Design and analysis of a learning control system

A learning control scheme is presented. The control action in this scheme is computed by referring to a decision table which is on-line modified, rather than by mathematically analyzing the dynamics of the control problems and then solving some equations. The controller combines the discretised input commands (if any) and feedback variables into a vector to address a memory where the appropriate output variables are stored. The learning algorithm is designed using the idea of sliding mode. Convergence conditions of a learning algorithm are given. The quantisation effects of measurement variables on convergence of learning are analysed