This ongoing postdoctoral research focuses on closing the loop between flood-state estimation and stormwater control. The core idea is that state uncertainty and control feasibility should be treated as a coupled systems problem rather than as two independent modules.

Key directions

• Developing uncertainty-aware flood-state reconstruction from sparse, heterogeneous, and failure-prone observations
• Designing projected diffusion policies that generate diverse control actions while enforcing hard actuator and safety constraints during sampling
• Building closed-loop stormwater digital twins that propagate uncertainty into decision-making rather than optimizing against a single best-guess state

Why it matters

Stormwater operations during extreme events require decisions in minutes, often under partial observability and strict physical constraints. This line of work is aimed at making real-time control more reliable, risk-aware, and operationally deployable in urban flood settings.