The world is not
a black box.

Modern data-driven methods are remarkable approximators. Yet the systems we most want to control — fusion reactors, lithography stages, the human brain, large-scale energy grids — are governed by partial differential equations, by conservation laws, by symmetries that data alone cannot reveal. Treating them as black boxes is not just inefficient. For safety-critical applications, it is unacceptable.

"How can we combine known invariants with real-world data to guarantee the best performance of physical systems?"

My answer is to build a control science that respects the continuum nature of distributed physical systems. Algorithms that exploit passivity, symmetry, and dispersion relations as structural priors rather than rediscovering them from scratch. Methods that come with quantifiable guarantees on stability, performance, and uncertainty.

This is the agenda of spatio-temporal control: algorithms that are discretization-independent, that scale with the physics rather than fight against it, and that meet the demands of high-tech systems where failure is not an option. My research advances this vision through four interlocking themes: