Data-driven stabilization and safe control of nonlinear systems

The recent successes of machine learning solutions have inspired the research of new control algorithms derived directly from the available data without any intermediate step. Being able to design a stabilizing controller directly from data has the main advantage that, since it does not rely on a model of the system to control, the controller design is not influenced by any modeling error.
Most of the time real systems are simplified with linear models to reduce the overall complexity in the controller design discarding all the complex nonlinear behaviors. A linear approximation could be an excessive simplification for complex system where the presence of nonlinear dynamics are important to understand those processes and nonlinearities can not be ignored. However, the analysis and control of a nonlinear model is often challenging. This thesis investigates data-based control methods for continuous and discrete-time nonlinear systems that do not require to model the system. In particular, we have developed a solution to obtain a stabilizing state feedback controller for the case of nonlinear systems.
Stabilizing a closed-loop system is critical, but sometimes it is not enough. Safety is another important criteria considered in the design of a controller. We were able to formulate a new data-driven procedure to find a stabilizing controller that can also guarantee that the state of the system never violates the safety constraints.
For all the solutions presented, we discussed how to handle real noisy measurements.