Characterizing Input Methods for Human-to-robot Demonstrations