GoSafeOpt: Scalable Safe Exploration for Global Optimization of Dynamical Systems

Bhavya Sukhija; Matteo Turchetta; David Lindner; Andreas Krause; Sebastian Trimpe; Dominik Baumann

doi:10.1016/j.artint.2023.103922

GoSafeOpt: Scalable Safe Exploration for Global Optimization of Dynamical Systems

Bhavya Sukhija, Matteo Turchetta, David Lindner, Andreas Krause, Sebastian Trimpe, Dominik Baumann

Research output: Contribution to journal › Article › Scientific › peer-review

5 Citations (Scopus)

56 Downloads (Pure)

Abstract

Learning optimal control policies directly on physical systems is challenging. Even a single failure can lead to costly hardware damage. Most existing model-free learning methods that guarantee safety, i.e., no failures, during exploration are limited to local optima. This work proposes GOSAFEOPT as the first provably safe and optimal algorithm that can safely discover globally optimal policies for systems with high-dimensional state space. We demonstrate the superiority of GOSAFEOPT over competing model-free safe learning methods in simulation and hardware experiments on a robot arm.

Original language	English
Article number	103922
Number of pages	25
Journal	Artificial Intelligence
Volume	320
Early online date	20 Apr 2023
DOIs	https://doi.org/10.1016/j.artint.2023.103922
Publication status	Published - Jul 2023
MoE publication type	A1 Journal article-refereed

Keywords

Computer Science - Machine Learning
Electrical Engineering and Systems Science - Systems and Control

Access to Document

10.1016/j.artint.2023.103922Licence: CC BY

1-s2.0-S0004370223000681-mainFinal published version, 1.69 MBLicence: CC BY

Cite this

@article{bd80eb1d187b44bc91dea99c3314249b,

title = "GoSafeOpt: Scalable Safe Exploration for Global Optimization of Dynamical Systems",

abstract = "Learning optimal control policies directly on physical systems is challenging. Even a single failure can lead to costly hardware damage. Most existing model-free learning methods that guarantee safety, i.e., no failures, during exploration are limited to local optima. This work proposes GOSAFEOPT as the first provably safe and optimal algorithm that can safely discover globally optimal policies for systems with high-dimensional state space. We demonstrate the superiority of GOSAFEOPT over competing model-free safe learning methods in simulation and hardware experiments on a robot arm.",

keywords = "Computer Science - Machine Learning, Electrical Engineering and Systems Science - Systems and Control",

author = "Bhavya Sukhija and Matteo Turchetta and David Lindner and Andreas Krause and Sebastian Trimpe and Dominik Baumann",

year = "2023",

month = jul,

doi = "10.1016/j.artint.2023.103922",

language = "English",

volume = "320",

journal = "Artificial Intelligence",

issn = "0004-3702",

publisher = "Elsevier Science B.V.",

}

TY - JOUR

T1 - GoSafeOpt: Scalable Safe Exploration for Global Optimization of Dynamical Systems

AU - Sukhija, Bhavya

AU - Turchetta, Matteo

AU - Lindner, David

AU - Krause, Andreas

AU - Trimpe, Sebastian

AU - Baumann, Dominik

PY - 2023/7

Y1 - 2023/7

N2 - Learning optimal control policies directly on physical systems is challenging. Even a single failure can lead to costly hardware damage. Most existing model-free learning methods that guarantee safety, i.e., no failures, during exploration are limited to local optima. This work proposes GOSAFEOPT as the first provably safe and optimal algorithm that can safely discover globally optimal policies for systems with high-dimensional state space. We demonstrate the superiority of GOSAFEOPT over competing model-free safe learning methods in simulation and hardware experiments on a robot arm.

AB - Learning optimal control policies directly on physical systems is challenging. Even a single failure can lead to costly hardware damage. Most existing model-free learning methods that guarantee safety, i.e., no failures, during exploration are limited to local optima. This work proposes GOSAFEOPT as the first provably safe and optimal algorithm that can safely discover globally optimal policies for systems with high-dimensional state space. We demonstrate the superiority of GOSAFEOPT over competing model-free safe learning methods in simulation and hardware experiments on a robot arm.

KW - Computer Science - Machine Learning

KW - Electrical Engineering and Systems Science - Systems and Control

UR - http://adsabs.harvard.edu/abs/2022arXiv220109562S

UR - http://www.scopus.com/inward/record.url?scp=85153578129&partnerID=8YFLogxK

U2 - 10.1016/j.artint.2023.103922

DO - 10.1016/j.artint.2023.103922

M3 - Article

SN - 0004-3702

VL - 320

JO - Artificial Intelligence

JF - Artificial Intelligence

M1 - 103922

ER -

GoSafeOpt: Scalable Safe Exploration for Global Optimization of Dynamical Systems

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this