Design and Implementation of Humanoid Robot Behavior Imitation System Based on Skeleton Tracking

Abstract

This paper builds and implements a robot humanoid motion imitation system, which consists of three parts. Firstly, a Microsoft Kinect 2.0 somatosensory camera is applied to capture human skeleton data. Secondly, using the EWMA (exponentially weighted moving averages, EWMA), we can smooth the acquired skeleton data considering the local and entire process without prior knowledge of noise. And the Holt-Winters second-order EWMA is employed to track the original signal slope and incorporate its trend into the smoothed data. Then, the smoothed data are mapped to the robot joint angles by geometric calculations. Finally, a NAO robot implements motion imitation based on the joint angles. Experimental results demonstrate that the NAO robot can imitate the action of the instructor smoothly and steadily using the proposed motion imitation system.