This dissertation addresses the topic of intuitive human-robot interaction. In particular, this dissertation takes the use case of navigating robots with gestures for example, conducts a user study with 40 participants. A taxonomy of gestures for robot navigation is gathered from this user study. A dataset of 33 videos regarding using gestures to navigate robot is annotated. In addition, feeding DiffFrame, a novel way to feed the neural network, is proposed to increase the speed of reducing cost during training. Four distinct novel deep learning neural networks, which are proposed as DiffFrameNet, are designed to recognize human's gestures from the video.