TR2025-105
Policy Optimization for PDE Control with a Warm Start
-
- "Policy Optimization for PDE Control with a Warm Start", American Control Conference (ACC), July 2025.BibTeX TR2025-105 PDF
- @inproceedings{Zhang2025jul,
- author = {Zhang, Xiangyuan and Mowlavi, Saviz and Benosman, Mouhacine and Basar, Tamer},
- title = {{Policy Optimization for PDE Control with a Warm Start}},
- booktitle = {American Control Conference (ACC)},
- year = 2025,
- month = jul,
- url = {https://www.merl.com/publications/TR2025-105}
- }
,
- "Policy Optimization for PDE Control with a Warm Start", American Control Conference (ACC), July 2025.
-
MERL Contact:
-
Research Areas:
Computational Sensing, Control, Dynamical Systems, Machine Learning, Signal Processing
Abstract:
Dimensionality reduction is crucial for control- ling nonlinear partial differential equations (PDE) through a “reduce-then-design” strategy, which identifies a reduced-order model and then implements model-based control solutions. However, inaccuracies in the reduced-order modeling can substantially degrade controller performance, especially in PDEs with chaotic behavior. To address this issue, we augment the reduce-then-design procedure with a policy optimization (PO) step. The PO step fine-tunes the model-based controller to compensate for the modeling error from dimensionality reduction. This augmentation shifts the overall strategy into reduce-then- design-then-adapt, where the model-based controller serves as a warm start for PO. Specifically, we study the state-feedback tracking control of PDEs that aims to align the PDE state with a specific constant target subject to a linear-quadratic cost. Through extensive experiments, we show that a few iterations of PO can significantly improve the model-based controller performance. Our approach offers a cost-effective alternative to PDE control using end-to-end reinforcement learning.
Related News & Events
-
NEWS MERL researchers present 13 papers at ACC 2025 Date: July 8, 2025 - July 10, 2025
Where: Denver, USA
MERL Contacts: Ankush Chakrabarty; Vedang M. Deshpande; Stefano Di Cairano; Purnanand Elango; Jordan Leung; Saviz Mowlavi; Diego Romeres; Abraham P. Vinod; Yebin Wang; Avishai Weiss
Research Areas: Control, Dynamical Systems, Electric Systems, Machine Learning, Multi-Physical Modeling, RoboticsBrief- MERL researchers presented 13 papers at the recently concluded American Control Conference (ACC) 2025 in Denver, USA. The papers covered a wide range of topics including Bayesian optimization for personalized medicine, machine learning for battery performance in eVTOLs, model predictive control for space and building systems, process systems engineering for sustainability, GNSS-RTK optimization, convex set manipulation, PDE control, servo system modeling, battery fault diagnosis, truck fleet coordination, interactive motion planning, and satellite station keeping. Additionally, MERL researchers (Vedang Deshpande and Ankush Chakrabarty) organized an invited session on design and optimization of energy systems.
As a sponsor of the conference, MERL maintained a booth for open discussions with researchers and students, and hosted a special session to discuss highlights of MERL research and work philosophy.
- MERL researchers presented 13 papers at the recently concluded American Control Conference (ACC) 2025 in Denver, USA. The papers covered a wide range of topics including Bayesian optimization for personalized medicine, machine learning for battery performance in eVTOLs, model predictive control for space and building systems, process systems engineering for sustainability, GNSS-RTK optimization, convex set manipulation, PDE control, servo system modeling, battery fault diagnosis, truck fleet coordination, interactive motion planning, and satellite station keeping. Additionally, MERL researchers (Vedang Deshpande and Ankush Chakrabarty) organized an invited session on design and optimization of energy systems.