Ensemble based closed-loop optimal control using physics-informed neural networks

Jostein Barry-Straume

doi:10.1016/j.cnsns.2026.110063

Ensemble based closed-loop optimal control using physics-informed neural networks

2026-04-24

0

PRE

AI

J

Jostein Barry-Straume *

DOI：10.1016/j.cnsns.2026.110063

Original

Original request for help

Save

Abstract

En 中文

• Closed-loop ensemble CPINN learns infinite-horizon HJB value functions. • Analytic feedback policies are recovered from learned value gradients. • Robust ensemble control rejects outlier actions and reintegrates them later. • Warm starts, boundary anchors, and alpha scheduling improve HJB training stability. • Benchmarks show accurate value recovery and robust closed-loop performance.

Keywords:

closed-loop optimal control

physics-informed neural networks

HJB value functions

ensemble methods

robust control

Ensemble based closed-loop optimal control using physics-informed neural networks

Abstract

Journal

Organization