Plucking Motions for Tea Harvesting Robots Using Probabilistic Movement Primitives

This letter proposes a harvesting robot capable of plucking tea leaves. In order to harvest high-quality tea, the robot is required to pluck the petiole of the leaf without cutting it using blades. To pluck the leaves, it is necessary to reproduce a complicated human hand motion of pulling while rotating. Furthermore, the rotation and pulling of the hand, and the time taken, vary greatly depending on conditions that include the maturity of the leaves, thickness of the petioles, and thickness and length of the branches. Therefore, it is necessary to determine the amount of translational and rotational movements, and the length of time of the motion, according to each situation. In this study, the complicated motion is reproduced by learning from demonstration. The condition is judged in terms of the stiffness of the branches, which is defined as the force received from the branches per unit length when the gripped leaf is slightly pulled up. Combining the learned motions probabilistically at a ratio determined by the branch stiffness, the appropriate motion is generated, even for situations where no motion is taught. We compared the motions generated by the proposed method with the motions taught by humans, and verified the effectiveness of the proposed method. It was confirmed by experiment that the proposed method can harvest high-quality tea.