Moving body control apparatus, moving body control method, and training method

1. A moving body control apparatus for controlling a moving body, the moving body control apparatus comprising:

an acquisition device for acquiring one control command among N control commands for the moving body and an image of a view in a traveling direction of the moving body, where N is an integer greater than or equal to 2; and

an information processing device that includes a feature amount extraction machine learning model and N control command machine learning models respectively corresponding to the N control commands,

wherein the information processing device uses the feature amount extraction machine learning model to calculate, from the image acquired by the acquisition device, a feature amount associated with the one control command acquired by the acquisition device and indicating a feature of an object, selects one control command machine learning model corresponding to the one control command from among the N control command machine learning models, and uses the one control command machine learning model to output a control parameter for controlling the moving body, using the one control command acquired by the acquisition device and the feature amount as inputs,

each of the N control commands is a signal indicating an intention to maneuver the moving body,

the N control commands include a first control command which is one of turn left, turn right, stop, go forward, make a U-turn, and change lanes, and

the N control command machine learning models include a first control command machine learning model which is one of a machine learning model for turning left, a machine learning model for turning right, a machine learning model for stopping, a machine learning model for going forward, a machine learning model for making a U-turn, and a machine learning model for changing lanes that corresponds to the first control command.

2. The moving body control apparatus according to claim 1 , wherein

each of the N control command machine learning models is a neural network model.

3. The moving body control apparatus according to claim 1 , wherein

the control parameter includes at least one of a speed and a steering angle.

4. A training method for the moving body control apparatus according to claim 1 , the training method comprising:

a first step of training the feature amount extraction machine learning model and the N control command machine learning models using first training data that takes the N control commands for the moving body and a plurality of images of a view in a traveling direction of the moving body as inputs, and includes a plurality of control parameters for controlling the moving body as correct answers, the plurality of images respectively corresponding to the N control commands, and the plurality of control parameters respectively corresponding to the N control commands.

5. The training method according to claim 4 , further comprising:

a second step of training the feature amount extraction machine learning model using second training data that takes the plurality of images as inputs and includes the N control commands as correct answers,

wherein the second step is performed before the first step.

6. The training method according to claim 5 , further comprising:

a third step of training the feature amount extraction machine learning model and the N control command machine learning models using a constraint condition that restricts a possible operating state of the moving body,

wherein the third step is performed after the second step.

7. The training method according to claim 5 , further comprising:

a fourth step of training the feature amount extraction machine learning model and the N control command machine learning models using a traffic rule to be obeyed by the moving body,

wherein the fourth step is performed after the second step.

8. A moving body control method for controlling a moving body, the moving body control method comprising:

a first step of acquiring one control command among N control commands for the moving body and an image of a view in a traveling direction of the moving body, where N is an integer greater than or equal to 2;

a second step of using a feature amount extraction machine learning model to calculate, from the image acquired in the first step, a feature amount associated with the one control command acquired in the first step and indicating a feature of an object; and

a third step of selecting one control command machine learning model corresponding to the one control command from among N control command machine learning models respectively corresponding to the N control commands, and using the one control command machine learning model to output a control parameter for controlling the moving body, using the one control command acquired in the first step and the feature amount as inputs,

wherein each of the N control commands is a signal indicating an intention to maneuver the moving body,

the N control commands include a first control command which is one of turn left, turn right, stop, go forward, make a U-turn, and change lanes, and

the N control command machine learning models include a first control command machine learning model which is one of a machine learning model for turning left, a machine learning model for turning right, a machine learning model for stopping, a machine learning model for going forward, a machine learning model for making a U-turn, and a machine learning model for changing lanes that corresponds to the first control command.