I consider several problems related to the analysis and control of cyber-physical systems from data. This is relevant in many applications where we do not have an accurate model of the system. To address this, I study the synthesis of safety certificates (such as Lyapunov functions and barrier functions) and safe controllers (such as feedback controller and MPC) from data with probabilistic guarantees of correctness.
I consider the problem of certifying the safety and correctness of nonlinear and hybrid systems, which is crucial for safety-critical applications (such as medical devices and autonomous vehicles). For that, I leverage powerful tools from Optimization, Computer Science and Machine Learning, such as Sum-of-Squares Optimization, Counterexample-Guided Inductive Synthesis (CeGIS), Neural Networks and SMT solvers.
I address the problem of learning models for cyber-physical systems. This is known to be a very challenging computational problem, but very important as many cyber-physical systems used in important applications are difficult to model from first-order principles (e.g., energy grids, biological systems, etc.).