抄録
We developed a robotic hand that folds an origami form “Tadpole”. However, the robot, which simply replays a given trajectory, often fails in folding due to the fluctuation of origami paper behavior. In this paper, we propose a novel method to synthesize a desired trajectory and sensory feedback control laws for robots based on the statistical feature of direct teaching data demonstrated by a human. Hidden Markov Model (HMM) is used to model the statistical feature of human motion. Nominal desired trajectory is obtained by temporally normalizing and spatially averaging the teaching data. Sensory feedback control laws are then synthesized based on the output probability density function parameters of the HMM. Considering velocity variance and canonical correlation between velocity and force of the teaching data, important motion segments are extracted and feedback control laws are applied only for those segments. Experimental results showed that the success rate and folding quality of “Valley-fold” were improved by the proposed method. The proposed method enables robot motion teachers to simply perform direct teaching several times to transfer their skill, which is difficult to describe explicitly, to the robot.
