Model-free Predictive Optimal Iterative Learning Control using Reinforcement Learning

Iterative learning control (ILC) is a high-performance control design method for systems working in repetitive manner and has seen many applications in practice. Predictive optimal ILC, a well-known design algorithm, updates the input for the next trial by optimising a performance index defined over (predicted) future trials and has many appealing convergence properties, e.g. monotonic error norm convergence guarantee. This is achieved, however, using a system model which can be difficult or expensive to obtain in practice. To address this problem, this paper develops a model-free predictive optimal ILC algorithm using recent developments in reinforcement learning. The algorithm can learn the predictive optimal ILC controller without using any system model. We provide a rigorous convergence proof of the developed algorithm which is generally not trivial for reinforcement learning based control design. A numerical example is presented to demonstrate the effectiveness of the proposed algorithm.