CaltechAUTHORS
  A Caltech Library Service

Feedback Linearization for Uncertain Systems via Reinforcement Learning

Westenbroek, Tyler and Fridovich-Keil, David and Mazumdar, Eric and Arora, Shreyas and Prabhu, Valmik and Sastry, S. Shankar and Tomlin, Claire J. (2020) Feedback Linearization for Uncertain Systems via Reinforcement Learning. In: 2020 IEEE International Conference on Robotics and Automation (ICRA). IEEE , Piscataway, NJ, pp. 1364-1371. ISBN 978-1-7281-7395-5. https://resolver.caltech.edu/CaltechAUTHORS:20210903-222215650

[img] PDF - Published Version
Creative Commons Public Domain Dedication.

2MB

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20210903-222215650

Abstract

We present a novel approach to control design for nonlinear systems which leverages model-free policy optimization techniques to learn a linearizing controller for a physical plant with unknown dynamics. Feedback linearization is a technique from nonlinear control which renders the input-output dynamics of a nonlinear plant linear under application of an appropriate feedback controller. Once a linearizing controller has been constructed, desired output trajectories for the nonlinear plant can be tracked using a variety of linear control techniques. However, the calculation of a linearizing controller requires a precise dynamics model for the system. As a result, model-based approaches for learning exact linearizing controllers generally require a simple, highly structured model of the system with easily identifiable parameters. In contrast, the model-free approach presented in this paper is able to approximate the linearizing controller for the plant using general function approximation architectures. Specifically, we formulate a continuous-time optimization problem over the parameters of a learned linearizing controller whose optima are the set of parameters which best linearize the plant. We derive conditions under which the learning problem is (strongly) convex and provide guarantees which ensure the true linearizing controller for the plant is recovered. We then discuss how model-free policy optimization algorithms can be used to solve a discrete-time approximation to the problem using data collected from the real-world plant. The utility of the framework is demonstrated in simulation and on a real-world robotic platform.


Item Type:Book Section
Related URLs:
URLURL TypeDescription
https://doi.org/10.1109/icra40945.2020.9197158DOIArticle
ORCID:
AuthorORCID
Westenbroek, Tyler0000-0003-1111-3118
Fridovich-Keil, David0000-0002-5866-6441
Mazumdar, Eric0000-0002-1815-269X
Tomlin, Claire J.0000-0003-3192-3185
Additional Information:U.S. Government work not protected by U.S. copyright. This work was supported by HICON-LEARN (design of HIgh CONfidence LEARNing-enabled systems), Defense Advanced Research Projects Agency award number FA8750-18-C-0101, and Provable High Confidence Human Robot Interactions, Office of Naval Research award number N00014-19-1-2066.
Funders:
Funding AgencyGrant Number
Defense Advanced Research Projects Agency (DARPA)FA8750-18-C-0101
Office of Naval Research (ONR)N00014-19-1-2066
DOI:10.1109/ICRA40945.2020.9197158
Record Number:CaltechAUTHORS:20210903-222215650
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20210903-222215650
Official Citation:T. Westenbroek et al., "Feedback Linearization for Uncertain Systems via Reinforcement Learning," 2020 IEEE International Conference on Robotics and Automation (ICRA), 2020, pp. 1364-1371, doi: 10.1109/ICRA40945.2020.9197158
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:110734
Collection:CaltechAUTHORS
Deposited By: George Porter
Deposited On:07 Sep 2021 21:20
Last Modified:07 Sep 2021 21:20

Repository Staff Only: item control page