Humanoid robot gait imitation

Abstract

This research proposed a learning model for stably walking of humanoid robots by imitation of human gaits. First, use Kinect to capture the skeleton information of human walking. Then extract key postures of a walking cycle from these abundant joints information, captured by Kinect. Q-Learning was used to learn stable walking efficiently, in order to conquer the lack of ankles information from these captured joints information and the walking differences between human and robots. This approach was implemented to both a simulated robot model and an actual humanoid robot. The results from the simulations show that the humanoid robots can efficient and stable walking by the imitation of human gaits.