Data-driven Trajectory Tracking Of Robotic Arm With Event-triggered Model Updating

In recent years,robot industry has developed rapidly,which has been widely used in industrial production and social life.The control problem of manipulator system with strong nonlinearity has become a hot spot.Based on the nonlinear system,this paper studies the data-driven method of Gaussian process regression,the feedback linearization method of nonlinear system,the event triggering mechanism and the trajectory tracking control of manipulator.The main contents of this paper are as follows:(1)In view of the situation that the nonlinear system cannot accurately obtain its model,a data-driven method is introduced,and the system is identified by Gaussian process regression.For the problem of large calculation amount of Gaussian process regression under large data samples,sparse and online Gaussian process regression is introduced to carry out.Gaussian process regression is a non-parametric model.Its predicted output has probabilistic properties and can predict the uncertainty of the output.It is widely used in nonlinear problems.Therefore,the algorithm of induced point input is used to sparse the prediction phase of the Gaussian process,which greatly reduces the calculation of prediction.Secondly,for the training process,the sparse FITC algorithm and the more flexible PITC algorithm are studied to reduce the training calculation.Finally,for the actual application,the data is continuously accumulated online,and the sparse Gaussian process regression in the online situation is analyzed.(2)In order to identify the system when the sparse Gaussian process regression is used to identify the system and the system is of a composite structure,a combined kernel function is designed.That is,the system equation is a composite structure of two functions and control inputs,so when it is not possible to obtain separate measurements of two independent unknown functions,by analyzing the structure of the Gaussian process sum,a combined kernel function that can represent the structure of the system is designed,so that when only the sum of two unknown functions can be obtained,the estimates of the two separate functions can still be obtained.(3)Based on the Gaussian process regression identification to obtain the system parameters,the feedback control law is designed,and the feedback linearization is performed on the nonlinear system,so that the technology in the linear control theory can be used to control the actual nonlinear system.In response to the model changes caused by external environmental factors or system aging and other problems,resulting in unstable system operation,an event-based model update is proposed.Only when the accuracy of the estimated model is insufficient and the system stability is affected,The online update of the model,compared with the traditional time-triggered update,not only greatly reduces the amount of calculation,but also ensures the stability of the system.(4)For a complex manipulator system,on the basis of the aforementioned algorithm,after using Gaussian process regression for system identification,for the identification of the Gaussian process regression system of the multi-input and multi-output manipulator system,the basis of multiinput is naturally supported in the Gaussian process In the above,a multi-output Gaussian process considering the correlation between different outputs is introduced.At the same time,the eventtriggered feedback linearization is used to linearize the system,and the corresponding linear controller is designed to finally achieve the tracking control of the robot arm.