Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

Abstract

Autonomous robot navigation and dynamic obstacle avoidance in complex, cluttered, indoor environments is a challenging task. A robust solution would allow robots to be deployed in hospitals, airports or shopping centres to serve as guides and fulfil other functions requiring safe human--robot interaction. Previous studies have explored various approaches to selecting sensor types, collecting data, and training models capable of safely avoiding unmapped, possibly dynamic obstacles in an indoor environment. In this paper we address the problem of recognizing and anticipating collisions, in order to determine when avoidance manoeuvres are required. We propose and compare two sensor-fusion and neural-network-based solutions, one in which models are trained separately on static and dynamic samples and another in which a model is trained on samples of collisions with both dynamic and static obstacles. The measured accuracies confirmed that the separately trained, ensemble models had better recognition performance, but were slower at calculation than the models trained without taking the obstacle types into account.
Original languageEnglish
Title of host publication27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
EditorsMichel Verleysen
Number of pages6
Place of PublicationBelgium
PublisherCiaco - i6doc.com
Publication date28 Apr 2019
Pages655-660
ChapterDynamical systems and reinforcement learning
ISBN (Print)9782875870650
ISBN (Electronic)9782875870667
Publication statusPublished - 28 Apr 2019
EventEuropean Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning - Bruges, Belgium
Duration: 24 Apr 201926 Apr 2019
Conference number: 27
https://www.elen.ucl.ac.be/esann/

Conference

ConferenceEuropean Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
Number27
CountryBelgium
CityBruges
Period24/04/201926/04/2019
Internet address

Fingerprint

Collision avoidance
Navigation
Robots
Shopping centers
Human robot interaction
Sensors
Airports
Fusion reactions
Neural networks

Keywords

  • dynamic object avoidance
  • Robot Navigation
  • convolutional neural networks
  • deep learning
  • robotics

Cite this

Schmuck, V., & Meredith, D. (2019). Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation. In M. Verleysen (Ed.), 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (pp. 655-660). Belgium: Ciaco - i6doc.com.
Schmuck, Viktor ; Meredith, David. / Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation. 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. editor / Michel Verleysen. Belgium : Ciaco - i6doc.com, 2019. pp. 655-660
@inproceedings{3668984648354f7aa781f108da516e93,
title = "Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation",
abstract = "Autonomous robot navigation and dynamic obstacle avoidance in complex, cluttered, indoor environments is a challenging task. A robust solution would allow robots to be deployed in hospitals, airports or shopping centres to serve as guides and fulfil other functions requiring safe human--robot interaction. Previous studies have explored various approaches to selecting sensor types, collecting data, and training models capable of safely avoiding unmapped, possibly dynamic obstacles in an indoor environment. In this paper we address the problem of recognizing and anticipating collisions, in order to determine when avoidance manoeuvres are required. We propose and compare two sensor-fusion and neural-network-based solutions, one in which models are trained separately on static and dynamic samples and another in which a model is trained on samples of collisions with both dynamic and static obstacles. The measured accuracies confirmed that the separately trained, ensemble models had better recognition performance, but were slower at calculation than the models trained without taking the obstacle types into account.",
keywords = "dynamic object avoidance, Robot Navigation, convolutional neural networks, deep learning, robotics",
author = "Viktor Schmuck and David Meredith",
year = "2019",
month = "4",
day = "28",
language = "English",
isbn = "9782875870650",
pages = "655--660",
editor = "Michel Verleysen",
booktitle = "27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning",
publisher = "Ciaco - i6doc.com",

}

Schmuck, V & Meredith, D 2019, Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation. in M Verleysen (ed.), 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. Ciaco - i6doc.com, Belgium, pp. 655-660, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, Bruges, Belgium, 24/04/2019.

Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation. / Schmuck, Viktor; Meredith, David.

27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. ed. / Michel Verleysen. Belgium : Ciaco - i6doc.com, 2019. p. 655-660.

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

TY - GEN

T1 - Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation

AU - Schmuck, Viktor

AU - Meredith, David

PY - 2019/4/28

Y1 - 2019/4/28

N2 - Autonomous robot navigation and dynamic obstacle avoidance in complex, cluttered, indoor environments is a challenging task. A robust solution would allow robots to be deployed in hospitals, airports or shopping centres to serve as guides and fulfil other functions requiring safe human--robot interaction. Previous studies have explored various approaches to selecting sensor types, collecting data, and training models capable of safely avoiding unmapped, possibly dynamic obstacles in an indoor environment. In this paper we address the problem of recognizing and anticipating collisions, in order to determine when avoidance manoeuvres are required. We propose and compare two sensor-fusion and neural-network-based solutions, one in which models are trained separately on static and dynamic samples and another in which a model is trained on samples of collisions with both dynamic and static obstacles. The measured accuracies confirmed that the separately trained, ensemble models had better recognition performance, but were slower at calculation than the models trained without taking the obstacle types into account.

AB - Autonomous robot navigation and dynamic obstacle avoidance in complex, cluttered, indoor environments is a challenging task. A robust solution would allow robots to be deployed in hospitals, airports or shopping centres to serve as guides and fulfil other functions requiring safe human--robot interaction. Previous studies have explored various approaches to selecting sensor types, collecting data, and training models capable of safely avoiding unmapped, possibly dynamic obstacles in an indoor environment. In this paper we address the problem of recognizing and anticipating collisions, in order to determine when avoidance manoeuvres are required. We propose and compare two sensor-fusion and neural-network-based solutions, one in which models are trained separately on static and dynamic samples and another in which a model is trained on samples of collisions with both dynamic and static obstacles. The measured accuracies confirmed that the separately trained, ensemble models had better recognition performance, but were slower at calculation than the models trained without taking the obstacle types into account.

KW - dynamic object avoidance

KW - Robot Navigation

KW - convolutional neural networks

KW - deep learning

KW - robotics

M3 - Article in proceeding

SN - 9782875870650

SP - 655

EP - 660

BT - 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning

A2 - Verleysen, Michel

PB - Ciaco - i6doc.com

CY - Belgium

ER -

Schmuck V, Meredith D. Training networks separately on static and dynamic obstacles improves collision avoidance during indoor robot navigation. In Verleysen M, editor, 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. Belgium: Ciaco - i6doc.com. 2019. p. 655-660