Unifying Model-Based and Neural Network Feedforward: Physics-Guided Neural Networks with Linear Autoregressive Dynamics

TL;DR

This paper introduces a unified feedforward control framework combining physics-based models and neural networks with shared linear autoregressive dynamics, enabling efficient compensation of unknown nonlinear dynamics.

Contribution

It proposes a novel parametrization that integrates physics models and neural networks sharing AR dynamics, improving interpretability and optimization in feedforward control.

Findings

Enhanced tracking performance with unknown nonlinear dynamics

Efficient output-error optimization via SK iterations

Neural network captures only unmodelled dynamics

Abstract

Unknown nonlinear dynamics often limit the tracking performance of feedforward control. The aim of this paper is to develop a feedforward control framework that can compensate these unknown nonlinear dynamics using universal function approximators. The feedforward controller is parametrized as a parallel combination of a physics-based model and a neural network, where both share the same linear autoregressive (AR) dynamics. This parametrization allows for efficient output-error optimization through Sanathanan-Koerner (SK) iterations. Within each SK-iteration, the output of the neural network is penalized in the subspace of the physics-based model through orthogonal projection-based regularization, such that the neural network captures only the unmodelled dynamics, resulting in interpretable models.