A Novel Reinforcement Learning Architecture for Continuous State and Action Spaces

Figure 3

12 function approximators used to implement the action type actor, the parameter actor, and the critic.