{"id":{"repo_id":"vt","oai_identifier":"oai:vtechworks.lib.vt.edu:10919/143496"},"canonical_url":"https://search.dev.ndltd.org/etd/vt/oai:vtechworks.lib.vt.edu:10919/143496","repository":{"repo_id":"vt","name":"Virginia Tech","base_url":"https://vtechworks.lib.vt.edu/oai/request"},"display":{"title":"Beyond the Visible: Transforming AR/VR Software Engineering Through Empirical Discovery, Metamorphic Testing, and Multi-Agent Intelligence","abstract":"Augmented Reality (AR) technologies are reshaping interactive computing, yet software engineering support for immersive applications has not kept pace with their growing complexity. Because AR systems produce outputs that depend on dynamic environments, sensor input, and real-world context, the expected outputs are often impossible to determine in advance -a challenge known as the oracle problem -making conventional testing approaches insufficient. To understand how this affects development in practice, we first conduct an empirical study of 48 core AR developers, identifying recurring challenges across the software development lifecycle, with especially persistent difficulties in testing, bug reproduction, and handling environmental variability. Building on these findings, we introduce a metamorphic testing framework for mobile AR that validates context-dependent behavior through domain- specific behavioral relations rather than fixed expected outputs, and we show that the same oracle-avoiding abstraction extends to LLM-generated code, where property-based testing exposes correctness failures that unit tests miss. We then investigate how LLMs themselves can support correctness reasoning in this setting: first, through a stability-driven evalua- tion framework showing that multi-agent debate improves the reliability of metamorphic relation identification, and then through a repository-aware generation framework that uses hierarchical code context and deliberative refinement to discover new AR-specific relations directly from software repositories. Taken together, this work advances a correctness-oriented foundation for AR software engineering, grounded in developer practice, expressed through behavioral abstractions, and supported by structured LLM-based reasoning.","abstract_html":"Augmented Reality (AR) technologies are reshaping interactive computing, yet software engineering support for immersive applications has not kept pace with their growing complexity. Because AR systems produce outputs that depend on dynamic environments, sensor input, and real-world context, the expected outputs are often impossible to determine in advance -a challenge known as the oracle problem -making conventional testing approaches insufficient. To understand how this affects development in practice, we first conduct an empirical study of 48 core AR developers, identifying recurring challenges across the software development lifecycle, with especially persistent difficulties in testing, bug reproduction, and handling environmental variability. Building on these findings, we introduce a metamorphic testing framework for mobile AR that validates context-dependent behavior through domain- specific behavioral relations rather than fixed expected outputs, and we show that the same oracle-avoiding abstraction extends to LLM-generated code, where property-based testing exposes correctness failures that unit tests miss. We then investigate how LLMs themselves can support correctness reasoning in this setting: first, through a stability-driven evalua- tion framework showing that multi-agent debate improves the reliability of metamorphic relation identification, and then through a repository-aware generation framework that uses hierarchical code context and deliberative refinement to discover new AR-specific relations directly from software repositories. Taken together, this work advances a correctness-oriented foundation for AR software engineering, grounded in developer practice, expressed through behavioral abstractions, and supported by structured LLM-based reasoning.","abstract_has_math":false,"creators":["Bose, Dibyendu Brinto"],"institution":"Virginia Tech","degree_name":"Doctor of Philosophy","degree_level":"doctoral","degree_discipline":"Computer Science & Applications","degree_department":"Computer Science and#38; Applications","school":null,"contributors":[],"advisors":[],"committee_chairs":["Brown, Dwayne Christian"],"committee_members":["Gracanin, Denis","Meng, Na","David-John, Brendan Matthew","Wang, Xiaoyin"],"year":2026,"date_issued":"2026-06-24","date_published":"2026-06-24","updated_at":"2026-07-24T05:56:26Z","subjects":["AR development","Metamorphic testing","Multi Agent System"],"languages":["en"],"rights":["In Copyright"],"rights_urls":["http://rightsstatements.org/vocab/InC/1.0/"],"identifier_entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:46154"],"render_values":[{"text":"vt_gsexam:46154","href":null,"code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/10919/143496","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.committeechair","label":"Committee Chair","values":["Brown, Dwayne Christian"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Gracanin, Denis","Meng, Na","David-John, Brendan Matthew","Wang, Xiaoyin"]},{"key":"dc:contributor.department","label":"Department","values":["Computer Science and#38; Applications"]},{"key":"dc:creator","label":"Author","values":["Bose, Dibyendu Brinto"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-06-25T08:00:33Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-06-25T08:00:33Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-06-24"]},{"key":"dc:publisher","label":"Institution","values":["Virginia Tech"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science & Applications"]},{"key":"thesis:degree_level","label":"Degree Level","values":["doctoral"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Virginia Polytechnic Institute and State University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["AR development","Metamorphic testing","Multi Agent System"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://rightsstatements.org/vocab/InC/1.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:46154"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10919/143496"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Augmented Reality (AR) technologies are reshaping interactive computing, yet software engineering support for immersive applications has not kept pace with their growing complexity. Because AR systems produce outputs that depend on dynamic environments, sensor input, and real-world context, the expected outputs are often impossible to determine in advance -a challenge known as the oracle problem -making conventional testing approaches insufficient. To understand how this affects development in practice, we first conduct an empirical study of 48 core AR developers, identifying recurring challenges across the software development lifecycle, with especially persistent difficulties in testing, bug reproduction, and handling environmental variability. Building on these findings, we introduce a metamorphic testing framework for mobile AR that validates context-dependent behavior through domain- specific behavioral relations rather than fixed expected outputs, and we show that the same oracle-avoiding abstraction extends to LLM-generated code, where property-based testing exposes correctness failures that unit tests miss. We then investigate how LLMs themselves can support correctness reasoning in this setting: first, through a stability-driven evalua- tion framework showing that multi-agent debate improves the reliability of metamorphic relation identification, and then through a repository-aware generation framework that uses hierarchical code context and deliberative refinement to discover new AR-specific relations directly from software repositories. Taken together, this work advances a correctness-oriented foundation for AR software engineering, grounded in developer practice, expressed through behavioral abstractions, and supported by structured LLM-based reasoning."]},{"key":"dc:description.abstractgeneral","label":"General Abstract","values":["Augmented Reality (AR) software is becoming more common in areas such as healthcare, education, entertainment, and training. However, building reliable AR applications is difficult because these systems must work correctly while responding to changing physical environ- ments, user movement, sensors, and complex 3D interactions. This makes them much harder to test than traditional software, where the correct output for a given input is usually known in advance. My research studies this problem from several angles. First, I investigated the real-world challenges faced by professional AR developers and found that many of their most persistent difficulties arise during testing and debugging—particularly because the expected behavior of an AR application often cannot be determined in advance. Building on these findings, I developed a testing approach for mobile AR applications that checks whether software continues to behave correctly when inputs or environmental conditions are changed in meaningful ways, rather than requiring a known correct answer. I then explored how modern AI tools, especially large language models (LLMs), can help identify these kinds of behavioral rules more reliably — including studying when AI models are consistent in their reasoning, how multiple AI agents can debate to improve decisions, and how code repository context can help generate better testing knowledge automatically. I also applied this same idea — checking behavior under change rather than checking against a fixed answer — to code written by AI systems, showing that this reveals correctness problems that standard testing methods often miss. Together, this work contributes new ways to understand, test, and reason about correctness in immersive software, showing how developer-grounded re- search, advanced testing methods, and AI-based reasoning can be combined to make AR applications more dependable in practice."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Doctor of Philosophy"]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["ETD"]},{"key":"dc:title","label":"Title","values":["Beyond the Visible: Transforming AR/VR Software Engineering Through Empirical Discovery, Metamorphic Testing, and Multi-Agent Intelligence"]}]}],"canonical_facts":{"dc:contributor.committeechair":["Brown, Dwayne Christian"],"dc:contributor.committeemember":["Gracanin, Denis","Meng, Na","David-John, Brendan Matthew","Wang, Xiaoyin"],"dc:contributor.department":["Computer Science and#38; Applications"],"dc:creator":["Bose, Dibyendu Brinto"],"dc:date.accessioned":["2026-06-25T08:00:33Z"],"dc:date.available":["2026-06-25T08:00:33Z"],"dc:date.issued":["2026-06-24"],"dc:description.abstract":["Augmented Reality (AR) technologies are reshaping interactive computing, yet software engineering support for immersive applications has not kept pace with their growing complexity. Because AR systems produce outputs that depend on dynamic environments, sensor input, and real-world context, the expected outputs are often impossible to determine in advance -a challenge known as the oracle problem -making conventional testing approaches insufficient. To understand how this affects development in practice, we first conduct an empirical study of 48 core AR developers, identifying recurring challenges across the software development lifecycle, with especially persistent difficulties in testing, bug reproduction, and handling environmental variability. Building on these findings, we introduce a metamorphic testing framework for mobile AR that validates context-dependent behavior through domain- specific behavioral relations rather than fixed expected outputs, and we show that the same oracle-avoiding abstraction extends to LLM-generated code, where property-based testing exposes correctness failures that unit tests miss. We then investigate how LLMs themselves can support correctness reasoning in this setting: first, through a stability-driven evalua- tion framework showing that multi-agent debate improves the reliability of metamorphic relation identification, and then through a repository-aware generation framework that uses hierarchical code context and deliberative refinement to discover new AR-specific relations directly from software repositories. Taken together, this work advances a correctness-oriented foundation for AR software engineering, grounded in developer practice, expressed through behavioral abstractions, and supported by structured LLM-based reasoning."],"dc:description.abstractgeneral":["Augmented Reality (AR) software is becoming more common in areas such as healthcare, education, entertainment, and training. However, building reliable AR applications is difficult because these systems must work correctly while responding to changing physical environ- ments, user movement, sensors, and complex 3D interactions. This makes them much harder to test than traditional software, where the correct output for a given input is usually known in advance. My research studies this problem from several angles. First, I investigated the real-world challenges faced by professional AR developers and found that many of their most persistent difficulties arise during testing and debugging—particularly because the expected behavior of an AR application often cannot be determined in advance. Building on these findings, I developed a testing approach for mobile AR applications that checks whether software continues to behave correctly when inputs or environmental conditions are changed in meaningful ways, rather than requiring a known correct answer. I then explored how modern AI tools, especially large language models (LLMs), can help identify these kinds of behavioral rules more reliably — including studying when AI models are consistent in their reasoning, how multiple AI agents can debate to improve decisions, and how code repository context can help generate better testing knowledge automatically. I also applied this same idea — checking behavior under change rather than checking against a fixed answer — to code written by AI systems, showing that this reveals correctness problems that standard testing methods often miss. Together, this work contributes new ways to understand, test, and reason about correctness in immersive software, showing how developer-grounded re- search, advanced testing methods, and AI-based reasoning can be combined to make AR applications more dependable in practice."],"dc:description.degree":["Doctor of Philosophy"],"dc:format.medium":["ETD"],"dc:identifier.other":["vt_gsexam:46154"],"dc:identifier.uri":["https://hdl.handle.net/10919/143496"],"dc:language.iso":["en"],"dc:publisher":["Virginia Tech"],"dc:rights":["In Copyright"],"dc:rights.uri":["http://rightsstatements.org/vocab/InC/1.0/"],"dc:subject":["AR development","Metamorphic testing","Multi Agent System"],"dc:title":["Beyond the Visible: Transforming AR/VR Software Engineering Through Empirical Discovery, Metamorphic Testing, and Multi-Agent Intelligence"],"dc:type":["Dissertation"],"thesis:degree_discipline":["Computer Science & Applications"],"thesis:degree_level":["doctoral"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["Virginia Polytechnic Institute and State University"]},"updated_at":"2026-07-24T05:56:26Z"}