The Robot’s Motor Learning By Imitation Based On Human-computer Interaction

Robotics in the ongoing process of development, has been widely used in the aerospace, medical, education, and service industries, robots and human relations are getting closer. Intelligent robot is a continuous improvement in the level of knowledge of the outside world learning process, human-computer interaction techniques have been proposed since 1959 for human studies provide a growing number of robot technology support. This article achieve the gola of robot motor learning based on HCI.Content of this paper includes the following four parts:(1) Using motion interaction to achieve the goal of motor imitation. Getting the skeleton data and the depth image through the Kinect sensor, and recording the skeleton position, then sending these datas to the Nao robot after coordinate transformation, which realize the gola of robot motor imitation follow the teacher;(2) When the Nao robot imitate the motor, the data Nao received from the Kinect arises jitter because of the noise, which leads to the difference between the Nao robot and the teacher. The method named Kalman filter arised in this article to avoid this problem, which improve the effect of the motor imitation;(3) Due to the human body joint structure and the Nao robot joint structure difference, for example the head and the arm. The teacher can’t satisfy the imitation result which the Nao robot imitate through the motion interaction. We put forward the way named voice interaction to adjust the robot posture in the real time until the teacher satisfy. In the process of multiple voice interaction robot can learn to the higher matching degree, which can improve the robot’s motor imitation. To move forward a single step, we can make the robot reappear the motor by voice signal;(4)The robot build its motor library by learning contantly, to improve the motor imitation efficiency and avoid learning one motor repeatly, the robot must recognize the learned motion. In this article we arised the BP artifical neural network, which using skeleton data as trained data, and the network output the skeleton name. By comparing the basic action library, when the robot identify the action sequence have been studied, then learn new motor. Experiments show that the BP network’action recognition rate can reach more than 90%.