A new way to control systems using noisy data
Researchers developed a method to create stable controllers directly from measurements without needing to identify system dynamics first.
A new probabilistic framework creates stochastic linear-quadratic output-feedback control for linear systems. This method works even when there is process and measurement noise.
The approach uses delayed input-output measurements. It uses a lifted parameterization and a lifted linear quadratic cost to synthesize the controller without explicitly identifying the lifted dynamics.
Two closed-loop formulations are used based on trajectory and covariance representations. These formulations have uncertainties that depend only on the realized but unknown lifted noise.
Gaussian zonotopes refine uncertainty sets by removing noise realizations inconsistent with measured data. A novel intersection rule fuses data-driven and physics-informed information.
This adaptive framework unifies indirect identification-based and direct control-oriented data-driven linear-quadratic output-feedback control. It achieves robust stability and superior performance under noisy measurements.
Key facts
- The method achieves probabilistic mean-square stability (p-MSS) and performance guarantees.
- The approach uses Gaussian zonotopes to reduce conservatism.
- The controller approaches the ideal optimal solution in simulations.