×

Regularization and robustness of learning-based control algorithms. (English. Russian original) Zbl 1092.93612

J. Comput. Syst. Sci. Int. 37, No. 2, 338-345 (1998); translation from Izv. Akad. Nauk, Teor. Sist. Upr. 1998, No. 2, 183-190 (1998).
Summary: A problem of learning-based control of cyclic operations applied to mechanical objects with weak damping is considered. It is demonstrated that the well-known conditions of convergence for the learning procedures require unrealizable operators. Weakened conditions of convergence are established. A method for regularization of the inverse operator is proposed. Algorithms of learning for the basic elastic inertial model and for the two-link robot manipulator are proposed and implemented. A problem of optimal choice of the learning operator is formulated.

MSC:

93E35 Stochastic learning and adaptive control
PDFBibTeX XMLCite