This PhD focuses on the development of distributionally robust control algorithms by building ambiguity sets that capture local distributional variations of the uncertainty.
Position description
Real-world systems are challenged by constantly increasing complexity and uncertainty. This uncertainty is often unknown and dynamically varying and may need to be described locally with sufficient detail. Traditionally, decisions are taken by assuming a given probability distribution of the uncertainty, which may be rather arbitrary and deviate significantly from the real model.
The goal of this PhD project is to build ambiguity sets of probability distributions that hedge against plausible variations of stochastic uncertainty models and optimize their geometry to prevent their potential conservativeness. Special emphasis will be devoted to data-driven formulations and on how to infer the unknown uncertainty models across sub-regions of interest while retaining formal statistical guarantees. The developed methods will be exploited to design efficient control algorithms for the safe deployment of autonomous systems in uncertain environments.
Application & additional information
More information about this position and the application procedure can be found at this link.
For any questions, please contact Dimitris Boskos (d.boskos@tudelft.nl).
