Control device, control method, and recording medium
A control device calculates a first estimate value of a first monitored amount of a mobile object, from sensor information of a controlled object, using a transformed first model obtained by transformation of a first model having a multi-layered neural structure and activation functions of a ReLU structure, and predicting dynamic behavior of the mobile object. The transformed first model is a model for which weight coefficients, biases, and upper and lower limit values of input/output variables of neurons are set based on the first model and in which activation functions of the neurons are transformed using a linear inequality function having a binary variable. The control device calculates a second estimate value of a second monitored amount of the controlled object from a candidate value of an operation amount for controlling the second monitored amount. The control device determines a value of the operation amount based on a target value of the second monitored amount, the second estimate value, and the first estimate value.
1. A control device, comprising:
a transformed first model for which a weight coefficient, a bias, and upper and lower limit values of an input/output variable of a neuron are set based on a first model and in which an activation function of the neuron is transformed using a linear inequality function that includes a binary variable, the first model having a multi-layered neural structure and the activation function of a rectified linear unit (ReLU) structure, the first model receiving, as input, information of a sensor equipped on a controlled object, the first model predicting dynamic behavior of a mobile object in a surroundings of the controlled object;
a memory; and
a processor coupled to the memory, the processor configured to:
obtain the information of the sensor and a target value of a monitored amount of the controlled object;
use the transformed first model to calculate an estimate value, as a future value, of a monitored amount of the mobile object, from the information of the sensor;
calculate an estimate value, as a future value, of the monitored amount of the controlled object, from a candidate value of an operation amount for controlling the monitored amount of the controlled object; and
determine a value of the operation amount for controlling the monitored amount of the controlled object, based on the target value of the monitored amount of the controlled object, the estimate value of the monitored amount of the controlled object, and the estimate value of the monitored amount of the mobile object.
2. The control device according to claim 1 , wherein
the estimate value of the monitored amount of the controlled object is calculated from the candidate value of the operation amount, using a second model that receives the operation amount of the controlled object as input and predicts dynamic behavior of the controlled object.
3. The control device according to claim 1 , wherein
the candidate value of the operation amount includes a plurality of candidate values, and
the processor:
for each of the plurality of candidate values, calculates a respective control evaluation value representing a cost for controlling the monitored amount of the controlled object, the respective control evaluation value being calculated based on the calculated estimate value of the monitored amount of the controlled object, the target value of the monitored amount of the controlled object, and the estimate value of the monitored amount of the mobile object; and
based on the calculated respective control evaluation value, determines the value of the operation amount for controlling the monitored amount of the controlled object, from among the plurality of candidate values.
4. The control device according to claim 3 , wherein
the monitored amount of the controlled object includes a path of travel and a speed of the controlled object,
the monitored amount of the mobile object includes a path of travel and a speed of the mobile object, and
the processor:
calculates the respective control evaluation value, based on an error of the estimate value of the monitored amount of the controlled object with respect to the target value of the monitored amount of the controlled object, and a positional relationship between the controlled object and the mobile object, the positional relationship being identified from the estimate value of the monitored amount of the controlled object and the estimate value of the monitored amount of the mobile object.
5. The control device according to claim 4 , wherein
the positional relationship is expressed by a presence or absence of an intersection of the estimate value of the path of travel of the controlled object and the estimate value of the path of travel of the mobile object.
6. The control device according to claim 4 , wherein
the positional relationship is expressed by an error between a position of the controlled object based on an estimate value of the speed and the path of travel of the controlled object, and a position of the mobile object based on an estimate value of the speed and the path of travel of the mobile object.
7. The control device according to claim 4 , wherein
the monitored amount of the controlled object includes fuel consumption of the controlled object,
the monitored amount of the mobile object includes fuel consumption of the mobile object, and
the processor calculates the respective control evaluation value, based on an estimate value of the fuel consumption of the controlled object and an estimate value of the fuel consumption of the mobile object.
8. The control device according to claim 4 , wherein
the processor calculates the respective control evaluation value, based on a difference of a current value of the operation amount and the candidate value of the operation amount.
9. The control device according to claim 1 , wherein
the operation amount is a steering amount, a braking amount, and an acceleration amount.
10. The control device according to claim 1 , wherein
the processor outputs the determined value of the operation amount.
11. The control device according to claim 1 , wherein
the information of the sensor includes image information of a camera equipped on the controlled object.
12. The control device according to claim 1 , wherein
the information of the sensor includes information of at least any one of light detection and ranging (LiDAR), millimeter wave radar, and an ultrasonic sensor.
13. The control device according to claim 1 , wherein
the upper and lower limit values of the input/output variable are set based on a maximum value and a minimum value of the input/output variable, obtained by providing a chirp signal or an amplitude pseudo random binary signal (APRBS) signal as the input to the first model.
14. The control device according to claim 1 , wherein
the activation function is a Leaky ReLU structure.
15. The control device according to claim 1 , wherein
the target value of the monitored amount of the controlled object is obtained from a navigation system equipped on the controlled object.
16. A control method executed by a processor, the control method comprising:
obtaining information of a sensor equipped on a controlled object and a target value of a monitored amount of the controlled object;
calculating an estimate value, as a future value, of a monitored amount of a mobile object, from the obtained information of the sensor, using a transformed first model for which a weight coefficient, a bias, and upper and lower limit values of an input/output variable of a neuron are set based on a first model and in which an activation function of the neuron is transformed using a linear inequality function that includes a binary variable, the first model having a multi-layered neural structure and the activation function of a rectified linear unit (ReLU) structure, the first model receiving, as input, the information of the sensor, the first model predicting dynamic behavior of the mobile object in a surroundings of the controlled object;
calculating an estimate value, as a future value, of the monitored amount of the controlled object, from a candidate value of an operation amount for controlling the monitored amount of the controlled object; and
determining a value of the operation amount for controlling the monitored amount of the controlled object, based on the obtained target value of the monitored amount of the controlled object, the calculated estimate value of the monitored amount of the controlled object, and the calculated estimate value of the monitored amount of the mobile object.
17. A computer-readable recording medium storing therein a control program executable by a processor, the control program comprising:
an instruction for obtaining information of a sensor equipped on a controlled object and a target value of a monitored amount of the controlled object;
an instruction for calculating an estimate value, as a future value, of a monitored amount of a mobile object, from the obtained information of the sensor, using a transformed first model for which a weight coefficient, a bias, and upper and lower limit values of an input/output variable of a neuron are set based on a first model and in which an activation function of the neuron is transformed using a linear inequality function that includes a binary variable, the first model having a multi-layered neural structure and the activation function of a rectified linear unit (ReLU) structure, the first model receiving, as input, the information of the sensor, the first model predicting dynamic behavior of the mobile object in a surroundings of the controlled object;
an instruction for calculating an estimate value, as a future value, of the monitored amount of the controlled object, from a candidate value of an operation amount for controlling the monitored amount of the controlled object; and
an instruction for determining a value of the operation amount for controlling the monitored amount of the controlled object, based on the obtained target value of the monitored amount of the controlled object, the calculated estimate value of the monitored amount of the controlled object, and the calculated estimate value of the monitored amount of the mobile object.