MS Teams (please email firstname.lastname@example.org or the student for the meeting link)
Ruikun Zhou | Applied Mathematics, University of Waterloo
Learning-based Lyapunov Control of Unknown Nonlinear Systems with Stability Guarantees
How to guarantee the safety and stability of nonlinear systems with unknown dynamics remains a challenging problem for control systems. One of the most common scenarios can be reach-avoid tasks for robots with uncertainties. Thanks to state-of-the-art machine learning techniques and classic control theories, this issue is being addressed by data-driven control approaches.
In this talk, I will present how to learn the unknown dynamics and obtain stability guarantees for nonlinear dynamical systems with the help of neural networks. First, I will show how to learn the unknown dynamics using neural networks as a function approximator. Secondly, I will show how to train another neural network as a neural control Lyapunov function to stabilize the learned dynamics with verifiable guarantees. Lastly, other promising learning-based methods for tackling this issue, e.g., Gaussian process regression and Koopman operator theory, will be mentioned. I will briefly discuss how to implement them as the potential future work for my PhD studies.