Research On Imitation Of NAO Robot Based On Kinect Sensor

Robot imitation is one of the hot topics in the field of human-computer interaction.It is of great value to find out how to improve the level of artificial intelligence of robot and how to adapt to the environment around.With the development of society and the refinement of social division of labor,it becomes difficult to control the robot to accomplish high precision and complicated content in different environments.And the interactive way of imitation provides a potential solution to this problem.In this paper,the imitation algorithm based on NAO robot is studied.In this paper,the NAO robot is used as the experimental platform to realize the imitation of human motion based on the Kinect sensor.Then optimizes the imitation action aiming at improving the gliding property and time-space consistency.After that,taking advantage of the accuracy of the human skeleton from Kinect detection,the face recognition algorithm is combined with imitation to realize the imitation of the specified person.Finally,the BP neural network is used to predict the steady state of the imitation action,so as to avoid the falling of the robot during the experiment.Specific research contents are as follows:First of all,connect the Kinect and NAO robot SDK with the computer to build the experimental platform.An improved inverse kinematics formula is used to realize the imitation of the arm action on human.Secondly,based on the imitation,this paper optimizes the action.Moving average filter and limit breadth filter are used to make the robot movement smooth,reduce the sudden change of joint angle leading to motor damage.Solutions called speed proportion are used to ensure velocity changes in real time,so as to keep the time-space consistency.By storing a series of human joints data,the robot can reproduce the human action at any time.The simulation algorithm and face recognition are combined to realize the imitation of a specified person.Finally,we control the leg joints on the basis of arm imitation.Due to the difference of the human and the robot structure,the joint angle corresponding to the human standing posture is set as the initial attitude.Considering robot may have instability problems because of gravity caused leg joint control.In order to avoid the robot fall due to instability,we puts forward a method of BP neural network to identify the stable state which will be the acted next time.Experiments show that by filtering and speed control,there is a good inhibitory effect to the angle mutation caused by the structure difference and noise,and ensure the time-space consistency of imitation at the same time.Meanwhile,finding the location of the face by the judgment to human skeletal framework from Kinect,the computer can effectively avoid the effects of rotation and illumination,achieving a specified person imitation;And after applying Back-Propagation neural network to whole-body imitation,there is an effective avoidance of the unstable action.