Robust Data-Driven Receding-Horizon Control for LQR with Input Constraints

Jian Zheng, Mario SznaierIEEE Control. Syst. Lett 2025

TL;DR

This paper introduces a robust data-driven control framework for linear quadratic regulators (LQR) that explicitly handles input constraints. By leveraging all available execution data in a receding-horizon fashion and employing semidefinite programming (SDP) with duality, it achieves formal robust stability guarantees while reducing conservatism compared to existing methods.

What Problem Does This Paper Address?

Traditional control methods for systems with constraints often rely on accurate models, which can be costly to obtain and prone to errors. While data-driven control (DDC) offers an alternative by learning from data, many existing DDC approaches design controllers offline and apply them without adaptation, leading to conservative performance. Furthermore, handling input constraints in a data-driven, robust manner with formal stability guarantees has been a significant challenge. This paper aims to bridge this gap by developing a data-driven LQR controller that can adapt online, respect input limits, and provide provable stability.

Methods

The core of the proposed method lies in extending existing robust data-driven LQR frameworks, which are based on semidefinite programming (SDP), to incorporate input constraints. Key methodological contributions include:

  • Receding-Horizon (RH) Data Usage: Unlike prior DDC methods that use a fixed window of data, this approach continuously updates the controller using all available execution data. This allows the controller to adapt to system behavior over time and capture more information.
  • Semidefinite Programming (SDP) Formulation: The problem is formulated as an SDP, which allows for the efficient computation of optimal control policies. This is achieved by defining a "consistency set" that encapsulates all possible system dynamics consistent with the collected data.
  • Input Constraint Handling: The framework explicitly incorporates input constraints into the SDP formulation. This is done by introducing a scaling parameter that ensures the control input stays within its bounds.
  • Duality for Complexity Reduction: To address the computational complexity of solving the SDP with a growing number of data samples, the authors leverage duality. This significantly reduces the computational burden, making the approach more practical.
  • Formal Robust Stability Guarantees: By using SDP and carefully formulating the problem, the method provides formal guarantees of robust stability, meaning the system will remain stable even in the presence of unknown dynamics and bounded noise.

Datasets

The paper does not introduce new datasets. Instead, it relies on simulated data generated from a discrete-time linear system with specified parameters and noise. The experiments use "noisy measurements" and "experimental training data" to demonstrate the controller's performance.

Key Insights and Ideas

  • Online Adaptation is Crucial for Performance: By continuously updating the controller with new data (receding-horizon approach), the framework can achieve less conservative performance than methods that rely on a fixed set of initial data.
  • Duality is a Powerful Tool for DDC Complexity: The application of duality significantly reduces the computational complexity of the data-driven SDP, making it feasible for practical applications where data accumulates over time.
  • Formal Guarantees are Achievable: The paper demonstrates that it's possible to achieve formal robust stability guarantees in a data-driven setting, even with input constraints, by carefully leveraging SDP and set-membership approaches.
  • Shrinking Consistency Sets Improve Bounds: As more data is collected, the consistency set (representing possible system dynamics) shrinks. This tighter representation leads to a reduced worst-case LQR cost bound.

Key Results

  • The proposed robust data-driven receding-horizon LQR controller successfully enforces input constraints, as demonstrated in simulation.
  • The method provides formal robust stability guarantees for the closed-loop system.
  • Updating the consistency set with execution data leads to a decrease in the worst-case upper bound of the LQR cost over time.
  • The use of duality significantly reduces computational complexity compared to naive SDP relaxations.

Limitations and Future Work

  • Line Search for Scaling Parameter: The paper acknowledges that the line search for the scaling parameter $\tau$ is a limitation, although it can be efficiently implemented.
  • Multi-input Systems and General Control Bounds: The current framework is primarily demonstrated for single-input systems with an $l_\infty$ control bound. Extending it to multi-input systems with more general control bounds is a direction for future work.
  • State Constraints: Incorporating state constraints into the framework is another area for future research.
  • Efficient Computation of $\tau$: Developing more efficient methods for computing the scaling parameter $\tau$ is also suggested.

Summary updated

Abstract

This letter presents a robust data-driven receding-horizon control framework for the discrete time linear quadratic regulator (LQR) with input constraints. Unlike existing data-driven approaches that design a controller from initial data and apply it unchanged throughout the trajectory, our method exploits all available execution data in a receding-horizon manner, thereby capturing additional information about the unknown system and enabling less conservative performance. Prior data-driven LQR and model predictive control methods largely rely on Willem's fundamental lemma, which requires noise-free data, or use regularization to address disturbances, offering only practical stability guarantees. In contrast, the proposed approach extends semidefinite program formulations for the data-driven LQR to incorporate input constraints and leverages duality to provide formal robust stability guarantees. Simulation results demonstrate the effectiveness of the method.
BibTeX
@article{zheng2025robust,
  author = {Jian Zheng and Mario Sznaier},
  title = {{Robust Data-Driven Receding-Horizon Control for LQR with Input Constraints}},
  journal = {IEEE Control. Syst. Lett.},
  year = {2025},
}
Figure