Control apparatus for vibration actuator, and vibration driving apparatus, interchangeable lens, imaging apparatus, and automatic stage including same
A control apparatus for a vibration actuator configured to relatively move a contact body in contact with a vibrator with respect to the vibrator using vibration occurring on the vibrator includes a control amount output unit including a trained model trained to, if an instruction about a first speed to relatively move the contact body with respect to the vibrator at is issued, output a first control amount to relatively move the contact body with respect to the vibrator.
1. A control apparatus for a vibration actuator configured to relatively move a contact body in contact with a vibrator with respect to the vibrator using vibration occurring on the vibrator, the control apparatus comprising
a control amount output unit including:
a trained model trained to, if a target speed to relatively move the contact body with respect to the vibrator at is input, output a first control amount to relatively move the contact body with respect to the vibrator, and
a learning model including a neural network (NN) configuration with a relative speed of the vibration actuator,
wherein the learning model is trained using training data with a detected speed detected in a case where the contact body is relatively moved with respect to the vibrator by the first control amount and with a target deviation as input, and a second control amount as output, and
wherein, if an instruction about the target speed is issued, the target speed and a position deviation are input to the trained model, the position deviation being a difference between a target position about which an instruction is issued to relatively move the contact body with respect to the vibrator and a detected position detected in a case where the contact body is relatively moved with respect to the vibrator by the first control amount.
2. The control apparatus for the vibration actuator according to claim 1 , wherein, if an instruction about the target speed is issued, the target speed and a speed deviation are input to the trained model, the speed deviation being a difference between the target speed and the detected speed.
3. The control apparatus for the vibration actuator according to claim 2 , wherein the trained model is trained using training data with the detected speed and a value having a same data format as that of the speed deviation as input and the first control amount as output, the speed deviation being the difference between the target speed and the detected speed.
4. The control apparatus for the vibration actuator according to claim 1 , wherein the trained model is trained using training data with the detected speed and a value having a same data format as that of the position deviation as input and the first control amount as output, the position deviation being the difference between the target position and the detected position.
5. The control apparatus for the vibration actuator according to claim 1 , further comprising a proportional-integral-derivative (PID) controller,
wherein, if the instruction about the target speed is issued, the target speed and an output of the PID controller to which the position deviation is input are input to the trained model.
6. The control apparatus for the vibration actuator according to claim 1 , wherein the first control amount is at least one of a phase difference, a frequency, and a pulse width.
7. The control apparatus for the vibration actuator according to claim 1 , wherein the first control amount is a phase difference.
8. The control apparatus for the vibration actuator according to claim 1 , wherein the first control amount is a phase difference and a frequency.
9. The control apparatus for the vibration actuator according to claim 1 , wherein the first control amount is a phase difference, a frequency, and a pulse width.
10. The control apparatus for the vibration actuator according to claim 1 ,
wherein the trained model includes an NN configuration including an input layer including one or a plurality of first neurons, a hidden layer including a plurality of second neurons, and an output layer including one or a plurality of third neurons, and
wherein the neural network configuration has a plurality of first weights connecting the first neuron(s) and the second neurons, a plurality of second weights connecting the second neurons and the third neuron(s), thresholds for the second neurons, and a threshold or thresholds for the third neuron(s).
11. The control apparatus for the vibration actuator according to claim 10 , wherein the first weights, the second weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) of the trained model are optimized.
12. The control apparatus for the vibration actuator according to claim 11 , wherein the trained model is trained by optimizing the first weights, the second weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) that are set using a random function, based on an optimization algorithm.
13. The control apparatus for the vibration actuator according to claim 11 , wherein the trained model is trained by optimizing the first weights, the second weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) that are trained based on an optimization algorithm.
14. The control apparatus for the vibration actuator according to claim 10 , further comprising a storage unit configured to store the first weights, the second weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) of the trained model,
wherein the trained model is trained by replacing the first weights, the second weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) of the trained model with the first weights, the second weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) stored in the storage unit.
15. The control apparatus for the vibration actuator according to claim 1 , wherein the trained model includes a recurrent neural network structure including an input layer, a hidden layer, an output layer, and a state layer configured to store output from the hidden layer or the output layer and output the stored output from the hidden layer or the output layer to the hidden layer.
16. The control apparatus for the vibration actuator according to claim 15 ,
wherein the trained model includes, as the recurrent neural network structure, an input layer including one or a plurality of first neurons, a hidden layer including a plurality of second neurons, an output layer including one or a plurality of third neurons, and a state layer including a plurality of fourth neurons configured to store a plurality of outputs from the plurality of second neurons or a plurality of outputs from the one or plurality of third neurons, and
wherein the trained model has a plurality of first weights assigned to a plurality of outputs from the one or plurality of first neurons to the plurality of second neurons, a plurality of second weights assigned to a plurality of outputs from the plurality of second neurons to the one or plurality of third neurons, a plurality of third weights assigned to a plurality of outputs from the plurality of fourth neurons to the plurality of second neurons, thresholds for the second neurons, and a threshold or thresholds for the third neuron(s) as parameters of the recurrent neural network structure.
17. The control apparatus for the vibration actuator according to claim 16 , wherein the first weights, the second weights, the third weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) of the trained model are optimized.
18. The control apparatus for the vibration actuator according to claim 17 , wherein the trained model is trained by optimizing the first weights, the second weights, the third weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) that are set using a random function, based on an optimization algorithm.
19. The control apparatus for the vibration actuator according to claim 17 , wherein the trained model is trained by optimizing, based on an optimization algorithm, the first weights, the second weights, the third weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) that are trained based on the optimization algorithm.
20. The control apparatus for the vibration actuator according to claim 17 , further comprising a storage unit configured to store the first weights, the second weights, the third weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) of the trained model,
wherein the trained model is trained by replacing the first weights, the second weights, the third weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) of the trained model with the first weights, the second weights, the third weights, the thresholds for the second neurons, and the threshold(s) for the third neuron(s) stored in the storage unit.
21. The control apparatus for the vibration actuator according to claim 15 , wherein the hidden layer includes a long short-term memory (LSTM) structure.
22. The control apparatus for the vibration actuator according to claim 12 , wherein the optimization algorithm is any one of adaptive moment estimation (Adam), Momentum, Root Mean Square Propagation (RMSProp), and stochastic gradient descent (SGD).
23. The control apparatus for the vibration actuator according to claim 1 , wherein the trained model is trained other than in a case where the contact body is relatively moved with respect to the vibrator.
24. The control apparatus for the vibration actuator according to claim 1 , wherein the trained model is trained in a case where the contact body is relatively moved with respect to the vibrator.
25. The control apparatus for the vibration actuator according to claim 1 , further comprising an environment sensor configured to detect a state of an environment,
wherein the trained model is trained in a case where the environment sensor detects a change in the environment.
26. The control apparatus for the vibration actuator according to claim 25 , wherein the environment sensor is at least either one of a temperature sensor and a humidity sensor.
27. The control apparatus for the vibration actuator according to claim 1 , further comprising:
a speed instruction unit configured to issue an instruction about the target speed; and
a speed detection unit configured to detect the detected speed.
28. The control apparatus for the vibration actuator according to claim 5 , further comprising:
a position instruction unit configured to issue an instruction about the target position; and
a position detection unit configured to detect the detected position.
29. A vibration driving apparatus comprising:
a vibration actuator configured to relatively move a contact body in contact with a vibrator with respect to the vibrator using vibration occurring on the vibrator; and
the control apparatus for the vibration actuator according to claim 1 .
30. An interchangeable lens comprising:
the vibration driving apparatus according to claim 29 ; and
a lens to be driven by relatively moving the contact body with respect to the vibrator.
31. An imaging apparatus comprising:
the vibration driving apparatus according to claim 29 ; and
an image sensor to be driven by relatively moving the contact body with respect to the vibrator.
32. An automatic stage comprising:
the vibration driving apparatus according to claim 29 ; and
a stage to be driven by relatively moving the contact body with respect to the vibrator.
33. A method for outputting a control amount for a vibration actuator configured to relatively move a contact body in contact with a vibrator with respect to the vibrator using vibration occurring on the vibrator and configured to have a control amount output unit, the method comprising
performing machine learning using training data with a second speed as input and a first control amount as output, the second speed being detected in a case where the contact body is relatively moved with respect to the vibrator by the first control amount, the first control amount being intended to relatively move the contact body with respect to the vibrator in a case where an instruction about a first speed to relatively move the contact body with respect to the vibrator at is issued
wherein the control amount output unit includes:
a trained model trained to, if a target speed to relatively move the contact body with respect to the vibrator at is input, output the first control amount to relatively move the contact body with respect to the vibrator; and
a learning model including an NN configuration with a relative speed of the vibration actuator,
wherein the learning model is trained using training data with a detected speed detected in a case where the contact body is relatively moved with respect to the vibrator by the first control amount and with a target deviation as input, and a second control amount as output,
wherein, if an instruction about the target speed is issued, the target speed and a position deviation are input to the trained model, the position deviation being a difference between a target position about which an instruction is issued to relatively move the contact body with respect to the vibrator and a detected position detected in a case where the contact body is relatively moved with respect to the vibrator by the first control amount.
34. A storage medium storing a program for causing a computer to execute the method for outputting the control amount for the vibration actuator according to claim 33 .