Navigating the unknown: a SLAM approach for dynamic rough terrains
Keywords
Loading...
Authors
Issue Date
2024-08-19
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Despite significant advancements in autonomous robotics, navigating rough
and dynamic outdoor environments, such as forests, remains a largely unexplored
challenge. Current approaches often fall short in these complex settings, where
traditional methods can be costly and environmentally intrusive. In response,
this thesis aims to introduce a system designed specifically for these challenging
environments. The system combines visual and depth data to detect and classify
obstacles using a YOLOv5 neural network. Detected objects, particularly trees,
are tracked and clustered using the DBSCAN algorithm, with confirmed landmarks
added to a 2D occupancy grid. Loop closure detection helps maintain map accuracy
as the robot revisits areas. While the approach offers a promising framework for
mapping in forestry environments, the results indicate that further refinement is
needed. The system encountered challenges in maintaining real-time performance
and handling the intricacies of rough terrains, suggesting that additional work is
required to fully realize its potential.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen