Robot Manipulator Control And Virtual Realizing Based On Deterministic Learning

With the wide application of the robot manipulators in the manufacturing industry, theresearch of robot manipulator control has received more and more attention. Kinds of NNcontrol methods of robot manipulators have been proposed. These methods have obtainedgood control performance in the respect of adaptation, but they do not have the ability oflearning. In this case, these adaptive neural controllers have to recalculate the controlparameters even for repeating the same control task.Recently, a deterministic learning theory is presented. The theory provides systematicdesign approaches for knowledge acquisition, representation, and utilization in uncertaindynamical environments. An NN controller based on deterministic learning theory canimplement true learning ability, and the learned knowledge is saved in a way of constant RBFneural networks. The knowledge obtained can be utilized in another similar control task.This paper designs a direct adaptive NN controller based on deterministic learning forrobot manipulators control and combines ADAMS (Automatic Dynamic analysis ofMechanical System) and MATLAB/Simulink to realize a virtual simulation for theclosed-loop control system, which makes full use of the power modeling ability of ADAMSand the great control function of MATLAB/Simulink. Compared with traditional MATLABsimulation, virtual simulation can show the control performance of robot manipulators morevisually and verify the control algorithm designed in this paper is effective.The virtual simulation system includes three main parts: controller, tip referencetrajectory of robot manipulators, and virtual prototype of robot manipulators. The controllerand the tip reference trajectory of robot manipulators both are realized by S-function inSimulink. The entire algorithm is encapsulated in the independent functional module whichnot also simplifies the simulation framework but also facilitates the adjustment of parameters.The virtual prototype of robot manipulators is created and exported in ADAMS and then thecorresponding module of the virtual prototype is generated in MATLAB. In order to realizethe virtual simulation successfully, the problem of simulation speed must be solved. As thenumber of neurons included in the controller based on deterministic learning will increaseexponentially with the increasing complexity of the controlled object which needs massivecircle calculation, the simulation speed becomes extremely slow. In order to solve thisproblem, this paper adopts C-MEX technique and realizes the loop part using C program.The virtual simulation results not only visually display the process of robot manipulatorcontrol, but also show the controller designed in this paper has good control peroformance and learning ability which provides a new approach to the robot manipulator control.