real-time-optimal-control-shallow-recurrent-decoder

name	real-time-optimal-control-shallow-recurrent-decoder
description	Real-time optimal control framework using SHallow REcurrent Decoder networks-based Reduced Order Modeling (SHRED-ROM) for high-dimensional and parametric dynamical systems. Enables synthesis of closed-loop controllers from limited state sensor readings with effective distributed control actions in new scenarios.
metadata	{"arxiv_id":"2607.19302","published":"2026-07-21","authors":["Matteo Tomasetto","Francesco Braghin","J. Nathan Kutz","Andrea Manzoni"],"subjects":["Machine Learning (cs.LG)","Optimization and Control (math.OC)"],"tags":["SHRED-ROM","real-time-control","recurrent-decoder","reduced-order-modeling","optimal-control","systems-engineering"]}
license	Complete terms in LICENSE.txt

Real-time Optimal Control with Shallow Recurrent Decoder Networks

Overview

This methodology exploits SHallow REcurrent Decoder networks-based Reduced Order Modeling (SHRED-ROM) to synthesize real-time closed-loop controllers for high-dimensional and parametric dynamics, relying solely on limited state sensor readings.

Core Methodology

Problem Context

Controlling dynamical systems in real-time across multiple scenarios is critical for adaptive control strategies, ensuring stability and efficiency. Traditional optimal control problems require several system simulations, which are computationally demanding due to high-dimensionality of underlying spatio-temporal dynamics.

Solution Approach

SHRED-ROM synthesizes real-time closed-loop controllers by:

1. Training on a few optimal examples from an expert demonstrator