TR2022-064
VABO: Violation-Aware Bayesian Optimization for Closed-Loop Control Performance Optimization with Unmodeled Constraints
-
- , "VABO: Violation-Aware Bayesian Optimization for Closed-Loop Control Performance Optimization with Unmodeled Constraints", American Control Conference (ACC), June 2022, pp. 5288-5293.BibTeX TR2022-064 PDF
- @inproceedings{Xu2022jun,
- author = {Xu, Wenjie and Jones, Colin and Svetozarevic, Bratislav and Laughman, Christopher R. and Chakrabarty, Ankush},
- title = {VABO: Violation-Aware Bayesian Optimization for Closed-Loop Control Performance Optimization with Unmodeled Constraints},
- booktitle = {American Control Conference (ACC)},
- year = 2022,
- pages = {5288--5293},
- month = jun,
- isbn = {978-1-6654-5197-0},
- url = {https://www.merl.com/publications/TR2022-064}
- }
- , "VABO: Violation-Aware Bayesian Optimization for Closed-Loop Control Performance Optimization with Unmodeled Constraints", American Control Conference (ACC), June 2022, pp. 5288-5293.
-
MERL Contacts:
-
Research Areas:
Abstract:
We study the problem of performance optimiza- tion of closed-loop control systems with unmodeled dynamics. Bayesian optimization (BO) has been demonstrated effective for improving closed-loop performance by automatically tuning controller gains or reference setpoints in a model-free manner. However, BO methods have rarely been tested on dynamical systems with unmodeled constraints. In this paper, we propose a violation-aware BO algorithm (VABO) that optimizes closed- loop performance while simultaneously learning constraint- feasible solutions. Unlike classical constrained BO methods which allow an unlimited constraint violations, or ‘safe’ BO algorithms that are conservative and try to operate with near- zero violations, we allow budgeted constraint violations to improve constraint learning and accelerate optimization. We demonstrate the effectiveness of our proposed VABO method for energy minimization of industrial vapor compression systems.
Related News & Events
-
NEWS Karl Berntorp gave Spotlight Talk at CDC Workshop on Gaussian Process Learning-Based Control Date: December 5, 2022
Where: Cancun, Mexico
MERL Contact: Karl Berntorp
Research Areas: Control, Machine LearningBrief- Karl Berntorp was an invited speaker at the workshop on Gaussian Process Learning-Based Control organized at the Conference on Decision and Control (CDC) 2022 in Cancun, Mexico.
The talk was part of a tutorial-style workshop aimed to provide insight into the fundamentals behind Gaussian processes for modeling and control and sketching some of the open challenges and opportunities using Gaussian processes for modeling and control. The talk titled ``Gaussian Processes for Learning and Control: Opportunities for Real-World Impact" described some of MERL's efforts in using Gaussian processes (GPs) for learning and control, with several application examples and discussing some of the key benefits and limitations with using GPs for learning-based control.
- Karl Berntorp was an invited speaker at the workshop on Gaussian Process Learning-Based Control organized at the Conference on Decision and Control (CDC) 2022 in Cancun, Mexico.
-
NEWS MERL researchers win ASME Energy Systems Technical Committee Best Paper Award at 2022 American Control Conference Date: June 8, 2022
Where: 2022 American Control Conference
MERL Contacts: Ankush Chakrabarty; Christopher R. Laughman
Research Areas: Control, Machine Learning, Multi-Physical Modeling, OptimizationBrief- Researchers from EPFL (Wenjie Xu, Colin Jones) and EMPA (Bratislav Svetozarevic), in collaboration with MERL researchers Ankush Chakrabarty and Chris Laughman, recently won the ASME Energy Systems Technical Committee Best Paper Award at the 2022 American Control Conference for their work on "VABO: Violation-Aware Bayesian Optimization for Closed-Loop Performance Optimization with Unmodeled Constraints" out of 19 nominations and 3 finalists. The paper describes a data-driven framework for optimizing the performance of constrained control systems by systematically re-evaluating how cautiously/aggressively one should explore the search space to avoid sustained, large-magnitude constraint violations while tolerating small violations, and demonstrates these methods on a physics-based model of a vapor compression cycle.
-
NEWS MERL researchers presented 9 papers at the American Control Conference (ACC) Date: June 8, 2022 - June 10, 2022
Where: Atlanta, GA
MERL Contacts: Karl Berntorp; Scott A. Bortoff; Ankush Chakrabarty; Stefano Di Cairano; Christopher R. Laughman; Abraham P. Vinod; Avishai Weiss
Research Areas: Control, Machine Learning, OptimizationBrief- At the American Control Conference in Atlanta, GA, MERL presented 9 papers on subjects including autonomous-vehicle decision making and motion planning, realtime Bayesian inference and learning, reference governors for hybrid systems, Bayesian optimization, and nonlinear control.
Related Publication
BibTeX arXiv
- @article{Xu2021oct,
- author = {Xu, Wenjie and Jones, Colin and Svetozarevic, Bratislav and Laughman, Christopher R. and Chakrabarty, Ankush},
- title = {VABO: Violation-Aware Bayesian Optimization for Closed-Loop Control Performance Optimization with Unmodeled Constraints},
- journal = {arXiv},
- year = 2021,
- month = oct,
- url = {https://arxiv.org/abs/2110.07479}
- }