises:
updating, by online learning techniques, the machine-learned model based on the actual movement corresponding to the target movement instructions.
7. The method of claim 1 , wherein the comparison is performed responsive to switching between an autonomous operation mode and a manual operation mode.
8. The method of claim 1 , wherein the actual movement corresponding to the target movement instructions is determined by measuring a set of action conditions based on collected sensor data.
9. A non-transitory computer-readable storage medium storing executable instructions that, when executed by a processor, cause the processor to:
access, by a training system, information describing movement instructions provided to one or more earth moving vehicles (EMV);
determine, by the training system, measures of EMV movement corresponding to the movement instructions;
generate, by the training system, a training set of data correlating movement instructions to the measures of EMV movement;
train, by the training system, a machine-learned model based on the generated training set of data, the machine-learned model configured to predict an EMV movement resulting from a movement instruction;
compare, by the training system and for target movement instructions, an EMV movement predicted by the machine-learned model to an actual EMV movement corresponding to the target movement instructions; and
in response to the actual EMV movement drifting out of line with the EMV movement predicted by the machine-learned model, update the machine-learned model based on the comparison of the EMV movement predicted by the machine-learned model to the actual EMV movement until the machine-learned model predicts movement of the EMV to within a threshold level of precision.
10. The non-transitory computer-readable storage medium of claim 9 , wherein the machine-learned model is further configured to generate a control signal for actuating a control surface of the EMV based on the predicted EMV movement resulting from the movement instruction.
11. The non-transitory computer-readable storage medium of claim 10 , wherein the actual EMV movement is measured based on a state of the control surface at a time when the control signal is executed.
12. The non-transitory computer-readable storage medium of claim 9 , wherein the comparison of EMV movements predicted by the machine-learned model to actual movements is performed periodically.
13. The non-transitory computer-readable storage medium of claim 9 , wherein the machine-learned model is further trained based on baseline training data gathered from a set of additional EMVs similar to the EMV.
14. The non-transitory computer-readable storage medium of claim 9 , wherein instructions for updating the machine-learned model further cause the processor to:
update, by online learning techniques, the machine-learned model based on the actual movement corresponding to the target movement instructions.
15. The non-transitory computer-readable storage medium of claim 9 , wherein the comparison is performed responsive to switching between an autonomous operation mode and a manual operation mode.
16. The non-transitory computer-readable storage medium of claim 9 , wherein the actual movement corresponding to the target movement instructions is determined by measuring a set of action conditions based on collected sensor data.
17. A method comprising:
accessing, by a training system, information describing movement instructions provided to one or more earth moving vehicles (EMV) and a training set of data correlating movement instructions to measures of EMV movement;
predicting, by a machine-learned model of the training system, an EMV movement resulting from a movement instruction, the machine-learned model trained based on the accessed training set of data; and
comparing, by the training system and for target movement instructions, an EMV movement predicted by the machine-learned model to an actual EMV movement corresponding to the target movement instructions; and
in response to the actual EMV movement drifting out of line with the EMV movement predicted by the machine-learned model, updating the machine-learned model based on the comparison of the EMV movement predicted by the machine-learned model to the actual EMV movement until the machine-learned model predicts movement of the EMV to within a threshold level of precision.
18. The method of claim 17 , wherein the machine-learned model is further configured to generate a control signal for actuating a control surface of the EMV based on the predicted EMV movement resulting from the movement instruction.