Robust output regulation of linear system subject to modeled and unmodeled uncertainty

Zhicheng Zhang, Zhiqiang Zuo, Xiang Chen, Ying Tan, Yijing WangarXiv 2022 · DOI · arXiv

Abstract

This paper introduces a multi-loop control framework designed to maintain high-precision tracking in linear systems facing a "triple threat" of disturbances: stochastic white noise, modeled periodic oscillations, and unmodeled worst-case dynamics. By nesting an $H_\infty$ compensator and a Kalman filter within a traditional output regulation loop, the authors provide a unified architecture that mathematically bounds tracking errors. The framework is validated through high-fidelity simulations of a Furuta Inverted Pendulum, demonstrating significant improvements in stability and precision over existing state-of-the-art methods.

Introduction

In modern control applications, such as autonomous drones or precision robotics, systems are rarely subjected to a single type of interference. Instead, they face a complex environment characterized by modeled disturbances (e.g., periodic oscillations), unmodeled "worst-case" dynamics, and random measurement noise. While specialized techniques like $H_\infty$ control or Kalman filtering address these issues individually, combining them into a unified, non-conservative framework remains a significant challenge. This paper addresses this gap by proposing a multi-loop architecture that provides both robustness and performance guarantees without sacrificing nominal tracking precision.

The Complementary Control Architecture

The core of the proposed approach is the Complementary Control Structure, which decomposes the control task into three distinct, nested loops that operate in tandem:

  1. Output Regulation Loop: Employs the Internal Model Principle to track reference signals and cancel disturbances with known dynamics (modeled uncertainty).
  2. $H_\infty$ Compensation Loop: Based on Youla parameterization, this loop suppresses unmodeled dynamics. It is driven by a "residual signal"—the discrepancy between the observer's state prediction and actual sensor data.
  3. Kalman Filter Loop: Replaces standard state observers with optimal filters to handle stochastic white noise, ensuring reliable state estimation even in "fuzzy" environments.
The complete multi-loop Complementary Control Structure integrating output regulation, $H_\infty$ compensation, and optimal filtering.

The integration of these loops is achieved by solving a set of coupled Riccati equations, which define the optimal gains for each component while ensuring global stability and a defined "power norm" bound on tracking error.

Key Insights: Isolation via Residuals

A pivotal innovation in this framework is the Isolation via Residuals. Traditional $H_\infty$ controllers often operate on the raw tracking error, which can lead to overly conservative performance that degrades nominal tracking. In the proposed architecture, the $H_\infty$ loop specifically targets the residual $\epsilon_f$.

Basic structure of the $H_\infty$ compensation loop utilizing residual signals for disturbance rejection.

This allows the controller to react specifically to the "surprise" elements—discrepancies between the mathematical model and reality—without interfering with the primary tracking objectives. Furthermore, the authors derive a closed-form expression for the tracking error bound $\|e\|_P$, providing engineers with a quantitative "performance budget" to predict how noise and unmodeled disturbances will impact accuracy.

Experimental Results: The Furuta Pendulum Benchmark

The framework was evaluated using a high-fidelity simulation of a Furuta Inverted Pendulum, a classic benchmark for complex control dynamics. The proposed "Robust Output Regulation" approach was compared against a standard state-of-the-art method (Huang, 2004):

  • Stability: Under heavy unmodeled disturbance, the standard method became unstable, with the vertical arm exhibiting an overshoot of 520°. The proposed method maintained stability, keeping the deviation under 43°.
  • Precision: The proposed controller narrowed the horizontal arm's tracking error to less than 0.5° in steady-state, whereas the baseline method maintained an error exceeding 27°.
  • Noise Handling: The integration of the Kalman filter allowed the system to maintain this high precision even when sensors were subjected to significant white noise.

Conclusion

The proposed multi-loop framework offers a robust solution for linear systems operating in unpredictable real-world environments. By isolating unmodeled disturbances through residual-based compensation and integrating optimal filtering, it provides a clear path toward high-performance control with guaranteed safety margins. Future research will explore extending this theory to highly non-linear dynamics and multi-agent robotic swarms.

Summary updated

BibTeX
@misc{zhang2022robust,
  author = {Zhicheng Zhang and Zhiqiang Zuo and Xiang Chen and Ying Tan and Yijing Wang},
  title = {{Robust output regulation of linear system subject to modeled and unmodeled uncertainty}},
  year = {2022},
  eprint = {2210.15152},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2210.15152},
}
Figure