Active Learning of Dynamics for Data-Driven Control Using Koopman Operators

Ian Abraham, Todd David Murphey

Research output: Contribution to journalArticle

Abstract

This paper presents an active learning strategy for robotic systems that takes into account task information, enables fast learning, and allows control to be readily synthesized by taking advantage of the Koopman operator representation. We first motivate the use of representing nonlinear systems as linear Koopman operator systems by illustrating the improved model-based control performance with an actuated Van der Pol system. Information-theoretic methods are then applied to the Koopman operator formulation of dynamical systems where we derive a controller for active learning of robot dynamics. The active learning controller is shown to increase the rate of information about the Koopman operator. In addition, our active learning controller can readily incorporate policies built on the Koopman dynamics, enabling the benefits of fast active learning and improved control. Results using a quadcopter illustrate single-execution active learning and stabilization capabilities during free fall. The results for active learning are extended for automating Koopman observables and we implement our method on real robotic systems.

Original languageEnglish (US)
JournalIEEE Transactions on Robotics
DOIs
StatePublished - Jan 1 2019

Fingerprint

Controllers
Robotics
Problem-Based Learning
Mathematical operators
Nonlinear systems
Dynamical systems
Stabilization
Robots

Keywords

  • Active learning
  • information theoretic control
  • Koopman operators
  • single execution learning

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Computer Science Applications
  • Electrical and Electronic Engineering

Cite this

@article{dd9d6726143042d5abb6f3a7aa3f8e50,
title = "Active Learning of Dynamics for Data-Driven Control Using Koopman Operators",
abstract = "This paper presents an active learning strategy for robotic systems that takes into account task information, enables fast learning, and allows control to be readily synthesized by taking advantage of the Koopman operator representation. We first motivate the use of representing nonlinear systems as linear Koopman operator systems by illustrating the improved model-based control performance with an actuated Van der Pol system. Information-theoretic methods are then applied to the Koopman operator formulation of dynamical systems where we derive a controller for active learning of robot dynamics. The active learning controller is shown to increase the rate of information about the Koopman operator. In addition, our active learning controller can readily incorporate policies built on the Koopman dynamics, enabling the benefits of fast active learning and improved control. Results using a quadcopter illustrate single-execution active learning and stabilization capabilities during free fall. The results for active learning are extended for automating Koopman observables and we implement our method on real robotic systems.",
keywords = "Active learning, information theoretic control, Koopman operators, single execution learning",
author = "Ian Abraham and Murphey, {Todd David}",
year = "2019",
month = "1",
day = "1",
doi = "10.1109/TRO.2019.2923880",
language = "English (US)",
journal = "IEEE Transactions on Robotics",
issn = "1552-3098",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

Active Learning of Dynamics for Data-Driven Control Using Koopman Operators. / Abraham, Ian; Murphey, Todd David.

In: IEEE Transactions on Robotics, 01.01.2019.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Active Learning of Dynamics for Data-Driven Control Using Koopman Operators

AU - Abraham, Ian

AU - Murphey, Todd David

PY - 2019/1/1

Y1 - 2019/1/1

N2 - This paper presents an active learning strategy for robotic systems that takes into account task information, enables fast learning, and allows control to be readily synthesized by taking advantage of the Koopman operator representation. We first motivate the use of representing nonlinear systems as linear Koopman operator systems by illustrating the improved model-based control performance with an actuated Van der Pol system. Information-theoretic methods are then applied to the Koopman operator formulation of dynamical systems where we derive a controller for active learning of robot dynamics. The active learning controller is shown to increase the rate of information about the Koopman operator. In addition, our active learning controller can readily incorporate policies built on the Koopman dynamics, enabling the benefits of fast active learning and improved control. Results using a quadcopter illustrate single-execution active learning and stabilization capabilities during free fall. The results for active learning are extended for automating Koopman observables and we implement our method on real robotic systems.

AB - This paper presents an active learning strategy for robotic systems that takes into account task information, enables fast learning, and allows control to be readily synthesized by taking advantage of the Koopman operator representation. We first motivate the use of representing nonlinear systems as linear Koopman operator systems by illustrating the improved model-based control performance with an actuated Van der Pol system. Information-theoretic methods are then applied to the Koopman operator formulation of dynamical systems where we derive a controller for active learning of robot dynamics. The active learning controller is shown to increase the rate of information about the Koopman operator. In addition, our active learning controller can readily incorporate policies built on the Koopman dynamics, enabling the benefits of fast active learning and improved control. Results using a quadcopter illustrate single-execution active learning and stabilization capabilities during free fall. The results for active learning are extended for automating Koopman observables and we implement our method on real robotic systems.

KW - Active learning

KW - information theoretic control

KW - Koopman operators

KW - single execution learning

UR - http://www.scopus.com/inward/record.url?scp=85069464845&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85069464845&partnerID=8YFLogxK

U2 - 10.1109/TRO.2019.2923880

DO - 10.1109/TRO.2019.2923880

M3 - Article

AN - SCOPUS:85069464845

JO - IEEE Transactions on Robotics

JF - IEEE Transactions on Robotics

SN - 1552-3098

ER -