A real-time human imitation system

Abstract

Robot imitation is a useful tool in humanoid robot research, it provides a natural way of teaching a complex humanoid robot to accomplish human-like behaviors. However, it is never an easy task to build such a system robustly due to the high degree of freedom (DOF) in human motion and difficulty in humanoid control. In this paper, we present a system built up by the Microsoft Kinect box and the Aldebaran NAO Humanoid robot, which can mimic full body human motion in real time. By applying inverse kinematics through an optimization process, the human motion was split up into critical frames and represented by a list of robot joint angles. Then a second order continuous trajectory is formed to drive each joint toward the target angle, followed by balance control and self-collision avoidance system. The tests show that the system is robust and flexible enough to imitate various human motions.