Back-end Performance Analysis of Large-Scale Simultaneous Localization and Mapping

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

Sähkötekniikan korkeakoulu | Master's thesis

Date

2021-05-17

Department

Major/Subject

Control, Robotics and Autonomous Systems

Mcode

ELEC3025

Degree programme

AEE - Master’s Programme in Automation and Electrical Engineering (TS2013)

Language

en

Pages

58

Series

Abstract

Simultaneous Localization and Mapping (SLAM) refers to the process in which a robot is mapping an unknown environment while simultaneously using maps for estimating poses. Two main methods to solve a SLAM problem are probabilistic estimation and optimization. As the probabilistic estimation is inconsistent in the large-scale environment, there are several optimization approaches to addressing the nonlinear large-scale SLAM problems.The optimization in SLAM is also called as the back-end. In order to improve the performance of the back-ends, this work mainly compared performance of different back-ends in the large-scale normal distributions transform (NDT) LiDAR-based SLAM. There are multiple factors which can affect the performance of the back-ends. This work was focus on the back-end performance analysis which includes the accuracy, the computation time, and the sensitivity to noise of the back-ends. Also, the incremental smoothing and mapping (ISAM) and incremental smoothing and mapping using Bayes tree (ISAM2) are chosen to be studied and compared as novel back-ends in this work. The goal of the work is to analyze how the two back-ends (ISAM or ISAM2) perform in large-scale 3D LiDAR-based NDT SLAM. Furthermore, We compared the efficiency and accuracy of two back-ends using the simulation and real-world platform. A system including software and hardware was introduced. The performance was evaluated in several groups in three different scenarios. The experimental results show that ISAM is more accurate than ISAM2 and the overall accuracy is quite close in the large-scale environment. The computation time of ISAM is considerably greater than ISAM2. The sensitivity to noise of both ISAM and ISAM2 is at the same level. Overall, ISAM2 has better performance in large-scale 3D LiDAR-based NDT SLAM.

Description

Supervisor

Kyrki, Ville

Thesis advisor

Ahtiainen, Juhana

Keywords

large-scale simultaneous localization and Mmpping, normal distribution transforms, incremental smoothing, optimization

Other note

Citation