BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//TYPO3/NONSGML Calendarize//EN


BEGIN:VEVENT
UID:calendarize-defense-of-the-doctoral-dissertation-1
DTSTAMP:20251222T142914Z

DTSTART:20251223T140000Z
DTEND:20251222T220000Z
    
SUMMARY:Defense of the doctoral dissertation: Pēteris Račinskis
DESCRIPTION:Topic: "Perception and control in autonomous robotics using vision-language embeddings".\nScientific supervisor: Dr. sc. comp. Modris Greitāns.\nAbstract:\nThis doctoral thesis presents advancements in perception and control for autonomous robotics\, integrating vision-language semantics to address challenges in unstructured environments. The research focuses on developing semantic Simultaneous Localization and Mapping (SLAM) systems. Key contributions include the application of visual open-set semantics within large-scale LiDAR maps for terrain segmentation in autonomous robot navigation\, and the integration of Large Language Model (LLM)-based high-level planning with these semantic maps in a tabletop object stacking scenario. The study aims to fuse various sensor data — LiDAR\, depth camera\, color images\, and GNSS — into a unified\, language-grounded environmental model\, facilitating human-robot interaction through voice or text. Experimental results also validate novel data set collection methodologies\, demonstrating that visual fiducial markers positioned using surveying tools can substitute for GNSS-INS-based ground truth position and orientation estimates in scenarios when these are unavailable or inadmissible. The results of this work have been published across three first-author articles in indexed journals\, two international conferences and a project report technology description publication. This work introduces two publicly available resources — the EDI-SLAM dataset and the open-source SLAMVDB software package. Two demonstrators developed in this dissertation — a semantic perception system and static manipulation-focused robot control architecture — have been positively assessed by industry representatives.\nReviewers:\n1) Dr. sc. comp. Inguna Skadiņa\, University of Latvia\;\n2) Dr. Vytautas Bučinskas\, Vilnius Gediminas Technical University\, Lithuania\;\n3) Dr. Po Ting Lin\, National Taiwan University of Science and Technology\, Taiwan.\nThe doctoral thesis is available at the University of Latvia Library\, Raiņa bulvāris 19.\nParticipation in the remote session is subject to prior registration by writing to sintija.silina@lu.lv by 20th of January.
X-ALT-DESC;FMTTYPE=text/html:<p class="text-justify">Topic: <strong>"Perception and control in autonomous robotics using vision-language embeddings"</strong>.</p>\n<p class="text-justify">Scientific supervisor: <i>Dr. sc. comp</i>. <strong>Modris Greitāns</strong>.</p>\n<p class="text-justify"><strong>Abstract:</strong></p>\n<p class="text-justify">This doctoral thesis presents advancements in perception and control for autonomous robotics\, integrating vision-language semantics to address challenges in unstructured environments. The research focuses on developing semantic Simultaneous Localization and Mapping (SLAM) systems. Key contributions include the application of visual open-set semantics within large-scale LiDAR maps for terrain segmentation in autonomous robot navigation\, and the integration of Large Language Model (LLM)-based high-level planning with these semantic maps in a tabletop object stacking scenario. The study aims to fuse various sensor data — LiDAR\, depth camera\, color images\, and GNSS — into a unified\, language-grounded environmental model\, facilitating human-robot interaction through voice or text. Experimental results also validate novel data set collection methodologies\, demonstrating that visual fiducial markers positioned using surveying tools can substitute for GNSS-INS-based ground truth position and orientation estimates in scenarios when these are unavailable or inadmissible. The results of this work have been published across three first-author articles in indexed journals\, two international conferences and a project report technology description publication. This work introduces two publicly available resources — the EDI-SLAM dataset and the open-source SLAMVDB software package. Two demonstrators developed in this dissertation — a semantic perception system and static manipulation-focused robot control architecture — have been positively assessed by industry representatives.</p>\n<p class="text-justify">Reviewers:</p>\n<p class="text-justify"><strong>1)</strong> <i>Dr. sc. comp.</i><strong> Inguna Skadiņa</strong>\, University of Latvia\;</p>\n<p class="text-justify"><strong>2) </strong><i>Dr. </i><strong>Vytautas Bučinskas</strong>\, Vilnius Gediminas Technical University\, Lithuania\;</p>\n<p class="text-justify"><strong>3) </strong><i>Dr.</i> <strong>Po Ting Lin</strong>\, National Taiwan University of Science and Technology\, Taiwan.</p>\n<p class="text-justify">The doctoral thesis is available at the University of Latvia Library\, Raiņa bulvāris 19.</p>\n<p class="text-justify">Participation in the remote session is subject to prior registration by writing to <a href="mailto:sintija.silina@lu.lv">sintija.silina@lu.lv</a> <strong>by 20th of January</strong>.</p>
LOCATION:
END:VEVENT

END:VCALENDAR