Progressive Automation of Periodic Movements

This paper presents the extension of the progressive automation framework for periodic movements, where an operator kinesthetically demonstrates a movement and the robotic manipulator progressively takes the lead until it is able to execute the task autonomously. The basic frequency of the periodic movement in the operational space is determined using adaptive frequency oscillators with Fourier approximation. The multi-dimensionality issue of the demonstrated movement is handled by using a common canonical system and the attractor landscape is learned online with periodic Dynamic Movement Primitives. Based on the robot’s tracking error and the operator’s applied force, we continuously adjust the adaptation rate of the frequency and the waveform learning during the demonstration, as well as the target stiffness of the robot, while progressive automation is achieved. In this way, we enable the operator to intervene and demonstrate either small modifications or entirely new tasks and seamless transition between guided and autonomous operation of the robot, without distinguishing among a learning and a reproduction phase. The proposed method is verified experimentally with an operator demonstrating periodic tasks in the free-space and in contact with the environment for wiping a surface.