Real-time optimal control with shallow recurrent decoder networks

2026-07-21

AI Summary

This work addresses the challenge of real-time optimal control in high-dimensional dynamical systems, where sparse, delayed, or failed sensor data often lead to prohibitive computational costs. To overcome this limitation, the authors propose the SHRED-ROM framework, which leverages a small number of expert demonstrations to construct a reduced-order model using a shallow recurrent decoder. Within a latent space, SHRED-ROM jointly learns a control policy and a sensor predictor, enabling closed-loop control from limited observations. The approach substantially reduces system dimensionality while accurately reproducing expert behavior and introduces, for the first time, a latent-space closed-loop mechanism robust to sensor anomalies. Evaluated on three high-dimensional parametric fluid and density field control tasks, SHRED-ROM demonstrates stable, efficient, and robust real-time performance.

Abstract

Controlling dynamical systems in real-time across multiple scenarios is critical to enabling adaptive control strategies, ensuring stability and efficiency. However, to tailor control actions in response to varying scenarios, traditional optimal control problems typically require several system simulations, which are often computationally demanding due to the high-dimensionality of the underlying spatio-temporal dynamics. In this work, we exploit SHallow REcurrent Decoder networks-based Reduced Order Modeling (SHRED-ROM) to synthesize a real-time closed-loop controller for high-dimensional and parametric dynamics, relying solely on limited state sensor readings. After training the model on a few optimal examples given by an expert demonstrator, SHRED-ROM mimics the expert behavior with effective distributed control actions in new scenarios, alleviating the curse of dimensionality. Moreover, a sensor forecaster is synthesized and used to close the loop at the latent level, thus efficiently mitigating possible sensor failures or delays. The performance of the proposed optimal control strategy is finally assessed on three challenging high-dimensional cases dealing with either parametric density control or fluid flow control.

Problem

• real-time optimal control
• high-dimensional dynamics
• parametric systems
• adaptive control
• sensor-based feedback

Innovation

• SHRED-ROM
• real-time optimal control
• reduced order modeling
• shallow recurrent networks
• sensor forecasting