Bilateral teleoperation is an efficient and powerful solution for conducting manipulation tasks through the robot in remote environments. However, performing repetitive manipulation tasks through bilateral teleoperation induces a heavy human workload. The typical repetitive and difficult task in a real teleoperation scenario is the rotational manipulation task. Therefore, we propose a framework to learn the skill of conducting rotational manipulation tasks from a single human demonstration through bilateral teleoperation. We have experienced that the existing Cartesian orientation-based trajectory learning method could not properly encode and reproduce the rotational trajectory. Therefore, a method that utilizes task parameters to encode the trajectory is applied to the framework. Moreover, the rotational manipulation task cannot be successfully performed without considering physical interaction, even if there exists only a very small estimation error in the goal pose. Thus, we suggest a method to learn and utilize physical interaction from the demonstration. The experimental result on simulation and real robot conducting vial capping task shows that the proposed framework can learn and reproduce human skill of performing rotational manipulation task even with estimation error.