Interactive robotics is a vast and expanding research field. Interactions must be sufficiently natural, with robots having socially acceptable behavior for humans, adaptable to user expectations, thus allowing easy integration in our daily lives in various fields (science, industry, domestic, health ...). In this context, we will achieve a system that involves the interaction between the human and the NAO robot. This system is based on gesture recognition via Kinect sensor. We choose the Hidden Markov Model (HMM) to recognize four gestures (move forward, move back, turn, and stop) in order to teleoperate the NAO robot. To improve recognition rate, data are extracted with Kinect depth camera under ROS, which provides a node that tracks human skeleton. We tried to choose a feature vector as relevant as possible to be the input of the HMM. We performed 3 different experiments with two types of features extracted from human skeleton. Experimental results indicates that the average recognition accuracy is near 100%.