Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Stochastic triangular mesh mapping
Thesis (PhD)--Stellenbosch University, 2020.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2020
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613851309047808 |
|---|---|
| access_status_str | Open Access |
| author | Lombard, Clint Daniel |
| author2 | Van Daalen, Corne E. |
| author_browse | Lombard, Clint Daniel Van Daalen, Corne E. |
| author_facet | Van Daalen, Corne E. Lombard, Clint Daniel |
| author_sort | Lombard, Clint Daniel |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (PhD)--Stellenbosch University, 2020. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/107853 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:42:41.823Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2020 |
| publishDateRange | 2020 |
| publishDateSort | 2020 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/107853 Stochastic triangular mesh mapping Lombard, Clint Daniel Van Daalen, Corne E. Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Stochastic triangular mesh Mobile robots Robot vision UCTD Stochastic matrices Numerical grid generation (Numerical analysis) Thesis (PhD)--Stellenbosch University, 2020. ENGLISH ABSTRACT: For a mobile robot to operate autonomously in general environments, the ability of the robot to perceive its surroundings is paramount. To perform this task of perception, a robot must have the ability to incrementally construct a model of its environment—otherwise known as online dense mapping. The existing dense mapping techniques do not account for all sources of uncertainty in the system, and also neglect to model any structure inherent in the environment. To address these shortcomings, we present the stochastic triangular mesh (STM) mapping technique: a 2.5-D representation of the surface of the environment using a continuous mesh of triangular surface elements, where each surface element models the mean plane and roughness of the underlying surface. In contrast to existing mapping techniques, an STM map models the structure of the environment by ensuring a continuous model that can be updated incrementally and in an efficient manner—with a linear computational cost in the number of measurements. This efficiency is due to the use of approximate message-passing techniques; specifically, a combination of loopy belief propagation (LBP) and variational message passing (VMP). The uncertainty in the measurements of the environment and robot pose (position and orientation) is accounted for by the use of these Bayesian inference techniques during the map update. We demonstrate that an STM map can be used with sensors that generate point measurements, such as light detection and ranging (LiDAR) sensors and stereo cameras. Simulated results show that, when comparing the log likelihoods of the models, an STM map is a more accurate model than the only comparable online surface-mapping technique—a standard elevation map—while also being as expressive as the offline Gaussian process (GP) mapping technique. We also provide qualitative results on a large-scale practical dataset. In addition to presenting the STM mapping technique, we demonstrate that performing dense mapping in the relative inertial reference frame (IRF) of a triangular submap has several advantages over the traditional approach using a single global IRF, and extend the 2-D hybrid metric map (HYMM) framework to three dimensions. We demonstrate that performing dense mapping in a relative IRF addresses issues like loop closure or the kidnapped robot problem, which affect mapping in a global IRF. AFRIKAANSE OPSOMMING: Vir ’n mobiele robot om outonoom in algemene omgewings te werk, is die vermoë van die robot om die omgewing waar te neem van groot belang. Om hierdie taak van waarneming te verrig, moet ’n robot die vermoë hê om op ’n inkrementele wyse ’n model van sy omgewing op te stel—dit staan bekend as aanlyn digte kartering. Die bestaande digte karteringstegnieke neem nie alle bronne van onsekerheid in die stelsel in ag nie, en modelleer ook nie enige struktuur inherent aan die omgewing nie. Om hierdie tekortkominge aan te spreek, stel ons die stochastiese-driehoekige-maas-(STM)-kaarteringstegniek voor: ’n 2.5-D-voorstelling van die oppervlak van die omgewing deur middel van ’n kontinue maas van driehoekige oppervlakelemente, waar elke oppervlakelement die gemiddelde vlak en grofheid van die onderliggende oppervlak modelleer In teenstelling met die bestaande karteringstegnieke, modelleer ’n STM-kaart die struktuur van die omgewing deur kontinuïteit af te dwing en dateer die kaart op ’n inkrementele en doeltreffende wyse op—met lineêre koste in die aantal metings. Hierdie doeltreffendheid is as gevolg van die gebruik van benaderde boodskap-oordrag-tegnieke; spesifiek ’n kombinasie van lusvormige kennisvoortplanting (“loopy belief propagation”) en veranderlike boodskap-oordrag (“variational message passing”). Die onsekerheid in die metings van die omgewing en robot-posisie en -oriëntasie word in ag geneem deur die gebruik van hierdie Bayesiese inferensietegnieke tydens die kaartopdatering. Ons demonstreer dat ’n STM-kaart gebruik kan word met sensors wat puntmetings genereer, byvoorbeeld LiDAR-sensors en stereokameras. Simulasieresultate wys dat ’n STM-kaart ’n meer akkurate model is, wanneer die log-waarskynlikheid vergelyk word, as die enigste vergelykbare aanlyn-oppervlak-karteringstegniek—’n gewone hoogtekaart—terwyl dit ook net so ekspressief is as die aflyn Gaussiese-proses-karteringstegniek. Ons wys ook kwalitatiewe resultate van ’n groot praktiese datastel. Benewens die voorstel van die STM-karteringstegniek wys ons ook dat kartering in die relatiewe inersiële koördinaatstelsel van ’n driehoekige subkaart verskeie voordele toon bo die tradisionele benadering, wat ’n enkele globale inersiële koordinaatstelsel gebruik, en brei die raamwerk vir 2-D hibriede metrieke kaart uit na drie dimensies. Ons wys dat deur digte kartering in ’n relatiewe koördinaatstelsel te doen, word probleme soos lusluiting of die ontvoerde-robot-probleem aangespreek. Doctoral 2020-02-24T05:08:09Z 2020-04-28T12:06:14Z 2020-02-24T05:08:09Z 2020-04-28T12:06:14Z 2020-03 Thesis http://hdl.handle.net/10019.1/107853 en Stellenbosch University viii, 82 leaves : illustrations (some color) application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Stochastic triangular mesh Mobile robots Robot vision UCTD Stochastic matrices Numerical grid generation (Numerical analysis) Lombard, Clint Daniel Stochastic triangular mesh mapping |
| title | Stochastic triangular mesh mapping |
| title_full | Stochastic triangular mesh mapping |
| title_fullStr | Stochastic triangular mesh mapping |
| title_full_unstemmed | Stochastic triangular mesh mapping |
| title_short | Stochastic triangular mesh mapping |
| title_sort | stochastic triangular mesh mapping |
| topic | Stochastic triangular mesh Mobile robots Robot vision UCTD Stochastic matrices Numerical grid generation (Numerical analysis) |
| url | http://hdl.handle.net/10019.1/107853 |
| work_keys_str_mv | AT lombardclintdaniel stochastictriangularmeshmapping |