Abstract
This paper proposes a data-driven method for infinite-horizon optimal control with unknown nonlinear dynamics. It introduces a Koopman-based gradient estimation framework integrated with actor-critic ideas to iteratively update policy parameters via gradient descent. Analysis and experiments show effective convergence and competitive control performance versus model-free and model-based baselines.
Demo
Demo coming soon.
Citation
Hao, Wenjian, Paulo C. Heredia, and Shaoshuai Mou. 2023. "Optimal Control of Nonlinear Systems with Unknown Dynamics." arXiv preprint arXiv:2305.15188.
@techreport{WHao2023optimalc,
author = {W Hao, PC Heredia, S Mou},
year = {2023},
title = {Optimal Control of Nonlinear Systems with Unknown Dynamics},
number = {arXiv preprint},
url = {https://www.researchgate.net/profile/Wenjian-Hao-2/publication/371009358_Optimal_Control_of_Nonlinear_Systems_with_Unknown_Dynamics/links/691cdcd8de814309827224ae/Optimal-Control-of-Nonlinear-Systems-with-Unknown-Dynamics.pdf}
}