Abstract
dc:description.abstractThis thesis addresses the optimal control of state‐dependent switched systems—a class of hybrid dynamical systems characterized by state‐triggered transitions across known interfaces. Such systems arise in applications ranging from robotics and chemical process control to electronic circuits, yet the classical Pontryagin Minimum Principle applies only to continuously differentiable dynamics. Moreover, existing numerical toolboxes often lack both ease of use and rigorous error guarantees. To overcome these challenges, this thesis develops new theoretical foundations and efficient computational algorithms. In the first part of the thesis, we propose a generalized Euler-Lagrange Equation (GEL) which clarifies a quantitative property of the co-state's behavior for systems with a time-invariant switching interface. We derive the optimal control for state-dependent switched systems using the calculus of variations technique. We further extend the theory to systems with time-varying switching interfaces and constant time delays in states. We illustrate our theoretical results through representative simulation examples. In the second part of the thesis, we develop and analyze numerical methods for the optimal control of state-dependent switched systems. First, we propose a model-based gradient descent method grounded in our hybrid optimality conditions and generalized Euler-Lagrange equation. To address its reliance on known switching sequences and its limited scalability, we then propose a model-free approximation method that uses basis functions to approximate the control input and gradient descent to update the parameters of the control approximation. A precise error bound of this approximation is derived under some assumptions. Simulation results are provided to show the power of this model-free method. We next extend this model-free algorithm to systems with fixed terminal states using the augmented Lagrangian method and dual decomposition. We illustrate our approach on a range of examples, including systems with differentiable dynamics, systems with state-dependent switches, systems with time delays, and systems with both switches and time delays. In addition, we offer a user-friendly MATLAB toolbox that implements this model-free method. In the third part of the thesis, we investigate reinforcement learning techniques for hybrid systems. We demonstrate that applying the deterministic policy gradient directly to state-dependent switched systems results in high variance and instability. To remedy this, we propose a piecewise reinforcement learning scheme, analogous to a piecewise control approximation, which achieves lower variance and improved suboptimal control performance.
Degree
thesis:*- Level thesis:degree_level
- Doctoral
- Department dc:contributor.department
- Electrical and Computer Engineering
- Grantor dc:publisher
- Georgia Institute of Technology
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhou, Mi
- Advisors dc:contributor.advisor
-
- Abdallah, Chaouki T.
- Verriest, Erik I.
- Committee members dc:contributor.committeemember
-
- Wardi, Yorai
- Dieci, Luca
- Tucker, Maegan
- Hale, Matthew
Subjects
dc:subject × 3Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/78692