Research On Dynamic Parameter Identification And Teaching Learning Of Collaborative Robots Based On GMM

With the promotion of policies such as ’ Made in China 2025 ’ and ’ 14 th Five-Year ’ robot industry development planning,intelligent flexible cooperative robots have become a research hotspot.As a new type of industrial robot,collaborative robots can work safely with humans in a shared space,improving work efficiency and quality.Dynamic model parameter identification,drag teaching and learning are important research directions of cooperative robots.At present,dragging teaching mostly depends on external sensors,and the effect of sensorless dragging teaching depends on the accuracy of dynamic model.However,due to the complexity and uncertainty of robot system,the accuracy of dynamic model parameter identification is difficult to guarantee.At the same time,the traditional teaching process requires professionals to use the teaching instrument for programming,which is inefficient and not adaptive.Aiming at these problems,this paper studies the parameter identification of the dynamic model of the cooperative robot,the drag teaching without external sensors and the teaching learning with adaptive ability.The details are as follows:Taking the six-degree-of-freedom cooperative robot as the research object,the kinematics model and dynamic model of the robot are established.The Denavit-Hartenberg(DH)method is used to derive the robot kinematics and Jacobian matrix.The Newton-Euler method is used to model the dynamics,and the friction force is modeled by improving the Coulomb-viscous friction model.The established dynamic model is linearized and the minimum identifiable parameter set is obtained by QR(Orthogonal)decomposition.In order to fully stimulate the dynamic characteristics of the robot in the parameter identification experiment,this paper uses the finite Fourier series to design the trajectory,and takes the block regression matrix as the optimization objective.Considering the joint limit and start-stop limit as the optimization constraints,the iterative least squares method is used to optimize the trajectory parameters to obtain the excitation trajectory that better stimulates the dynamic characteristics.Taking the condition number of the regression matrix as the criterion of the excitation trajectory,it is proved that the proposed method can stimulate the dynamic characteristics more effectively,and the feasibility of the excitation trajectory is verified on the experimental platform.Aiming at the problem of low accuracy of dynamic parameter identification,this paper proposes a dynamic model identification and compensation method based on three-iteration identification.The iterative weighted least squares method is used to identify the parameters of the basic dynamic model and the improved friction model,and the residual torque is compensated by the Gaussian Mixture Model(GMM).The root mean square of the actual measured torque and the model estimated torque residual is used as the criterion for dynamic parameter identification,and the accuracy of the identification results is proved by the experimental platform.In this paper,the disturbance observer and admittance control are combined to realize the drag teaching without external sensors,and the dynamic movement primitives(DMP)algorithm is improved to improve the robustness of teaching learning.Firstly,the basic dynamic model is decoupled to obtain the standard robot dynamic model.The external force of the joint is estimated by the disturbance observer based on momentum,and the zero force teaching of the cooperative robot is realized by combining the admittance control technology.After that,through manual guidance,multiple teaching trajectories of the same task are obtained,and DTW is used to align the trajectories in the time dimension.Through DMP-GMM,multiple teaching trajectories of the same task are modeled,features are extracted and task trajectories are formed,and task trajectories are learned and generalized.Finally,the learning and generalization ability of DMP-GMM method for multiple teaching trajectories is verified by Python,and the feasibility of dragging teaching and teaching learning is verified on the experimental platform.