Guided Multi-Fidelity Bayesian Optimization for Data-driven Controller Tuning with Digital Twins

TL;DR

This work tackles data-efficient controller tuning under model mismatch by proposing Guided Multi-Fidelity Bayesian Optimization (GMFBO), which integrates real-system data with corrected digital twin simulations. A correction GP refines DT outputs, an adaptive multi-source kernel modulates cross-source correlations according to estimated DT accuracy, and a cost-aware EI acquisition balances fidelity, improvement, and sampling costs. The framework actively updates as new measurements arrive, enabling rapid adaptation to changing dynamics and reducing reliance on costly real-system evaluations. Empirical results on robotic drive hardware and numerical studies show that GMFBO accelerates convergence and improves data efficiency over standard BO and baseline MFBO methods, even in non-stationary scenarios.

Abstract

We propose a \textit{guided multi-fidelity Bayesian optimization} framework for data-efficient controller tuning that integrates corrected digital twin simulations with real-world measurements. The method targets closed-loop systems with limited-fidelity simulations or inexpensive approximations. To address model mismatch, we build a multi-fidelity surrogate with a learned correction model that refines digital twin estimates using real data. An adaptive cost-aware acquisition function balances expected improvement, fidelity, and sampling cost. Our method ensures adaptability as new measurements arrive. The digital twin accuracy is re-estimated, dynamically adapting both cross-source correlations and the acquisition function. This ensures that accurate simulations are used more frequently, while inaccurate simulation data are appropriately downweighted. Experiments on robotic drive hardware and supporting numerical studies demonstrate that our method enhances tuning efficiency compared to standard Bayesian optimization and multi-fidelity methods.

Figures (11)

• Figure 1: Efficient MFBO controller tuning using multiple information sources
• Figure 2: Guided multi-fidelity Bayesian optimization for tuning controller parameters
• Figure 3: Block diagrams of information sources
• Figure 4: The top panel shows the true objective function given IS1; The bottom panel shows the relative absolute error (in %) of the estimated objective by IS2
• Figure 5: Sampling cost over BO iterations, averaged over $50$ Monte Carlo experiments. Shaded regions indicate $95\%$ confidence intervals.