Over the past decades, model-based approaches have been the dominant paradigm in rigorous analysis and systematic control design, which is reflected in numerous applications ranging from manufacturing to intelligent robots. A typical model-based control design is often relied on manually generated models that are based on human insights into physics. However, as modern engineering systems become increasingly sophisticated, such classical control designs could not catch up with the rapid development and have presented a number of difficulties regarding to the scalability of the models as well as the overall control systems. To address these challenges, researchers are actively looking for more integrated and automated control approaches which can efficiently and directly learn from data. In light of these initiatives, the research proposal launches a series of fundamental studies aiming to facilitate efficient and valuable integrations of data as well as powerful computation into diverse aspects of system analysis and control design. In particular, I will present in the first part of the proposal a vertically integrated control design framework in which a system can quickly learn a desirable control from scratch after only a small number of trials. To further tap on unrealized opportunities from computational data, in the second part, I will present several novel data-driven approaches to nonlinear analysis including positivity analysis of functions, computation of invariant sets, and quantification of system complexity. 

Over the past decades, model-based approaches have been the dominant paradigm in rigorous analysis and systematic control design, which is reflected in numerous applications ranging from manufacturing to intelligent robots. A typical model-based control design is often relied on manually generated models that are based on human insights into physics. However, as modern engineering systems become increasingly sophisticated, such classical control designs could not catch up with the rapid development and have presented a number of difficulties regarding to the scalability of the models as well as the overall control systems. To address these challenges, researchers are actively looking for more integrated and automated control approaches which can efficiently and directly learn from data. In light of these initiatives, the research proposal launches a series of fundamental studies aiming to facilitate efficient and valuable integrations of data as well as powerful computation into diverse aspects of system analysis and control design. In particular, I will present in the first part of the proposal a vertically integrated control design framework in which a system can quickly learn a desirable control from scratch after only a small number of trials. To further tap on unrealized opportunities from computational data, in the second part, I will present several novel data-driven approaches to nonlinear analysis including positivity analysis of functions, computation of invariant sets, and quantification of system complexity.