Defence Lidong Li: "Data-Driven Controller Learning in Noisy Environments"

Promotors: Prof. Nima Monshizadeh and Prof. Claudio De Persis

Abstract: The world is generating more and more data, leading to breakthroughs in many fields, including control engineering. Traditionally, controller design requires mathematical models, which can be expensive and time consuming. Instead, new methods allow controllers to be learned directly from data while ensuring robust guarantees. However, these methods struggle when there isn’t enough data or when the data is noisy. This thesis tackles these challenges: (1) Learning from Limited Data: When there is not enough data for a system, we use knowledge from other systems to help design the controller for the original system. (2) Dealing with Noisy Data: Real-world data often contains errors. These errors lead to a set of possible systems that are consistent with the data. Since the actual system is indistinguishable from all others in the set, our method aims to asymptotically stabilize all systems within the set.

Dissertation