Abstract: For joints mapping between the human arm and the manipulator in human-machine natural interaction, a direct teaching scheme for 7-DOF (degree of freedom) humanoid manipulator in joint space is proposed. In the proposed approach, the Kinect sensor is adopted to acquire the position coordinates of the feature points of the human arm in Cartesian space, and an inverse kinematics method of the redundant 7-DOF human arm based on elbow constraints is investigated to obtain the joint angle data. To solve the trembling problem of humanoid manipulator caused by performance limits of Kinect for tracking the human arm motion, a filtering algorithm is proposed to smooth the joint motion trajectory of the manipulator. Finally, an experiment platform based on Motoman humanoid is set up and relevant experiments are conducted to verify the feasibility and effectiveness of the proposed approach.