Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Maximising exploration and discovery in unknown large intricate 3D worlds
Thesis (MEng)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614075251326976 |
|---|---|
| access_status_str | Open Access |
| author | Legg, Tristan Josef |
| author2 | Engelbrecht, H. A. (Herman) |
| author_browse | Engelbrecht, H. A. (Herman) Legg, Tristan Josef |
| author_facet | Engelbrecht, H. A. (Herman) Legg, Tristan Josef |
| author_sort | Legg, Tristan Josef |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134672 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:46:15.146Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134672 Maximising exploration and discovery in unknown large intricate 3D worlds Legg, Tristan Josef Engelbrecht, H. A. (Herman) Lorido-Botran, T. Stellenbosch University. Faculty of Engineering. Dept. of Electrical & Electronic Engineering. Three-dimensional display systems Virtual reality in engineering Autonomous robots -- Navigation Computer network architectures Thesis (MEng)--Stellenbosch University, 2025. Legg, T. J. 2025. Maximising Exploration and Discovery in Unknown Large Intricate 3D Worlds. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/d0dfcfda-3e72-4029-a5a5-24a9626484b3 ENGLISH ABSTRACT: This thesis investigates autonomous exploration in intricate 3D environments. The research addresses the growing need for automated content discovery in virtual environments, particularly content moderation on gaming platforms where manual review becomes increasingly impractical. We develop a custom reinforcement learning [1] environment featuring interconnected rooms and hidden areas, and implement an autonomous agent that explores using only RGB-D camera information and intrinsic motivation [2]. We evaluate three different neural network architectures: a no sequential model baseline, a gated recurrent unit model [3] and a decision transformer [4]. Through comprehensive experiments, we demonstrate that sequential models, particularly the decision transformer, significantly outperform no sequential model approaches, achieving up to 79.87% environment coverage in single-environment scenarios. Transfer learning experiments show promising results, with the decision transformer maintaining strong performance in both zero-shot (60.15%) and fine-tuned (73.54%) transfer scenarios. The research reveals important information about the sensitivity of the learning rate and the persistent challenge of discovering hidden rooms across all architectures. While attempts to create a general environment agent showed limited success (40.12% coverage), the strong performance in transfer learning scenarios suggests potential for developing efficient deployment strategies for autonomous exploration systems. AFRIKAANSE OPSOMMING: Hierdie tesis ondersoek outonome verkenning in komplekse 3D-omgewings. Die navorsing spreek die toenemende behoefte aan geoutomatiseerde ontdekking van inhoud in virtueleomgewings aan, veral vir die moderering van inhoud in video speletjie platforms. Die skaal van video speletjie platforms maak die moderering van inhoud deur regte persone onprakties. Ons ontwikkel ’n pasgemaakte versterkingsleeromgewing [1] wat bestaan uit kamers asook versteekte areas, en implementeer ’n outonome agent wat slegs RGB-D-kamera-inligting en intrinsieke motivering [2] gebruik om te verken. Ons evalueer drie verskillende argitekture vir neurale netwerke: ’n nie-sekwensi¨ele model as basislyn, asook twee nie-sekwensi¨ele modelle nl. ’n GRU-gebaseerde model [3] , en ’n Besluitnemer-transformator [4]. Deur omvattende eksperimente demonstreer ons dat sekwensi¨ele modelle, veral die Besluitnemer-transformator, aansienlik beter presteer as niesekwensi ¨ele benaderings, met tot 79.87% omgewingsdekking in enkel-omgewingscenarios. Ons ondersoek ook eksperimenteel die modelle se vermo¨e om oordragsleer te behartig, en verkry belowende resultate waar die Besluitnemer-transformator se goeie prestasie gehandhaaf word in beide die nul-skoot scenario (60.15%) en die scenario waar die modelparameters verfyn word (73.54%). Die navorsing onthul belangrike inligting oor die sensitiwiteit van die leertempo en die voortdurende uitdaging om versteekte areas te ontdek. Alhoewel ons deels daarin geslaag het was om ’n veralgemeende omgewingsagent te skep (slegs 40.12% dekking), dui die goeie prestasie in die oordragsleer-scenarios op die potensiaal vir die ontwikkeling van doeltreffende ontplooiingstrategie¨e vir outonome verkenningsisteme. Masters 2025-12-23T06:39:46Z 2025-12-23T06:39:46Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134672 en Stellenbosch University xiv, 114 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Three-dimensional display systems Virtual reality in engineering Autonomous robots -- Navigation Computer network architectures Legg, Tristan Josef Maximising exploration and discovery in unknown large intricate 3D worlds |
| title | Maximising exploration and discovery in unknown large intricate 3D worlds |
| title_full | Maximising exploration and discovery in unknown large intricate 3D worlds |
| title_fullStr | Maximising exploration and discovery in unknown large intricate 3D worlds |
| title_full_unstemmed | Maximising exploration and discovery in unknown large intricate 3D worlds |
| title_short | Maximising exploration and discovery in unknown large intricate 3D worlds |
| title_sort | maximising exploration and discovery in unknown large intricate 3d worlds |
| topic | Three-dimensional display systems Virtual reality in engineering Autonomous robots -- Navigation Computer network architectures |
| url | https://scholar.sun.ac.za/handle/10019.1/134672 |
| work_keys_str_mv | AT leggtristanjosef maximisingexplorationanddiscoveryinunknownlargeintricate3dworlds |