{"id":{"repo_id":"mtu","oai_identifier":"oai:digitalcommons.mtu.edu:etdr-3278"},"canonical_url":"https://search.ndltd.org/etd/mtu/oai:digitalcommons.mtu.edu:etdr-3278","repository":{"repo_id":"mtu","name":"Michigan Technological University","base_url":"https://digitalcommons.mtu.edu/do/oai/"},"display":{"title":"Robust State Estimation and Scene Understanding for Autonomous Ground Vehicles in Challenging Environments","abstract":"Autonomous ground vehicles require reliable state estimation and scene understanding to operate safely in challenging environments. Poor illumination, adverse weather, degraded visibility, snow-covered roads, and sensor noise can reduce the reliability of conventional localization and perception pipelines. Addressing these challenges requires sensing and algorithmic strategies that exploit the complementary strengths of different modalities. Radar is particularly attractive because of its resilience to lighting and weather variations, while inertial, camera, and lidar measurements provide additional motion, appearance, and geometric information. This dissertation investigates compact radar-based and sensor-fusion approaches for improving autonomous ground-vehicle state estimation and scene understanding under degraded sensing conditions. The first part of the dissertation focuses on radar odometry and SLAM for autonomous ground vehicles. It begins with a comprehensive survey of radar odometry methods, datasets, evaluation metrics, and research trends, with particular attention to sparse, dense, hybrid, sensor-fusion, and learning-based approaches. Building on this foundation, the dissertation presents FD-RIO, a fast dense radar-inertial odometry method that estimates scan-to-scan motion using cross-correlation and fuses radar-derived motion estimates with high-rate IMU measurements through a Kalman filter. FD-RIO is evaluated on public radar datasets and shown to provide competitive accuracy with practical CPU-only runtime. The work is then extended into FD-SLAM, a dense radar-inertial SLAM system that adds frequency-domain loop-closure detection, multi-stage loop verification, and pose graph optimization. FD-SLAM improves long-term trajectory consistency while preserving the dense image-like structure of scanning radar measurements. The second part of the dissertation addresses selected tasks in scene understanding for winter and adverse-weather driving. A CNN-based multimodal fusion model is developed for drivable path detection in snow using camera, lidar, and radar data. A broader camera-lidar fusion study then compares multiple early, intermediate, and late fusion CNN architectures for drivable-area detection in winter conditions while considering accuracy, runtime, model size, and synthetic sensor degradation. Finally, a compact CNN-based road-weather classifier is developed using a narrow grayscale image band to support weather-aware ADAS operation and sensor weighting. Overall, this dissertation demonstrates that compact multimodal algorithms and models can improve autonomous ground-vehicle operation in challenging environments.","abstract_html":"Autonomous ground vehicles require reliable state estimation and scene understanding to operate safely in challenging environments. Poor illumination, adverse weather, degraded visibility, snow-covered roads, and sensor noise can reduce the reliability of conventional localization and perception pipelines. Addressing these challenges requires sensing and algorithmic strategies that exploit the complementary strengths of different modalities. Radar is particularly attractive because of its resilience to lighting and weather variations, while inertial, camera, and lidar measurements provide additional motion, appearance, and geometric information. This dissertation investigates compact radar-based and sensor-fusion approaches for improving autonomous ground-vehicle state estimation and scene understanding under degraded sensing conditions. The first part of the dissertation focuses on radar odometry and SLAM for autonomous ground vehicles. It begins with a comprehensive survey of radar odometry methods, datasets, evaluation metrics, and research trends, with particular attention to sparse, dense, hybrid, sensor-fusion, and learning-based approaches. Building on this foundation, the dissertation presents FD-RIO, a fast dense radar-inertial odometry method that estimates scan-to-scan motion using cross-correlation and fuses radar-derived motion estimates with high-rate IMU measurements through a Kalman filter. FD-RIO is evaluated on public radar datasets and shown to provide competitive accuracy with practical CPU-only runtime. The work is then extended into FD-SLAM, a dense radar-inertial SLAM system that adds frequency-domain loop-closure detection, multi-stage loop verification, and pose graph optimization. FD-SLAM improves long-term trajectory consistency while preserving the dense image-like structure of scanning radar measurements. The second part of the dissertation addresses selected tasks in scene understanding for winter and adverse-weather driving. A CNN-based multimodal fusion model is developed for drivable path detection in snow using camera, lidar, and radar data. A broader camera-lidar fusion study then compares multiple early, intermediate, and late fusion CNN architectures for drivable-area detection in winter conditions while considering accuracy, runtime, model size, and synthetic sensor degradation. Finally, a compact CNN-based road-weather classifier is developed using a narrow grayscale image band to support weather-aware ADAS operation and sensor weighting. Overall, this dissertation demonstrates that compact multimodal algorithms and models can improve autonomous ground-vehicle operation in challenging environments.","abstract_has_math":false,"creators":["Abu-Alrub, Nader J"],"institution":null,"degree_name":"Doctor of Philosophy in Computational Science and Engineering (PhD)","degree_level":"Open Access Dissertation","degree_discipline":"Department of Applied Computing","degree_department":null,"school":null,"contributors":["Nathir A. Rawashdeh"],"advisors":[],"committee_chairs":[],"committee_members":["Ashraf I. Saleem","Jung Yun Bae","Sidike Paheding"],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T08:19:32Z","subjects":["State Estimation","Odometry","SLAM","Sensor Fusion","Radar","Autonomous Driving","Computational Engineering"],"languages":[],"rights":[],"rights_urls":["http://creativecommons.org/licenses/by-nc/4.0/"],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.mtu.edu/etdr/2157"],"render_values":[{"text":"https://digitalcommons.mtu.edu/etdr/2157","href":"https://digitalcommons.mtu.edu/etdr/2157","code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.37099/mtu.dc.etdr/2157","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/10.37099/mtu.dc.etdr/2157","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://digitalcommons.mtu.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.mtu.edu%3Aetdr-3278","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Nathir A. Rawashdeh"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Ashraf I. Saleem","Jung Yun Bae","Sidike Paheding"]},{"key":"dc:creator","label":"Author","values":["Abu-Alrub, Nader J"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.created","label":"Dc Date Created","values":["2026-01-01T08:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Department of Applied Computing"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Open Access Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy in Computational Science and Engineering (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["State Estimation","Odometry","SLAM","Sensor Fusion","Radar","Autonomous Driving","Computational Engineering"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights.license","label":"Dc Rights License","values":["http://creativecommons.org/licenses/by-nc/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["10.37099/mtu.dc.etdr/2157","https://digitalcommons.mtu.edu/etdr/2157"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Autonomous ground vehicles require reliable state estimation and scene understanding to operate safely in challenging environments. Poor illumination, adverse weather, degraded visibility, snow-covered roads, and sensor noise can reduce the reliability of conventional localization and perception pipelines. Addressing these challenges requires sensing and algorithmic strategies that exploit the complementary strengths of different modalities. Radar is particularly attractive because of its resilience to lighting and weather variations, while inertial, camera, and lidar measurements provide additional motion, appearance, and geometric information. This dissertation investigates compact radar-based and sensor-fusion approaches for improving autonomous ground-vehicle state estimation and scene understanding under degraded sensing conditions. The first part of the dissertation focuses on radar odometry and SLAM for autonomous ground vehicles. It begins with a comprehensive survey of radar odometry methods, datasets, evaluation metrics, and research trends, with particular attention to sparse, dense, hybrid, sensor-fusion, and learning-based approaches. Building on this foundation, the dissertation presents FD-RIO, a fast dense radar-inertial odometry method that estimates scan-to-scan motion using cross-correlation and fuses radar-derived motion estimates with high-rate IMU measurements through a Kalman filter. FD-RIO is evaluated on public radar datasets and shown to provide competitive accuracy with practical CPU-only runtime. The work is then extended into FD-SLAM, a dense radar-inertial SLAM system that adds frequency-domain loop-closure detection, multi-stage loop verification, and pose graph optimization. FD-SLAM improves long-term trajectory consistency while preserving the dense image-like structure of scanning radar measurements. The second part of the dissertation addresses selected tasks in scene understanding for winter and adverse-weather driving. A CNN-based multimodal fusion model is developed for drivable path detection in snow using camera, lidar, and radar data. A broader camera-lidar fusion study then compares multiple early, intermediate, and late fusion CNN architectures for drivable-area detection in winter conditions while considering accuracy, runtime, model size, and synthetic sensor degradation. Finally, a compact CNN-based road-weather classifier is developed using a narrow grayscale image band to support weather-aware ADAS operation and sensor weighting. Overall, this dissertation demonstrates that compact multimodal algorithms and models can improve autonomous ground-vehicle operation in challenging environments."]},{"key":"dc:title","label":"Title","values":["Robust State Estimation and Scene Understanding for Autonomous Ground Vehicles in Challenging Environments"]}]}],"canonical_facts":{"dc:contributor":["Nathir A. Rawashdeh"],"dc:contributor.committeemember":["Ashraf I. Saleem","Jung Yun Bae","Sidike Paheding"],"dc:creator":["Abu-Alrub, Nader J"],"dc:date.created":["2026-01-01T08:00:00Z"],"dc:description.abstract":["Autonomous ground vehicles require reliable state estimation and scene understanding to operate safely in challenging environments. Poor illumination, adverse weather, degraded visibility, snow-covered roads, and sensor noise can reduce the reliability of conventional localization and perception pipelines. Addressing these challenges requires sensing and algorithmic strategies that exploit the complementary strengths of different modalities. Radar is particularly attractive because of its resilience to lighting and weather variations, while inertial, camera, and lidar measurements provide additional motion, appearance, and geometric information. This dissertation investigates compact radar-based and sensor-fusion approaches for improving autonomous ground-vehicle state estimation and scene understanding under degraded sensing conditions. The first part of the dissertation focuses on radar odometry and SLAM for autonomous ground vehicles. It begins with a comprehensive survey of radar odometry methods, datasets, evaluation metrics, and research trends, with particular attention to sparse, dense, hybrid, sensor-fusion, and learning-based approaches. Building on this foundation, the dissertation presents FD-RIO, a fast dense radar-inertial odometry method that estimates scan-to-scan motion using cross-correlation and fuses radar-derived motion estimates with high-rate IMU measurements through a Kalman filter. FD-RIO is evaluated on public radar datasets and shown to provide competitive accuracy with practical CPU-only runtime. The work is then extended into FD-SLAM, a dense radar-inertial SLAM system that adds frequency-domain loop-closure detection, multi-stage loop verification, and pose graph optimization. FD-SLAM improves long-term trajectory consistency while preserving the dense image-like structure of scanning radar measurements. The second part of the dissertation addresses selected tasks in scene understanding for winter and adverse-weather driving. A CNN-based multimodal fusion model is developed for drivable path detection in snow using camera, lidar, and radar data. A broader camera-lidar fusion study then compares multiple early, intermediate, and late fusion CNN architectures for drivable-area detection in winter conditions while considering accuracy, runtime, model size, and synthetic sensor degradation. Finally, a compact CNN-based road-weather classifier is developed using a narrow grayscale image band to support weather-aware ADAS operation and sensor weighting. Overall, this dissertation demonstrates that compact multimodal algorithms and models can improve autonomous ground-vehicle operation in challenging environments."],"dc:identifier":["10.37099/mtu.dc.etdr/2157","https://digitalcommons.mtu.edu/etdr/2157"],"dc:rights.license":["http://creativecommons.org/licenses/by-nc/4.0/"],"dc:subject":["State Estimation","Odometry","SLAM","Sensor Fusion","Radar","Autonomous Driving","Computational Engineering"],"dc:title":["Robust State Estimation and Scene Understanding for Autonomous Ground Vehicles in Challenging Environments"],"thesis:degree_discipline":["Department of Applied Computing"],"thesis:degree_level":["Open Access Dissertation"],"thesis:degree_name":["Doctor of Philosophy in Computational Science and Engineering (PhD)"]},"updated_at":"2026-11-19T08:19:32Z"}