Learning Safe Neural Network Controllers with Barrier Certificates

Hengjun Zhao; Xia Zeng; Taolue Chen; Zhiming Liu; Jim Woodcock

Learning Safe Neural Network Controllers with Barrier Certificates

Hengjun Zhao, Xia Zeng, Taolue Chen, Zhiming Liu, Jim Woodcock

Computer Science

Research output: Working paper › Preprint

Abstract

We provide a novel approach to synthesize controllers for
nonlinear continuous dynamical systems with control against safety properties. The controllers are based on neural networks (NNs). To certify
the safety property we utilize barrier functions, which are represented by
NNs as well. We train the controller-NN and barrier-NN simultaneously,
achieving a verification-in-the-loop synthesis. We provide a prototype
tool nncontroller with a number of case studies. The experiment results
confirm the feasibility and efficacy of our approach.

Original language	English
Volume	abs/2009.09826
Publication status	Published - 18 Sept 2020

Publication series

Name	CoRR

Keywords

Continuous dynamical systems
Controller synthesis
Neural networks
Safety verification
Barrier certificates

Access to Document

https://arxiv.org/abs/2009.09826

Cite this

@techreport{1ed0dd3a8c8b453483e29bc4a46d0eb7,

title = "Learning Safe Neural Network Controllers with Barrier Certificates",

abstract = "We provide a novel approach to synthesize controllers fornonlinear continuous dynamical systems with control against safety properties. The controllers are based on neural networks (NNs). To certifythe safety property we utilize barrier functions, which are represented byNNs as well. We train the controller-NN and barrier-NN simultaneously,achieving a verification-in-the-loop synthesis. We provide a prototypetool nncontroller with a number of case studies. The experiment resultsconfirm the feasibility and efficacy of our approach.",

keywords = "Continuous dynamical systems, Controller synthesis, Neural networks, Safety verification, Barrier certificates",

author = "Hengjun Zhao and Xia Zeng and Taolue Chen and Zhiming Liu and Jim Woodcock",

year = "2020",

month = sep,

day = "18",

language = "English",

volume = "abs/2009.09826",

series = "CoRR",

type = "WorkingPaper",

}

TY - UNPB

T1 - Learning Safe Neural Network Controllers with Barrier Certificates

AU - Zhao, Hengjun

AU - Zeng, Xia

AU - Chen, Taolue

AU - Liu, Zhiming

AU - Woodcock, Jim

PY - 2020/9/18

Y1 - 2020/9/18

N2 - We provide a novel approach to synthesize controllers fornonlinear continuous dynamical systems with control against safety properties. The controllers are based on neural networks (NNs). To certifythe safety property we utilize barrier functions, which are represented byNNs as well. We train the controller-NN and barrier-NN simultaneously,achieving a verification-in-the-loop synthesis. We provide a prototypetool nncontroller with a number of case studies. The experiment resultsconfirm the feasibility and efficacy of our approach.

AB - We provide a novel approach to synthesize controllers fornonlinear continuous dynamical systems with control against safety properties. The controllers are based on neural networks (NNs). To certifythe safety property we utilize barrier functions, which are represented byNNs as well. We train the controller-NN and barrier-NN simultaneously,achieving a verification-in-the-loop synthesis. We provide a prototypetool nncontroller with a number of case studies. The experiment resultsconfirm the feasibility and efficacy of our approach.

KW - Continuous dynamical systems

KW - Controller synthesis

KW - Neural networks

KW - Safety verification

KW - Barrier certificates

M3 - Preprint

VL - abs/2009.09826

T3 - CoRR

BT - Learning Safe Neural Network Controllers with Barrier Certificates

ER -