On the Role of Models in Learning Control: Actor-Critic Iterative Learning Control

TL;DR

This paper introduces a model-free actor-critic iterative learning control framework that learns feedforward control signals directly from data, achieving model-based performance without explicit models, demonstrated on a printer system.

Contribution

Develops a novel model-free ACILC framework that encodes implicit model knowledge via basis functions and learns control parameters without explicit modeling.

Findings

Achieves comparable performance to model-based methods

Demonstrates effectiveness on a printer setup

Enables fast and safe learning without explicit models

Abstract

Learning from data of past tasks can substantially improve the accuracy of mechatronic systems. Often, for fast and safe learning a model of the system is required. The aim of this paper is to develop a model-free approach for fast and safe learning for mechatronic systems. The developed actor-critic iterative learning control (ACILC) framework uses a feedforward parameterization with basis functions. These basis functions encode implicit model knowledge and the actor-critic algorithm learns the feedforward parameters without explicitly using a model. Experimental results on a printer setup demonstrate that the developed ACILC framework is capable of achieving the same feedforward signal as preexisting model-based methods without using explicit model knowledge.