How to guarantee that the actor would select a correct action?

Question

In the training phase of Deep Deterministic Policy Gradient (DDPG) algorithm, the action selection would be simply

action = actor(state)

wher