Safe Data-driven Reinforcement Learning Control Using Set-theoretic Approach

Please enable javascript in your browser.

Modares, Amir | 2022

120 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 55089 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Sadati, Naser
Abstract:
Successful deployment of the next-generation safety-critical systems (e.g., self-driving cars or assistive robots) requires certifying their safety during their operation. Therefore, there is an urgent need for developing safe controllers that respect the system’s constraints all the time. Safety, however, is the bare minimum requirement for any safety-critical system, and it is desired to design safe controllers that achieve as much performance as possible. Most existing results in the literature consider either optimal control design without safety specifications or safe control design with only short-sighted optimality. However, long-term optimality based on long-horizon optimization is highly desired. This paper considers design of state-feedback controllers with safety and optimality requirements using Reinforcement learning. We use data from a single open-loop experiment to design a RL-based feedback controllers enforcing that a given set of the state is invariant and successfully learn a high-performance controller for a long- horizon
Keywords:
Reinforcement Learning ; Optimal Control ; Safety ; Data Driven Control Systems ; Adaptive Control ; Dynamical Systems

No TOC