Safety Constrained Stochastic Systems

Real-world systems are often subject to uncertainties and disturbances, which may stem from external influences or unmodelled dynamics. Such uncertainties can lead to an unsafe control behavior if not properly managed. Yet, robustifying the controller to all possible uncertainties can lead to a poor performance.

Fortunately, it is often not necessary to robustify to all possible realizations of uncertain events or parameters, as some may be unlikely. Allowing for a neglegible probability of safety violations can thereby drastically improve performance. Motivated by this fact, the main focus of my PhD is to explore controller designs optimized for performance under a pre-specified safety probability. I synthesize such controllers utilizing dynamic programming and convex optimization.

Precision Agriculture

Related to my theoretical work, I am interested in practical applications involving biological systems, such as medical instruments as well as precision agriculture. Not only do such systems have highly nonlinear dynamics, making them challenging to control. But they are also subject to biological variability, introducing uncertainties in the evolution of the system state.

I am currently working on the design and control of a hydroponics system, which is a soil-less cultivation method that allows for a precise control over plant growth conditions. The goal is to develop an optimal, learning-based control strategy that maximizes crop yield vs. resource usage (water, electricity). I further study the optimal fertilization of fields.

Sustainable and efficient use of resources in agriculture will be of ever rising importance in the face of climate change and increasing food demand. Further, the control of biological systems is challenging; While models are hard to derive and imprecise due to biological variability, data is scarce and expensive to acquire. This calls for alternatives to popular data-hungry learning based control approaches.