{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_53434"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_53434","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"Development, Processing, and Synthetic Augmentation of Comprehensive Broad-Domain Autonomous Vehicle Datasets and Optimizing Autonomous Vehicle Path Following under Localization Uncertainty","abstract":"Thesis (Ph.D.)--Michigan State University. Electrical Engineering - Doctor of Philosophy, 2026","abstract_html":"Thesis (Ph.D.)--Michigan State University. Electrical Engineering - Doctor of Philosophy, 2026","abstract_has_math":false,"creators":["Kent, Daniel"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Radha, Hayder","Cheng, Betty","Morris, Daniel","Nanzer, Jeffrey"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:53Z","subjects":["Electrical engineering","Computer engineering","Computer science"],"languages":["eng"],"rights":["In Copyright"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53434","local:Kent_grad.msu_0128D_22503"],"render_values":[{"text":"etd:53434","href":null,"code":true},{"text":"local:Kent_grad.msu_0128D_22503","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/9w72-rq45","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/9w72-rq45","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_53434","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Radha, Hayder","Cheng, Betty","Morris, Daniel","Nanzer, Jeffrey"]},{"key":"dc:creator","label":"Author","values":["Kent, Daniel"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Electrical engineering","Computer engineering","Computer science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53434","local:Kent_grad.msu_0128D_22503","https://doi.org/doi:10.25335/9w72-rq45"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Electrical Engineering - Doctor of Philosophy, 2026","Autonomous vehicle (AV) datasets serve as a critical cornerstone for the development of highly automated vehicles and advanced driver assistance systems (ADAS). The total number of multi-sensor samples of such datasets publicly available to scientists and engineers far exceeds one million, serving as an invaluable resource for countless autonomous perception algorithms that have been deployed in real world driving scenarios. However, the vast majority of these datasets are taken in nominal, fair weather conditions, which limits the utility of the data for the broad and diverse set of scenarios under which AVs may operate. Meanwhile, autonomous vehicle development continues to push beyond optimal conditions into more challenging and more realistic scenarios. Furthermore, very wide performance gaps have emerged when state-of-the-art perception methods trained under favorable conditions are tested under different and/or more challenging conditions. Simply collecting more datarequires excessive and repetitive deployment of complex and costly advanced sensor equipment, specialized labor, and major time and financial investments, and may present major safety challenges if data representing hazardous operating conditions is required. In this dissertation, we present two sets of contributions towards improving autonomous vehicle operations using data-centric techniques. First, we identified important connections between localization uncertainty and path following, and using a novel adaptive control algorithm we employed evolutionary algorithms to learn and optimize parameters for mitigating localization uncertainty in path following algorithms in simulation. We then transferred and tested these parameters to a real platform, demonstrating that the learned parameters operate well in practice. For our second set of contributions, we recognize that despite the immense amount of multimodal autonomous vehicle data available for public research, there remains measurable gaps in both data domain coverage and downstream perception algorithm performance. We address these problems with three complimentary works: first, we designed and pursued a comprehensive effort for the collection, processing, and analysis of a novel multimodal dataset deliberately tailored for the training, development and testing of multi-weather, multi-season autonomous vehicle perception systems. We show that this dataset exhibits statistically significant domain shifts across seasons and weather conditions. Next, we propose a unified software framework for processing autonomous vehicle perception data using a node-graph based architecture. This framework offers increased flexibility while ensuring high performance and software robustness by reducing redundant code and device-to-accelerator memory transfers. Finally, we evaluate the use of large generative foundation models to augment the size and domain coverage of our image datasets. To avoid the pertinent labeling problem, we invert the visual data generation problem: instead of creating labels for the synthetic data, we instead use labels as a control method for generating the synthetic samples themselves.Starting with pretrained generative networks, we demonstrate an architecture that produces realistic imagery that conforms to input semantic segmentation labels, which we demonstrate can successfully improve detection algorithm performance when these data are used to augment real datasets for during training. We also demonstrate two extensions of our work, one towards generating data for cross-weather domain adaptation research, and the other towards using pretrained segmentation and vision language models to replace human labeling for application on datasets without rich semantic segmentation labels. Our work on autonomous vehicle datasets and data processing demonstrates a vertically integrated process for collection, processing, and augmentation of these data for vehicle perception applications.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (130 p.) : ill."]},{"key":"dc:title","label":"Title","values":["Development, Processing, and Synthetic Augmentation of Comprehensive Broad-Domain Autonomous Vehicle Datasets and Optimizing Autonomous Vehicle Path Following under Localization Uncertainty"]}]}],"canonical_facts":{"dc:contributor":["Radha, Hayder","Cheng, Betty","Morris, Daniel","Nanzer, Jeffrey"],"dc:creator":["Kent, Daniel"],"dc:date":["2026"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Electrical Engineering - Doctor of Philosophy, 2026","Autonomous vehicle (AV) datasets serve as a critical cornerstone for the development of highly automated vehicles and advanced driver assistance systems (ADAS). The total number of multi-sensor samples of such datasets publicly available to scientists and engineers far exceeds one million, serving as an invaluable resource for countless autonomous perception algorithms that have been deployed in real world driving scenarios. However, the vast majority of these datasets are taken in nominal, fair weather conditions, which limits the utility of the data for the broad and diverse set of scenarios under which AVs may operate. Meanwhile, autonomous vehicle development continues to push beyond optimal conditions into more challenging and more realistic scenarios. Furthermore, very wide performance gaps have emerged when state-of-the-art perception methods trained under favorable conditions are tested under different and/or more challenging conditions. Simply collecting more datarequires excessive and repetitive deployment of complex and costly advanced sensor equipment, specialized labor, and major time and financial investments, and may present major safety challenges if data representing hazardous operating conditions is required. In this dissertation, we present two sets of contributions towards improving autonomous vehicle operations using data-centric techniques. First, we identified important connections between localization uncertainty and path following, and using a novel adaptive control algorithm we employed evolutionary algorithms to learn and optimize parameters for mitigating localization uncertainty in path following algorithms in simulation. We then transferred and tested these parameters to a real platform, demonstrating that the learned parameters operate well in practice. For our second set of contributions, we recognize that despite the immense amount of multimodal autonomous vehicle data available for public research, there remains measurable gaps in both data domain coverage and downstream perception algorithm performance. We address these problems with three complimentary works: first, we designed and pursued a comprehensive effort for the collection, processing, and analysis of a novel multimodal dataset deliberately tailored for the training, development and testing of multi-weather, multi-season autonomous vehicle perception systems. We show that this dataset exhibits statistically significant domain shifts across seasons and weather conditions. Next, we propose a unified software framework for processing autonomous vehicle perception data using a node-graph based architecture. This framework offers increased flexibility while ensuring high performance and software robustness by reducing redundant code and device-to-accelerator memory transfers. Finally, we evaluate the use of large generative foundation models to augment the size and domain coverage of our image datasets. To avoid the pertinent labeling problem, we invert the visual data generation problem: instead of creating labels for the synthetic data, we instead use labels as a control method for generating the synthetic samples themselves.Starting with pretrained generative networks, we demonstrate an architecture that produces realistic imagery that conforms to input semantic segmentation labels, which we demonstrate can successfully improve detection algorithm performance when these data are used to augment real datasets for during training. We also demonstrate two extensions of our work, one towards generating data for cross-weather domain adaptation research, and the other towards using pretrained segmentation and vision language models to replace human labeling for application on datasets without rich semantic segmentation labels. Our work on autonomous vehicle datasets and data processing demonstrates a vertically integrated process for collection, processing, and augmentation of these data for vehicle perception applications.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (130 p.) : ill."],"dc:identifier":["etd:53434","local:Kent_grad.msu_0128D_22503","https://doi.org/doi:10.25335/9w72-rq45"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["In Copyright"],"dc:subject":["Electrical engineering","Computer engineering","Computer science"],"dc:title":["Development, Processing, and Synthetic Augmentation of Comprehensive Broad-Domain Autonomous Vehicle Datasets and Optimizing Autonomous Vehicle Path Following under Localization Uncertainty"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:53Z"}