To automate design of modularizing and layering of a controller in an autonomous learning agent.

In the step S1, a prediction apparatus learning section obtains a differential equation corresponding to the prediction apparatus formed of functions with learning. In the step S2, a variable group creating section groups variables based on the obtained differential equation. In the step S3, an action variable determining section determines behavior variable corresponding to respective variable groups. In the step S4, a state variable determining section determines state variables corresponding to respective variable groups, and outputs them with the behavior variable to a layering unit. In the step S5, the layering unit creates controllers corresponding to combinations of the variable groups and behavior variable, and layers the created controllers. The present invention can be applied to the autonomous agent.