Application of Neural Nets for Control of Robotic Manipulators

Abstract

Fine motion control of robotic manipulators has become a desired goal in the last few years as a result of new robot morphology and the definition of new tasks involving high velocity motions and end effector tracking precision. In order to achieve better performance of robotic manipulators the artificial intelligence can be introduced into the control system. One way for accomplishing this is application of neural nets. Main advantage claimed for neural based controllers is their ability to learn and generalize from partial data and ability to perform parall processing. Many papers present possibilities of neural nets application for control [Hunt,92],[Sontag 93],[Anstaklis,92]. They have discussed the use in adaptive control, predictive control, optimal control, gain sheduling in PID with variable gain. A large area of neural nets application for control is control of nonlinear systems. Neural nets can be applied as a forward or inverse model of manipulator. Such neural net can learn a behaviour of the real system, without knowledge about robot’s model structure.