Neural network control for direct-drive robot mechanisms

The theoretical development of a trajectory-tracking neural network control ler based on the theory of continuous sliding-mode controllers is shown in the paper. Derived equations of the on-line adaptive neural network control ler were verified on a real industrial direct-drive 3 degrees of freedom (D .O.F.) PUMA mechanism. The new neural network continuous sliding-mode contr oller was successfully tested for trajectory-tracking control tasks with re spect to three criteria: convergence properties of the proposed control alg orithm (high-speed cyclic movement, low-speed movement, high-speed PTP move ment), adaptation capability of the algorithm to sudden changes in the mani pulator dynamics (load), and generalization properties of the proposed cont rol scheme. An interesting effect of the lower position error after a trans ient time at sudden load changes is shown. (C) 1998 Elsevier Science Ltd. A ll rights reserved.