Learning object manipulation from demonstration through vision for the 7-DOF barrett WAM

This paper presents an experimentation with a vision based skill learning technique using Barrett WAM (a 7-DoF robot manipulator) and Microsoft's Kinect (a 3D sensor). This skill learning approach is based on Symbolic Encoding. Even though most of the imitation learning approaches focus on trajectory encoding, symbolic encoding offers a more concise representation of a skill which is easily transferable to different forms of embodiments. The paper also presents a novel, fast and accurate hand-camera calibration technique along with a parametrized Inverse Kinematic solution for the redundant manipulator.