Applying reinforcement learning (RL) algorithms to control systems design remains a challenging task due to the potential unsafe exploration and the low sample efficiency. In this paper, we propose a ...