Generation of human-aware navigation maps using graph neural networks

Daniel Rodriguez-Criado, Pilar Bachiller-Burgos, Luis J. Manso

Research output: Chapter in Book/Published conference outputConference publication


Minimising the discomfort caused by robots when navigating in social situations is crucial for them to be accepted. Graph Neural Net-works can process representations including arbitrarily complex relation-ships between entities such as human interactions. This is particularly interesting in the context of social navigation, where relational information should be considered. This paper presents a model combining Graph Neural Network (GNN) and Convolutional Neural Network (CNN) layers to produce cost maps for human-aware navigation in real-time. The model leverages the relational inductive bias of GNNs to generate scenario representations that can be efficiently exploited using CNNs. In addition, a framework to bootstrap existing zero-dimensional models to generate cost map datasets is proposed. The model is evaluated against the original zero-dimensional dataset and in simulated navigation tasks.The results outperform similar state-of-the-art-methods considering the accuracy for the dataset and the navigation metrics used. The applications of the proposed framework are not limited to human-aware navigation, it could be applied to other fields where cost map generation is needed.
Original languageEnglish
Title of host publicationArtificial Intelligence XXXVIII - 41st SGAI International Conference on Artificial Intelligence, AI 2021, Proceedings
EditorsMax Bramer, Richard Ellis
Number of pages14
ISBN (Electronic)978-3-030-91100-3
ISBN (Print)978-3-030-91099-0
Publication statusPublished - 6 Dec 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13101 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Dive into the research topics of 'Generation of human-aware navigation maps using graph neural networks'. Together they form a unique fingerprint.

Cite this