BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Robotics Institute Carnegie Mellon University - ECPv6.15.12.1//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Robotics Institute Carnegie Mellon University
X-ORIGINAL-URL:https://www.ri.cmu.edu
X-WR-CALDESC:Events for Robotics Institute Carnegie Mellon University
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20250309T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20251102T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20260308T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20261101T060000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:20270314T070000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:20271107T060000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260609T140000
DTEND;TZID=America/New_York:20260609T153000
DTSTAMP:20260708T225105
CREATED:20260601T172413Z
LAST-MODIFIED:20260601T172413Z
UID:151404-1781013600-1781019000@www.ri.cmu.edu
SUMMARY:Scaling Sim2Real Learning for Robot Manipulation
DESCRIPTION:Abstract: \nRecent progress in robot learning has led to impressively capable manipulation systems. Much of the progress has come from scaling up human demonstrations; however\, collecting such data through manual teleoperation is slow\, costly\, and hard to scale.\n\nPhysics-based Simulation offers a scalable\, safe\, and efficient alternative for generating large demonstration datasets. However\, some core challenges limit the full potential of this approach: the heavy manual effort required to design simulation tasks and rewards\, the gap between simulation and reality\, and the difficulty of learning policies that generalize well when trained on large\, diverse simulation datasets.\n\nIn this thesis\, we tackle these challenges through scalable\, generalizable\, and adaptive robot learning. First\, I will show how structured policy representations can enable simulation trained policies to achieve broad generalization in the real world and serve as a strong prior for downstream fine-tuning. Second\, I will introduce the Generative Simulation pipeline for automatic generation of large-scale simulation datasets with minimal human efforts. Finally\, I will discuss some novel algorithms for efficient adaptation of simulation-trained policies to the real world. Together\, these efforts move us toward robots that can learn broadly\, adapt quickly\, and assist people in real homes and workplaces.\n\nCommittee:\nZackory Erickson (co-chair)\nDavid Held (co-chair)\nKaterina Fragkiadaki\nChuang Gan (UMass Amherst and MIT-IBM Watson AI Lab)\nDieter Fox (University of Washington and Ai2)\nLink to draft thesis
URL:https://www.ri.cmu.edu/event/scaling-sim2real-learning-for-robot-manipulation/
LOCATION:1305 Newell Simon Hall
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260610T100000
DTEND;TZID=America/New_York:20260610T113000
DTSTAMP:20260708T225105
CREATED:20260601T184629Z
LAST-MODIFIED:20260601T184629Z
UID:151406-1781085600-1781091000@www.ri.cmu.edu
SUMMARY:Annotation-Free Learning for Mobile Robot Navigation in Unstructured Environments
DESCRIPTION:Abstract: Navigation in unstructured environments is a capability critical to many robotics applications such as forestry\, construction\, disaster response and defense. In these domains\, robots have the potential to eliminate much of the dull\, dirty and/or dangerous work that is currently performed by humans. Unfortunately\, these environments pose a unique set of challenges for navigation not present in controlled settings such as warehouses and highway driving.\n\nWe argue that the key challenge in deploying robots in unstructured environments is the difficulty of interpreting high-dimensional streams of multi-modal perception data\, and translating them into robust\, expressive representations for downstream path planning and control. This problem is exacerbated by the fact that unstructured environments are highly uncertain and lack a clean mapping from semantics to navigation decisions (i.e. traversability). \nWhile there have been notable successes in deploying robots in unstructured environments\, such systems are highly tuned to a given robot platform and environment\, and are the result of months to years of dedicated effort from a highly skilled team of engineers. As such\, deployment of these systems to new platforms or environments requires extensive tuning of perception-planning interfaces\, labeling of semantic images\, etc. Unfortunately\, this makes engineering expertise the bottleneck in the widespread adoption of mobile robots. \nI will present my work in designing a learning-based perception and traversability system for high-speed off-road driving from camera and LiDAR data. Importantly\, this approach is annotation-free\, enabling performance matching or exceeding current SoTA in off-road driving from a relatively small amount of tele-operated demonstrations. \n\nCommittee:\nSebastian Scherer\, chair\nAaron Johnson\nDrew Bagnell\nWenshan Wang\nDavid D. Fan (Field AI)\n \n\nThesis Draft
URL:https://www.ri.cmu.edu/event/annotation-free-learning-for-mobile-robot-navigation-in-unstructured-environments/
LOCATION:3305 Newell-Simon Hall
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260611T100000
DTEND;TZID=America/New_York:20260611T113000
DTSTAMP:20260708T225105
CREATED:20260528T204739Z
LAST-MODIFIED:20260529T163943Z
UID:151395-1781172000-1781177400@www.ri.cmu.edu
SUMMARY:Autonomous Crop Manipulation: From Model-Based Reasoning to Learned Interaction
DESCRIPTION:Abstract: Robots that manipulate crops must contend with plants that occlude themselves\, deform under contact\, and make the manipulator contact structures it neither targets nor sees in advance. This thesis argues that autonomous manipulation of crops in unstructured agricultural environments is best advanced not by choosing between model-based and learned approaches\, but by integrating them where each is strongest: model-based components provide explicit geometric priors and safety constraints for contact-rich interaction\, while learned components capture plant dynamics and visuomotor behaviors that resist hand-engineering. \nFour technical contributions develop this argument. The first is a model-based skeletonization method that recovers tree branch structure from RGB-D views using an occlusion-likelihood field and a min-cost path search to reason explicitly about unobserved geometry. The second is a graph neural network trained on mass-spring-damper simulations that predicts how a tree deforms under push. The third is an end-to-end diffusion policy trained on demonstrations collected with a handheld shear-gripper that drives pepper harvesting in an unprotected outdoor field. The fourth is a reactive safety layer that wraps a learned visuomotor policy with a control barrier function quadratic program whose constraints are drawn from an occupancy map updated online from contact estimates\, enabling safe reaching through cluttered canopies. \nTaken together\, the four contributions trace a path from offline geometric reasoning about static plants to safe contact-rich interaction in clutter\, showing that robust agricultural manipulation emerges not from end-to-end learning alone\, but from a synthesis in which model-based safety enables reliable learned interaction. \nCommittee:\nGeorge Kantor (chair)\nOliver Kroemer\nDave Held\nMaren Bennewitz (University of Bonn) \nThesis Draft
URL:https://www.ri.cmu.edu/event/autonomous-crop-manipulation-from-model-based-reasoning-to-learned-interaction/
LOCATION:3305 Newell-Simon Hall
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260615T110000
DTEND;TZID=America/New_York:20260615T123000
DTSTAMP:20260708T225105
CREATED:20260604T191624Z
LAST-MODIFIED:20260604T191624Z
UID:151486-1781521200-1781526600@www.ri.cmu.edu
SUMMARY:Integrating Structured Knowledge for State and Geometry Estimation
DESCRIPTION:Abstract:\n\nReliable state and geometry estimation from limited observations is a fundamental challenge in robotics and perception. Observations are often noisy\, partial\, or ambiguous\, making estimation ill-posed without additional structure. This thesis argues that robust estimation in these regimes is enabled by integrating structured knowledge into the inference process. \nEstimation can be viewed as inferring latent state or geometry by combining three complementary forms of structure: constraints that restrict the feasible solution space\, physical models that describe how observations are generated\, and priors that regularize underdetermined solutions. \nFirst\, constraints provide structure in the solution space. We develop methods for imposing hard constraints in real-time estimation through incremental optimization\, and for learning well-conditioned measurement noise models that shape the optimization landscape. We also study constraints learned from demonstrations and analyze when they can serve as general state constraints. \nSecond\, physical models provide structure in the observation process. We study settings where sensing physics fundamentally limits observability\, and show how incorporating physics-based forward models into neural rendering enables accurate 3D reconstruction and resolves ambiguities in acoustic and multimodal sensing. \nFinally\, priors provide structure when estimation remains underdetermined even after modeling constraints and physics. We demonstrate how priors learned from data-rich visual domains can be transferred to data-scarce sensing modalities\, enabling pose-free acoustic 3D reconstruction and tactile 3D reconstruction from sparse robot touches. \nTogether\, these contributions provide a unified perspective on estimation as the integration of constraints\, physical models\, and priors. \nThesis Committee Members:\n\nMichael Kaess (Chair)  \nIoannis Gkioulekas\nShubham Tulsiani\nNikolay Atanasov (University of California – San Diego)\n\n \nURL link to thesis: Thesis Document
URL:https://www.ri.cmu.edu/event/integrating-structured-knowledge-for-state-and-geometry-estimation/
LOCATION:Newell-Simon Hall 4305
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260615T130000
DTEND;TZID=America/New_York:20260615T143000
DTSTAMP:20260708T225105
CREATED:20260608T193629Z
LAST-MODIFIED:20260608T193629Z
UID:151519-1781528400-1781533800@www.ri.cmu.edu
SUMMARY:Physical Process-Informed Mapping for Robotic Exploration
DESCRIPTION:Abstract:\nMobile robots used for information gathering tasks rely on dense\, predictive mapping of large-scale regions to determine where to take measurements. Current approaches to mapping commonly rely on Gaussian process regression to spatially correlate data\, extrapolate from sparse samples\, and estimate uncertainty. However\, these approaches do not incorporate meaningful information about physical processes that contribute to the scientific observations being recorded\, instead focusing on only predicting measurements or relying on black box expert-based systems to incorporate physically meaningful information.\n\nIn this work\, we introduce physical process-informed mapping\, a framework for integrating scientific knowledge about physical processes into mapping for mobile exploration robots. This approach provides two advantages for exploration. First\, the hierarchical statistical model underlying the mapping framework is capable of providing predictions and uncertainty estimates for both the sensor measurements and the latent natural phenomena driving these measurements in a dense\, probabilistic manner. Second\, various components of the mapping approach provide methods to measure and adjust the importance of models of scientific knowledge\, enabling learning about the conditions under which these models are more or less relevant to the latent phenomenon. Taken together\, these methods improve the interpretability of mapping for autonomous scientific discovery\, enabling future robotic explorers to take advantage of a more in-depth representation of information\, and can provide increases in the predictive accuracy of mapping.\n\nThesis Committee Members:\nDavid Wettergreen (Chair)\nWennie Tabib\nMikael Kuusela\nTerrence Fong (NASA Ames Research Center)\n\n\nThesis Draft
URL:https://www.ri.cmu.edu/event/physical-process-informed-mapping-for-robotic-exploration-2/
LOCATION:3305 Newell-Simon Hall
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260616T140000
DTEND;TZID=America/New_York:20260616T153000
DTSTAMP:20260708T225105
CREATED:20260610T142430Z
LAST-MODIFIED:20260610T142621Z
UID:151566-1781618400-1781623800@www.ri.cmu.edu
SUMMARY:Scalable Oversight Across Generative Visual AI: Toward Visual Storytelling for Everyone
DESCRIPTION:Abstract: \nGenerative visual AI has advanced by scaling data and compute\, but its next bottleneck is oversight: the expert signals that evaluate\, reward\, and teach models what “good” looks like. Providing such oversight is increasingly difficult because foundation vision-language models now match or surpass most humans at the skills being judged. \nThis thesis develops scalable oversight for generative visual AI\, turning limited expert judgment into abundant and reliable evaluation\, reward\, and training signals. First\, I present VQAScore and GenAI-Bench\, evaluation standards for text-to-visual generation adopted by 100+ frontier labs such as DeepMind\, and show that a stronger judge directly improves generation through inference-time scaling. Second\, I extend oversight to video\, where precise language must be built rather than collected: CameraBench (NeurIPS’25 Spotlight) defines a structured cinematic vocabulary with professional creators\, and CHAI (CVPR’26 Highlight) introduces critique-based human-AI oversight in which experts critique model drafts rather than write from scratch. One recipe of specification\, oversight\, and post-training simultaneously improves captioning\, reward modeling\, and video generation\, enabling a small open model to surpass proprietary models like GPT and Gemini. Finally\, I present Moodio\, a deployed AI film studio with hundreds of daily active users\, whose cinematic video retrieval outperforms SOTA embedding models and measurably helps creators generate more compelling videos. I conclude with ongoing work on personalized oversight for end-to-end creative agents\, toward visual storytelling for everyone. \nThesis Committee Members: \n\nDeva Ramanan (Chair)\nDeepak Pathak\nGraham Neubig\nAli Farhadi (University of Washington and Microsoft)
URL:https://www.ri.cmu.edu/event/scalable-oversight-across-generative-visual-ai-toward-visual-storytelling-for-everyone/
LOCATION:Newell-Simon Hall 4305
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260618T110000
DTEND;TZID=America/New_York:20260618T123000
DTSTAMP:20260708T225105
CREATED:20260608T191139Z
LAST-MODIFIED:20260608T191612Z
UID:151508-1781780400-1781785800@www.ri.cmu.edu
SUMMARY:Integrating Learning and Collaboration for Human-Robot Alignment
DESCRIPTION:Abstract: The alignment problem for robots considers how robots can learn to behave in accordance with human values. Today\, robot learning paradigms enable humans to provide data (e.g.\, preference labels or demonstrations)\, which the robot uses to update its behavior (e.g.\, reward model or policy) to better align with human intentions.  However\, the current paradigm requires the human to constantly supervise\, provide new feedback\, and—more fundamentally—perfectly understand where the robot is misaligned. Even if the robot eventually learns a perfect model of how the human wanted it to behave\, the overall human-robot interaction during alignment could have been demanding\, confusing\, or arduous for the person.\n\nThis dissertation argues that alignment should not be viewed solely as a problem of learning the correct task behavior\, but also as a problem of supporting humans throughout the alignment process itself. We propose that the learning process should be fundamentally collaborative\, in which robots actively participate in alignment through introspection\, communication\, uncertainty estimation\, sharing control\, and interaction planning.Our early works investigate how robots can learn and adapt collaborative strategies from interaction\, including adapting to partner preferences and proactively communicating robot capabilities to improve coordination. We then introduce model-agnostic uncertainty quantification methods for interactive robot learning using online conformal prediction\, enabling robots to calibrate deployment-time uncertainty from intermittent human feedback. Building on these capabilities\, we develop collaborative active learning methods in which robots request assistance when uncertain\, calibrate intervention policies using human feedback\, strategically allocate human effort across multitask learning settings\, and negotiate control handoffs through shared-control interaction mechanisms. Finally\, we study the relationship between uncertainty and human intervention behavior\, showing that robot uncertainty and human judgments of intervention necessity are only partially aligned\, motivating methods that explicitly calibrate robot assistance requests to human preferences. The outcomes are robots that know when they do not know\, communicate uncertainty\, request help strategically\, support human teachers during learning\, and adapt interaction itself as part of the alignment process.Thesis Committee Members: \n\nHenny Admoni (Chair)\nReid Simmons (Chair)\nAndrea Bajcsy\nAnirudha Majumdar (Princeton)\n\n\n\nURL link to thesis
URL:https://www.ri.cmu.edu/event/integrating-learning-and-collaboration-for-human-robot-alignment/
LOCATION:Newell-Simon Hall 4305
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260623T120000
DTEND;TZID=America/New_York:20260623T133000
DTSTAMP:20260708T225105
CREATED:20260610T141042Z
LAST-MODIFIED:20260610T141042Z
UID:151561-1782216000-1782221400@www.ri.cmu.edu
SUMMARY:From Margin to Center: Designing Inclusive & Equitable Service Robots with Disabled Adults
DESCRIPTION:Abstract\n\n\nService robots – autonomous systems that perform personal and professional tasks – have become a common sight in the Global North. In Human-robot interaction (HRI)\, researchers rarely consider the design implications of service robots for people with disabilities (PwDs) beyond controlled assistive contexts\, such as private homes and assisted living facilities. Nevertheless\, the purview of PwDs transcends assistive contexts: PwDs know from lived experience that autonomous innovations can cause real harm to members of the disabled community when deployed. Therefore\, considering the lived experiences of PwDs\, many of whom exist at the social margins of society\, can enable the design of inclusive service robots. Even so\, robotics design practice can be exclusionary for people who are neither roboticists nor robot design practitioners.\n\nParticipatory Design (PD)\, or cooperative design (“co-design”)\, an equitable approach aimed at shifting power relations and integrating community interests in design processes outcomes\, is becoming commonplace in HRI. However\, HRI design practitioners who work with PwDs seldom interrogate whose lived experiences are represented in the design process – commonly privileging the experiences of professionals\, such as technical experts and medical practitioners; this further encodes normative biases about how PwDs may want robots to behave around them or the types of support they may desire. Instead\, HRI design practitioners should center on the experience-as-knowledge of members of the disabled community within co-design processes.\n\nIn this thesis\, I present the lived experiences as a PD methodological framework for designing service robots with disabled adults. I demonstrate the application of this framework by co-designing service robots with disabled adults across three domains: assistive navigation\, last-mile delivery\, and mental well-being. I highlight the principles of the lived experience framework\, which reflects an onto-epistemological commitment to re-position the experience-as-expertise of PwDs\, including shifting power towards PwDs through accessible design communication; highlighting power relations\, embracing tensions\, and fostering meaningful engagements with PwDs with intersecting social marginalization; and encouraging reflexivity in the co-design process. Furthermore\, I offer design guidelines and socio-technical considerations for developers and design practitioners who wish to create inclusive service robots and equitable human-robot experiences. This work is a purposeful demonstration of the meaningful contributions made by democratizing the design of enabling\, yet disruptive\, autonomous systems and a call-to-action for HRI practitioners who seek renewed commitments to PD as a transformative framework.\n\n\nThesis Committee Members:\nAaron Steinfeld\, Chair\nJean Oh\, RI\nPatrick Carrington\, HCII\nCynthia Bennett\, Google\n\nURL link to thesis
URL:https://www.ri.cmu.edu/event/from-margin-to-center-designing-inclusive-equitable-service-robots-with-disabled-adults/
LOCATION:Newell-Simon Hall 4305
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260626T100000
DTEND;TZID=America/New_York:20260626T113000
DTSTAMP:20260708T225105
CREATED:20260617T153533Z
LAST-MODIFIED:20260617T170801Z
UID:151617-1782468000-1782473400@www.ri.cmu.edu
SUMMARY:RI PhD Thesis Defense - Angela Chen
DESCRIPTION:Date: June 26\, 2026\nTime: 10:00 – 11:30 AM\nLocation: GHC – Room 4405\nZoom Link\nType: Ph.D. Thesis Defense\nWho: Angela Chen\nTitle: Behavioral Modeling of Interpersonal Dynamics as Controllable Agentic Systems: Empirically-grounded Adaptive Virtual Patients for Psychotherapy Training \nAbstract: The need for mental health care continues to outpace the supply of trained psychotherapists. Training is itself a bottleneck because supervision time is limited and trainees get few chances to practice repeatedly on realistic cases with feedback. Standardized patients played by actors are expensive and hard to scale\, and the hardest clinical moments are not ones a trainee can safely rehearse with real clients. Large language models (LLMs) have made interactive virtual patients practical\, yet most current systems are not empirically grounded: they carry little psychologically meaningful state and focus almost exclusively on one-on-one therapy\, leaving multi-party interaction such as couples therapy unsupported. \nThis thesis treats the design of an LLM-based virtual patient (VP)\, an interactive agent whose behavior must remain coherent over a long interaction\, as a control system. Rather than being steered by a fixed prompt\, the VP’s behavior is governed by a dynamics controller\, so its behavioral trajectory can be inspected and adjusted. The thesis develops this idea in three studies. The first builds LLM-based measures of therapist behavior and client behavior\, applies them to a large corpus of clinical transcripts\, and uses structural equation modeling to estimate how therapist micro-skills affect a client’s disclosure and emotion. The second turns these estimates into an adaptive virtual patient: detectors read the therapist’s empathy and exploration\, a dynamics controller updates the patient’s disclosure state using the estimated coefficients\, and the language model generates a reply at that level of disclosure. The third study extends to a multimodal\, multi-agent interaction system in which two VPs and the therapist interact through a stage controller built around the demand–withdraw pattern common in couples therapy. \nMethodologically\, this thesis shows how to build human–AI interaction systems for a specialized domain that requires explicit behavioral modeling. Concretely\, it delivers a validated pipeline for measuring therapy process\, two training systems for individual and couples therapy\, and a corpus of therapist–agent interactions. Together these broaden access to deliberate practice for trainees\, using AI to strengthen therapists’ preparation rather than substitute for it. \nCommittee:\nHaiyi Zhu (Chair)\, Carnegie Mellon University\nSherry Tongshuang Wu\, Carnegie Mellon University\nAaron Steinfeld\, Carnegie Mellon University\nHolly Swartz\, University of Pittsburgh Medical Center \nLINK TO THESIS DRAFT DOCUMENT
URL:https://www.ri.cmu.edu/event/ri-phd-thesis-defense-angela-chen/
LOCATION:Gates Hillman Center 4405
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=America/New_York:20260630T093000
DTEND;TZID=America/New_York:20260630T110000
DTSTAMP:20260708T225105
CREATED:20260622T175815Z
LAST-MODIFIED:20260622T175815Z
UID:151640-1782811800-1782817200@www.ri.cmu.edu
SUMMARY:Spatial Reasoning and Semantic Representations for Autonomous Exploration and Object Search
DESCRIPTION:Abstract: Autonomous robot exploration and object search in unknown environments are fundamental capabilities in robotics\, with applications ranging from search and rescue to structural inspection. A central challenge in both tasks is that robots often must make decisions based on information they have not yet directly observed–reasoning about unexplored space\, predicting future information gain\, or inferring where a target object is likely to be found. This thesis argues that autonomous robots can achieve efficient exploration and long-horizon object search by leveraging spatial and semantic representations of increasing expressiveness that enable reasoning beyond immediate sensor observations. \nWe develop this argument through a progression of representations. We begin by extracting geometric cues–such as rooms\, doorways\, and spatial connectivity–directly from sensor data\, and show how a team of robots can use these compact structural representations to coordinate room-based exploration in indoor environments. We then move beyond what has already been observed: by learning to predict unobserved portions of an environment\, a robot can estimate the information it would gain from exploring a candidate location or path before actually visiting it\, and a team of robots can use these predictions to coordinate exploration and manage information sharing under communication constraints. Finally\, we extend reasoning beyond the geometric altogether\, introducing a persistent 3D semantic representation that grounds open-vocabulary vision-language understanding into a structure capable of perceiving objects well beyond the range of onboard depth sensors. This representation enables a single robot to perform long-horizon object search in large outdoor environments\, and allows a team of robots to coordinate search behavior by sharing a sparse\, communication-efficient form of this representation rather than dense maps. \nWe demonstrate this progression through six contributions spanning indoor and outdoor environments and single- and multi-robot deployments\, showing that as representations grow richer–from geometric structure\, to predicted space\, to persistent semantic memory–robots become capable of more efficient exploration\, more effective object search\, and more naturally coordinated teamwork. \n\nThesis Committee:\nSebastian Scherer (Chair)\nYonatan Bisk\nWenshan Wang\nGraeme Best (University of Technology Sydney)\nMicah Corah (Colorado School of Mines)\n\nThesis draft
URL:https://www.ri.cmu.edu/event/spatial-reasoning-and-semantic-representations-for-autonomous-exploration-and-object-search/
LOCATION:Newell-Simon Hall 4305
CATEGORIES:PhD Thesis Defense,Student Talks
END:VEVENT
END:VCALENDAR