Research On The Balance Control Of Motion Posture In Imitation Learning Of Humanoid Robot

Imitation is a kind of learning mechanism widely existed in human society,and has the characteristics of fast and efficient.By imitating the experience,people can learn the process of behavior quickly and master the skills.This is true with robots,especially for humanoid robot which has a rapid development in recent years,because of its similar structure design with human body,it has a good application prospect of makes the humanoid robot mimic human behavior to learn and get behavior skills.However,due to the redundant structure,high-degrees of freedom and the difference of mass distribution between human and humanoid robot,the robot's motion stability need to be ensured while imitation the learning task.This paper focuses on the problem of posture balance control in imitation learning of humanoid robot.The main work of this paper is:First,the static balance control of humanoid robots in posture imitation is realized.Due to the obvious difference of mass distribution between human and robot,the human's posture data acquired by the body sense collector need to be adjusted for stability before it can be used for driving robot,in this paper,a method of centroid prediction and compensation control for humanoid robot based on human's demonstration is proposed: estimate centroid in the process of robot imitation by human's demonstration data,combine centroid-angle Jacobi matrix to adjust the centroid offset in the imitation process,and use quadratic programming to optimize driving angle.The results of experiment base on imitation system of robot NAO show that the proposed COM prediction and compensation control algorithm based on human demonstration can effectively guarantee the posture balance of humanoid robot in posture imitation,the quadratic programming can effectively guarantee the robot's imitation posture similarity with human's demonstration.Second,the dynamic balance control of humanoid robots in gait imitation is realized.Based on static balance control of posture imitation,the gait imitation of robot is studied.Due to the high dynamic performance requirements of gait imitation,in this paper,a method of gait imitation stability control based on the ZMP criterion is proposed: planning expectation ZMP trajectory by human's gait,the difference between the actual ZMP and the expected ZMP is calculated by the information of the robots centroid trajectory,and compensate centroid according to the difference.Finally,the hysteresis curve is introduced to determine the switching of the supporting foot in the walking process.The results of experiment base on imitation system of robot NAO show that the ZMP criterion effectively guarantees the stability of robot in imitation process,and the support foot selection based on hysteresis curve ensures the smooth switching of supporting feet,Then the humanoid gait imitation of robots is realized,and the calculation of gait planning is simplified.Third,research of humanoid robot's posture balance generalization based on SVR: through human demonstration data to estimate the COM of the robot,use ZMP criterion and Jacobi matrix can ensure the stability of the imitation process.However,this method does not have the ability to generalize,that is,in the face of imitating tasks,it is necessary to solve the problem by analytic method.To give the balance generalization ability of the humanoid robot in the imitation of motion posture,this paper constructs the regression model of human demonstration data and robot driving angle based on SVR,using a training set which consist of human demonstration data and robot driving angles to training the regression model,and optimization parameters through particle swarm optimization,finally,using the regression ability of the model to get robot driving angle.The results of experiment base on imitation system of robot NAO show that human COM trajectory as the model input data,the regression angle can ensure the stability of imitation processing,it is shown that the posture balance regression model has a certain role.This paper studied the posture balance control of humanoid robot in imitation learning process,the proposed control method can make the humanoid robot to maintain stability in the process of learning to imitate human posture,it has a certain guiding significance for the application and promotion of the humanoid robot imitation learning,and the research on behavior control of humanoid robot.