Accounts of using the Tustin-Net architecture on a rotary inverted pendulum

TL;DR

This paper evaluates the Tustin-Net neural architecture for modeling a rotary inverted pendulum, highlighting its advantages, limitations, and how transfer learning can improve its accuracy to match first-principles models.

Contribution

It introduces a transfer learning training strategy for Tustin-Nets, making them competitive with traditional physics-based models without extensive prior knowledge.

Findings

Tustin-Nets initially underperform compared to grey-box models.

Transfer learning significantly improves Tustin-Net accuracy.

Tustin-Nets with transfer learning can match first-principles models.

Abstract

In this report we investigate the use of the Tustin neural network architecture (Tustin-Net) for the identification of a physical rotary inverse pendulum. This physics-based architecture is of particular interest as it builds on the known relationship between velocities and positions. We here aim at discussing the advantages, limitations and performance of Tustin-Nets compared to first-principles grey-box models on a real physical apparatus, showing how, with a standard training procedure, the former can hardly achieve the same accuracy as the latter. To address this limitation, we present a training strategy based on transfer learning that yields Tustin-Nets that are competitive with the first-principles model, without requiring extensive knowledge of the setup as the latter.