Position uncertainty reduction in visualInertial navigation systems using multi-ML error compensation

Simple item page
dc.contributor.authorTabassum, Tarafder Elmi
dc.contributor.authorPetrunin, Ivan
dc.contributor.authorRana, Zeeshan A
dc.date.accessioned2025-02-11T14:01:52Z
dc.date.available2025-02-11T14:01:52Z
dc.date.freetoread2025-02-11
dc.date.issued2024-10-09
dc.date.pubOnline2024-10-09
dc.description.abstractIn the absence of signals from global navigation satellite systems (GNSS), visual-inertial navigation systems (VINS) are usually utilized in urban air mobility (UAM) applications which require reliable navigation in complex urban areas. This paper focuses on improving vision-based alternatives to GNSS navigation solutions with position uncertainty correction for safe and uninterrupted flights. The novel contribution introduces multiple machine-learning aided hybrid visual-inertial odometry (multi-ML hybrid VIO) that utilizes gated recurrent unit (GRU) based error compensators to enhance positioning within complex environments by reducing the impacts of various sources of uncertainty. Unlike state-of-the-art systems that lack evidence of demonstrating performance enhancement with position uncertainty correction within VIO architectures, the proposed framework simultaneously reduces position uncertainty and improves accuracy. Furthermore, training and testing datasets are generated using MATLAB incorporating unreal engine simulation environment for UAVs to replicate complex scenarios including environmental conditions, illumination variations, weather effects and flight dynamics where traditional VIO systems tend to fail. The proposed hybrid VIO architecture has been validated under combinations of complex scenarios including various sources of uncertainty such as sensor noise, feature tracking error, environmental dynamics, weather effects and lighting conditions for extended flights. The comparison results have demonstrated reduction in horizontal positioning RMSE errors: 1.7m for VIO with VO error compensation 2.18m for VIO with KF error compensation, 1.4m for multi-ML hybrid VIO. Furthermore, it demonstrates generalization ability over seen and unseen fault scenarios that indicates performance improvement of 89% in 3D position compared to VIO with VO error compensation, VIO with KF error compensation, and ESKF-based VIO. Additionally, experimental results demonstrate overall horizontal position uncertainty reduction by 79% for test 1 and 22% for test 2. Finally, this work represents a step forward in improving the safety and effectiveness of UAV navigation by providing vision-based alternative to GNSS solution for uninterrupted flights.
dc.description.conferencename37th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2024)
dc.format.extentpp. 1741-1755
dc.identifier.citationTabassum TE, Petrunin I, Rana ZA. (2024) Position uncertainty reduction in visualInertial navigation systems using multi-ML error compensation. Proceedings of the 37th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2024), 16 - 20 Sep 2024, Baltimore, Maryland, pp. 1741-1755
dc.identifier.elementsID555643
dc.identifier.issn2331-5954
dc.identifier.urihttps://doi.org/10.33012/2024.19790
dc.identifier.urihttps://dspace.lib.cranfield.ac.uk/handle/1826/23467
dc.language.isoen
dc.publisherInstitute of Navigation
dc.publisher.urihttps://www.ion.org/publications/abstract.cfm?articleID=19790
dc.rightsAttribution 4.0 Internationalen
dc.rights.urihttp://creativecommons.org/licenses/by/4.0/
dc.subject4007 Control Engineering, Mechatronics and Robotics
dc.subject40 Engineering
dc.subject10 Reduced Inequalities
dc.titlePosition uncertainty reduction in visualInertial navigation systems using multi-ML error compensation
dc.typeConference paper
dcterms.coverageBaltimore, Maryland
dcterms.temporal.endDate20 Sep 2024
dcterms.temporal.startDate16 Sep 2024

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
Position_uncertainty_reduction-2024.pdf
Size:
3.96 MB
Format:
Adobe Portable Document Format
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.63 KB
Format:
Plain Text
Description:
Download

Collections

Staff publications (SATM)