Comparison of SLAM algorithms and neural networks for autonomous navigation in simulated environments

McConnell, Brayden (2020) Comparison of SLAM algorithms and neural networks for autonomous navigation in simulated environments. [USQ Project]

[img]
Preview
Text (Project)
McConnell_B_Low_Redacted.pdf

Download (11MB) | Preview

Abstract

When looking at the research and industrial landscape, the tendency to favour SLAM based algorithms is ever present and for a good reason. SLAM algorithms are the most mature algorithms that we have available today. However, competitors to this are starting to arise to fill in the gaps that SLAM has in its capabilities at the moment. One of these competitors is the utilisation of Deep Neural Networks or DNN’s - a term loosely applied to Deep Learning algorithms.

The proposed study focussed on establishing the characteristic differences between SLAM based algorithms and CNN based algorithms within simulated environments.

The Deep Neural Network being tested was developed by the author of the project and trained using data collected within the chosen simulation’s engine. The simulation engine utilised was Unreal Engine 4 (version 4.24) using the latest version of the AirSim plugin (as of the March 2020 update v1.3.1). The SLAM algorithms being used for the comparison were MATLAB variants of ORB-SLAM and CEKF-SLAM.

The findings of the project are largely inconclusive, with the project conclusion defining that additional work needs to be complete in order to provide conclusive evidence for the project. However, the supporting findings of the project validate its position as having useful research findings. A clear consensus has been established on the methods required to operate within these environments and understanding the limitations that apply under these software conditions. This is information that is currently poorly defined in the Literature and as such this project serves as a strong base for future projects to build on.

Anecdotally, it has been found that the utilisation of CNNs is far more accessible in simulated environments such as AirSim. However, it is a direct result of immaturity for the software utilised. Additionally, the strong Literature Review provides evidence in the direction of CNN utilisation over SLAM based algorithms however these findings were not validated.

The project is aimed to be conducted across the formal 2020 academic year spread across the two courses ENG4111 and ENG4112 as part of the dissertation process for a graduating BENGH Engineer.


Statistics for USQ ePrint 42845
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Additional Information: Bachelors of Engineering (Honours) (Mechatronics) project.
Faculty/School / Institute/Centre: Current - Faculty of Health, Engineering and Sciences - School of Mechanical and Electrical Engineering (1 Jul 2013 -)
Faculty/School / Institute/Centre: Current - Faculty of Health, Engineering and Sciences - School of Mechanical and Electrical Engineering (1 Jul 2013 -)
Supervisors: Low, Tobias
Date Deposited: 22 Jul 2021 01:10
Last Modified: 16 Aug 2021 05:44
Uncontrolled Keywords: SLAM algorithms; neural networks; autonomous navigation
URI: http://eprints.usq.edu.au/id/eprint/42845

Actions (login required)

View Item Archive Repository Staff Only