The conception of quantum computation is raised almost one hundred ago and until several decades, after the experimental techniques have breakthroughs in the nearly ending of last century. Yet, there are still full of issues and topics in various directions of quantum computation researches, including both hardware and software fields. Among them, quantum control problem is one of the essential problems because of the fragile nature of quantum systems. In this thesis I will introduce the reinforcement learning methods to deal with quantum control problems, including state preparation in quantum metrology and smooth quantum gates in superconducting qubit systems. Essential fundamentals of the reinforcement learning methods are introduced, including concepts of basic objects in this area, as well as the policy-based algorithms. Ch. 3 has a fully introduction about the state preparation in parameter estimation (PE) topic, as well as the demonstration of the optimal control with RL method. Two kinds of noisy process, including the dephasing dynamics and spontaneous emission, are considered. The generalisability of the RL method is also discussed under biased initial parameter guess. Ch. 4 demonstrates the control problem in circuit quantum electrodynamics (cQED) systems, followed by the normal protocol of quantum gates. A new protocol of quantum control is introduced to deal with the arbitrary ZZ interaction issue. The protocol defined control pulse is parametrised and optimised with RL method. In the end of the thesis I discussed some current challenges and the potentially future works in the combination of quantum control and reinforcement learning methods.