Experience based acquisition of the initial value for the iterative learning control inputs

Abstract

The effects of the initial value of iterative learning control inputs to the convergence and stability are discussed. An idea of acquiring these values based on the systems control experience is presented. At the same time, three kinds of methods of getting these values--linear weighted average, curve fitting and intelligent methods, are given. Simulation results of a robotic system show that the robotic system can track the new task with less error at the first trial if the initial control input is chosen properly. So the trials of learning is decreasing, and the convergence and capacity of adapting to the new task and new environment are improved.