Adaptation system and adaptation method
Processing circuitry of an adaptation system executes a second process when a number of times of execution of a learning routine is greater than or equal to a specified number of times and less than a termination number of times. The second process performs a first trial and a second trial in each execution of the learning routine, and ends the learning routine by recording, in a storage device, a change in one of the first trial and the second trial in which a reward is larger. The processing circuitry executes a third process when the number of times of execution of the learning routine reaches a specified number of times. The third process is a process of calculating summary statistics of multiple changes and reflecting the summary statistics in a control map to complete optimization of the control map.
1 . An adaptation system for optimizing a function used to control a motor, the system comprising:
processing circuitry; and
a storage device, wherein
the processing circuitry is configured to repeat a learning routine, thereby optimizing the function to be stored in a controller for controlling the motor, the learning routine including:
a trial that drives the motor while acquiring a state variable by a sensor in a state in which a change has been added to the function for outputting a command value to the motor;
an evaluation that calculates a reward based on the acquired state variable; and
learning that updates the function based on the reward, and the processing circuitry is configured to execute
a first process that, in a case in which a number of times of execution of the learning routine is less than a specified number of times, performs a first trial and a second trial in which the change is added to the function so as to adjust, in a sign reversing direction, the command value output from the function in each execution of the learning routine, updates the function by reflecting, in the function, the change in one of the first trial and the second trial in which the reward is larger, and ends the learning routine,
a second process that, in a case in which the number of times of execution of the learning routine is greater than or equal to the specified number of times and is less than a termination number of times that is greater than the specified number of times, performs the first trial and the second trial in each execution of the learning routine, and ends the learning routine by recording, in the storage device, the change in one of the first trial and the second trial in which the reward is larger without reflecting the change in the function, and
a third process that, in a case in which the number of times of execution of the learning routine reaches the termination number of times, calculates a summary statistic of a plurality of the changes that are stored in the storage device without being reflected in the function, ends the learning routine by reflecting the change based on the summary statistic in the function, and completes the optimization of the function.
2 . The adaptation system according to claim 1 , wherein
the function is a control map that outputs the command value to the motor in accordance with an elapsed time from a start of control,
the change in the first trial randomly adjusts, within a specified adjustment range, the value of the command value for each elapsed time range in the control map, and
the change in the second trial performs the adjustment of the command value for each elapsed time range in the change of the first trial on the command value for each elapsed time range in the control map after reversing the sign of the adjustment.
3 . The adaptation system according to claim 1 , wherein the summary statistic is an average value.
4 . The adaptation system according to claim 1 , wherein
the function is used in control of cranking an engine mounted on a vehicle by driving a crankshaft of the engine by the motor,
the trials are trials of cranking the engine by the motor to start the engine,
the sensor includes:
a crank position sensor that detects an engine rotation speed of the engine;
a microphone that detects sound generated by the vehicle; and
an acceleration sensor that detects acceleration of the vehicle, and
the state variable includes:
the engine rotation speed detected by the crank position sensor;
a sound pressure detected by the microphone; and
an acceleration detected by the acceleration sensor.
5 . An adaptation method for optimizing a function used to control a motor using an adaptation system, wherein
the adaptation system includes:
processing circuitry; and
a storage device,
the adaptation method comprises causing the processing circuitry to repeat a learning routine, thereby optimizing the function to be stored in a controller for controlling the motor, the learning routine including:
a trial that drives the motor while acquiring a state variable by a sensor in a state in which a change has been added to the function for outputting a command value to the motor;
an evaluation that calculates a reward based on the acquired state variable; and
learning that updates the function based on the reward, and
the adaptation method further comprising causing the processing circuitry to execute
a first process that, in a case in which a number of times of execution of the learning routine is less than a specified number of times, performs a first trial and a second trial in which the change is added to the function so as to adjust, in a sign reversing direction, the command value output from the function in each execution of the learning routine, updates the function by reflecting, in the function, the change in one of the first trial and the second trial in which the reward is larger, and ends the learning routine,
a second process that, in a case in which the number of times of execution of the learning routine is greater than or equal to the specified number of times and is less than a termination number of times that is greater than the specified number of times, performs the first trial and the second trial in each execution of the learning routine, and ends the learning routine by recording, in the storage device, the change in one of the first trial and the second trial in which the reward is larger without reflecting the change in the function, and
a third process that, in a case in which the number of times of execution of the learning routine reaches the termination number of times, calculates a summary statistic of a plurality of the changes that are stored in the storage device without being reflected in the function, ends the learning routine by reflecting the change based on the summary statistic in the function, and completes the optimization of the function.