Moving body behavior prediction device

1. A moving body behavior prediction device that predicts a behavior of a moving body, the moving body behavior prediction device comprising:

a prediction model evaluation value storage that stores an evaluation value for each travel environment in relation to each of prediction models prepared in advance; and

at least one controller configured to

acquire external information and recognize a travel environment;

determine the prediction model corresponding to the travel environment recognized from among the prediction models, the determination being performed based on the travel environment recognized and the evaluation value stored in the prediction model evaluation value storage;

predict the behavior of the moving body using the prediction model determined; and

a safety model storage that stores at least one or more prediction models estimated to be safe for each of travel environments as safety models,

wherein the at least one controller is configured to:

determine one prediction model corresponding to the travel environment based on the evaluation value stored in the prediction model evaluation value storage according to a predetermined condition set in advance, or select one prediction model corresponding to the travel environment from among the prediction models stored as the safety models in the safety model storage, wherein

the predetermined condition is a case where it is possible to estimate that a predetermined safe state is maintained,

the at least one controller is configured to, when it is possible to estimate that the predetermined safe state is maintained, determine one prediction model corresponding to the travel environment, from among the prediction models stored as the safety models in the safety model storage, and

the at least one controller is configured to, when it is difficult to estimate that the predetermined safe state is maintained, select one prediction model having a highest evaluation value stored in the prediction model evaluation value storage from among prediction models corresponding to the travel environment,

the at least one controller is configured to, when it is possible to estimate that the predetermined safe state is maintained, randomly determine one prediction model corresponding to the travel environment from among the prediction models stored as the safety models in the safety model storage, and

the at least one controller is configured to control the behavior of the moving body based at least in part on the predicted behavior of the moving body.

2. The moving body behavior prediction device according to claim 1 ,

wherein the at least one controller is configured to:

evaluate the evaluation value for each travel environment stored in the prediction model evaluation value storage, and

update the evaluation value stored in the prediction model evaluation value storage according to a prediction result of the at least one controller.

3. The moving body behavior prediction device according to claim 2 , wherein the at least one controller is configured to:

manage the evaluation value stored in the prediction model evaluation value storage,

transmit the evaluation value for each travel environment in relation to each of the prediction models stored in the prediction model evaluation value storage to a management server, and

update the evaluation value stored in the prediction model evaluation value storage with an evaluation value received from the management server.

4. The moving body behavior prediction device according to claim 3 , further comprising:

a reference evaluation value storage that stores the evaluation value received by the at least one controller from the management server as a reference evaluation value; and

a safety model storage that stores at least one or more prediction models estimated to be safe for each of travel environments as safety models,

wherein the at least one controller is configured to, when it is possible to estimate that a predetermined safe state is maintained, determine one prediction model corresponding to the travel environment, from among the prediction models stored as the safety models in the safety model storage, and when it is difficult to estimate that the predetermined safe state is maintained, select one prediction model having a highest reference evaluation value among prediction models corresponding to the travel environment.

5. The moving body behavior prediction device according to claim 2 , wherein

the at least one controller is configured to update an evaluation value of the prediction model so as to lower the evaluation value when a predetermined situation set in advance occurs and to raise the evaluation value when the predetermined situation does not occur in accordance with the prediction result.

6. The moving body behavior prediction device according to claim 1 , wherein the at least one controller is configured to:

evaluate whether the prediction model stored as the safety model in the safety model storage is safe,

calculate a behavior in a case of inputting past external information to each of the prediction models, and

evaluate whether each of the prediction models is safe based on a result of the calculation.