Policy iteration alternates evaluation of a fixed decision policy with improvement of its actions using the resulting value function. A stable policy satisfies the Bellman equation when each chosen action attains the required optimum.
New to topics? Read the docs here!