Bridging robustness and performance guarantees in control under uncertainty

Develop a unified control framework that reconciles robustness to latent parametric uncertainty with strong closed-loop performance guarantees for nonlinear dynamical systems, thereby bridging the gap between conservative robust control designs and high-performance control under uncertainty.

Background

The paper contrasts classical robust control—often formulated via worst-case distributionally robust optimization—with stochastic and risk-sensitive control that optimizes expected performance given a prescribed distribution. The authors argue that worst-case formulations are overly conservative and can degrade task performance, while task-agnostic stochastic approaches may fail to capture task-relevant uncertainty.

Motivated by this tension, the paper proposes a Stein variational inference-based approach to adapt uncertainty representations toward task-sensitive parameters. The explicit open problem highlights the need for a principled framework that simultaneously ensures robustness and strong performance guarantees, rather than prioritizing one at the expense of the other.