Research On Trajectory Planning Of Robot Manipulator Based On IDMPs-AcaGMR

With the enormous economic value created by the robot industry,the strategic development targets of all countries have a place in the field of robotics.The rapid development of robot technology,especially the continuous breakthrough in the field of artificial intelligence,has driven a large number of industrial transformations and upgrades to the direction of intelligence.The trajectory planning of industrial manipulator is the basis of realizing the upgrading and transformation of automatic industrial production lines into an intelligent factory.At present,there are many problems in the motion planning of robot,such as weak learning ability,insufficient generalization ability,and insufficient flexibility in adjusting movement trajectory,etc.Based on this,we proposes a novel imitation learning method: research on trajectory planning of robot manipulator based on improved Dynamic Movement Primitives(iDMPs)– Active curve axis Gaussian Mixture Regression(AcaGMR).This method mainly consists of two parts: model optimization and algorithm optimization.Model optimization: Aiming at the shortage of DMPs,we change the nonlinear relationship between the phase variables and time to solve the problem that the feature extraction process in feature learning will extract the wrong information,optimize the Local Weighted algorithm and propose an improved Dynamic Motion Primitives(iDMPs);Algorithm optimization: Based on iDMPs,using Active curve axis Gaussian Mixture Regression(AcaGMR)instead of Local Weighted Regression.Experimental verify that this method has a better imitation learning effect.In this paper,the main research work are as follows:Firstly,we make a summary about the regression method,introduce base functions and kernel functions of machine learning in detail,and derive the basic regression expressions.Then we analysis Local Weighted algorithm,Gaussian Mixture algorithm,Active curve axis Gaussian Mixture algorithm respectively,and give a detailed analysis about internal workings of each algorithms.Finally,combine with regression method,giving the corresponding regression equation of each algorithm,and making some comparison about them.Secondly,in view of the deficiencies of the DMPs model,we make corresponding improvements and propose iDMPs.Specifically,under the premise of using Logistic function to guarantee the system stability,expression of phase variables is transformed to be evenly distributed in time,which has solved the problem that system is oversensitive to time,and further improve the local weighted algorithm fitting equation unreasonable problem.The results of the experiment show that iDMPs has a better imitation learning effect.Finally,we propose the research on trajectory planning of robot manipulator based on iDMPs-AcaGMR.It is achieved by using Active curve axis Gaussian Mixture algorithm in the forcing term of iDMPs,building an AcaGMM of forced component and phase variables,then we use AcaGMR algorithm to obtain the forced component fitting value,so as to obtain the corresponding movement trajectory.Finally,Matlab is used to carry out a multi-degree-of-freedom PM560 robot manipulator simulation experiment,and the results prove the superiority of iDMPs-AcaGMR.