Traffic Signal Control in partially observable environments using Graph Neural Networks with Variational Autoencoders.

Keywords

Loading...
Thumbnail Image

Issue Date

2021-06-18

Language

en

Document type

Journal Title

Journal ISSN

Volume Title

Publisher

Title

ISSN

Volume

Issue

Startpage

Endpage

DOI

Abstract

Tra c Signal Control (TSC) is used to improve the tra c e ciency of a district or city, which is more important than ever with populations rising. Current TSC methods, however, do not take the uncertainty of the real world into account and therefore perform sub-optimal with missing or incorrect data. This work proposes a Graph Neural Network Variational Autoencoder architecture to reconstruct missing or incorrect data, so that these TSC systems are still able to perform e ciently in the real world. In experiments is demonstrated that the proposed model improves the results of the state- of-the-art Colight model in an uncertain environment compared to the same model without the use of reconstructed data. However, it does not perform nearly as well as in an fully observable environment.

Description

Citation

Faculty

Faculteit der Sociale Wetenschappen