LEARNING-BASED CONTROL: THEORY AND PRACTICE

Credit Hours: 3 · Graduate

For several engineering systems, there are no good first-principles approaches for the design of controllers for them. While ad-hoc control strategies may work, we will investigate the potential for machine learning-based approaches. In this course, we will focus on when we can attain theoretical guarantees using learning-based approaches to control. The course will present a selected background in nonlinear control and neural networks and then spend most of the remaining time on neural Lyapunov functions, neural barrier functions, and neural contraction metrics.