Learning Stabilizing Controllers for High-dimensional Unknown Systems and Networked Dynamical Systems
Designing stabilizing controllers is a fundamental challenge in autonomous systems, particularly for high-dimensional, nonlinear systems that cannot be accurately modeled using differential equations because of the scalability and model transparency, and large-scale networked dynamical systems becau...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2024
|
| Online Access: | https://hdl.handle.net/1721.1/153783 https://orcid.org/0009-0005-6465-4833 |
| Summary: | Designing stabilizing controllers is a fundamental challenge in autonomous systems, particularly for high-dimensional, nonlinear systems that cannot be accurately modeled using differential equations because of the scalability and model transparency, and large-scale networked dynamical systems because of scalability and generalizability. To address the challenge, we develop (1) A Lyapunov-based guided exploration framework to learn stabilizing controllers for high-dimensional unknown systems; (2) A compositional neural certificate based on ISS (Input-to-State Stability) Lyapunov functions for finding decentralized stabilizing controllers in large-scale networked dynamical systems. Comprehensive experiments have shown that the proposed methods outperform the prior work in the case of stability, especially in high-dimensional unknown systems and large-scale networked systems. |
|---|