This paper describes the development of a hierarchical control system based on learning from demonstration applications. This system is capable of identifying and handling ambiguous situations that may have happened during the demonstration. An ambiguous situation is defined as a state of the robot in which the demonstrator executed different actions in spite of acquiring the same information from the sensors. The proposed solution uses a two stage unsupervised learning algorithm to generate the movement policies for the robot and it was tested using the Stage simulator on the Robot Operating System (ROS) for a Pioneer P3DX robot.