Adaptive Robot Control Using Artificial Neural Networks:An Application in the Theory of Cognition

During the last decade the problem of real-time robot control has proven to be of extreme difficulty. At present, available control systems are inadequate for the task. In addition, the application of sophisticated control schemes such as the Model-Reference Adaptive Control is prevented by the heavy computational task that is necessary to implement it. This paper offers a feasable solution for the real-time control of robot manipulators by adapting certain concepts of neural networks. An adaptive controller is presented which solves for the highly coupled dynamic equations of motion, which are known to present the heaviest obstacle in real-time computations. A symbolic representation of the Lagrange-Euler equations is adapted for this purpose. The neural controller is designed on a multi-layered network, in which the adaptation for environment changes could be accommodated via the back-propagation of errors throughout the network......