Humanoid robot locomotion control by posture recognition for human-robot interaction

During human-robot interaction, it is often necessary for human users to command the robot´s locomotion. In this paper, we design an intuitive interaction mechanism for users to command a NAO robot´s locomotion with predefined postures. Based on images taken by its onboard monocular RGB camera, the robot can localize the human user´s head, torso and arms, and recognize the posture displayed by the user with the k-NN algorithm. Then the robot can move according to the locomotion command (forward/backward/left/right) associated with the posture. We test these postures in an HRI experiment and get good recognition performance.