Adaptive Learning of Hand Movement in Human Demonstration for Robot Action

This paper describes a process for adaptive learning of hand movements in human demonstration for manipulation actions by robots using Dynamic Movement Primitives (DMPs) framework. The process includes 1) tracking hand movement from human demonstration, 2) segmenting hand movement, 3) adaptive learning with DMPs framework. We implement a extended DMPs model with a modified formulation for hand movement data observed from human demonstration including hand 3D position, orientation and fingers distance. We evaluate the generated movements by DMPs model which is reproduced without changes or adapted to change of goal of the movement. The adapted movement data is used to control a robot arm by spatial position and orientation of its end-effector with a parallel gripper.