Virginia Tech
From Pixels to Partners: A Hierarchical Approach to Adaptive Human-Robot Collaboration
Abstract
dc:description.abstractThe increasing adoption of robots across various industries, including warehousing, healthcare, and meat processing, highlights their efficiency and ability to handle hazardous, repetitive tasks. However, the deployment of these systems often remains confined to isolated robot cells. To ensure reliability, current frameworks impose significant environmental constraints, such as fixed object placement, uncluttered workspaces, and uniform backgrounds. While robot learning performs well in such structured environments, it fails in the inherently unstructured and unpredictable real-world industrial settings. This dissertation addresses this fragility of current robot learning frameworks. We propose a hierarchical approach to developing adaptive robot learners, moving from foundational data efficiency to environmental generalization and, ultimately, sophisticated human-robot coordination. The first tier of this research focuses on data-efficient skill acquisition. We introduce an autonomous reweighting framework designed to correct biases within imbalanced human datasets. This ensures the robot masters a full spectrum of behaviors, including critical edge cases, without the need for intensive manual labeling. The second tier addresses environmental generalization in the presence of static, task-irrelevant noise. We propose a masking approach that effectively "blinds" the robot to environmental distractions --- such as clutter or varying table texture --- allowing the policy to condition exclusively on the invariant features essential for expert decision-making. Building upon these foundations, the third tier tackles dynamic generalization through intentional human-robot collaboration. We develop an algorithmic framework capable of reasoning over high-level representations of human strategy and its evolution. This allows the robot to not only adapt to shifting human behaviors but also proactively influence partners toward high-reward, collaborative strategies. Finally, we validate the technical feasibility of collaborative robots through a pilot study in a high-stakes industrial context: collaborative meat processing. By implementing specialized vision and safety control protocols for complex tasks like slicing and trimming, this work demonstrates a robust roadmap for integrating robots as safe, adaptive, and truly collaborative partners in the modern workforce.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Mechanical Engineering
- Department dc:contributor.department
- Mechanical Engineering
- Grantor dc:publisher
- Virginia Tech
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Parekh, Sagar
- Chair dc:contributor.committeechair
-
- Losey, Dylan Patrick
- Committee members dc:contributor.committeemember
-
- Akbari Hamed, Kaveh
- Li, Suyi
- Eskandarian, Azim
- White, Robin
Subjects
dc:subject × 5Rights
dc:rights- Statement dc:rights
-
- Creative Commons Attribution-NonCommercial 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Other
- vt_gsexam:46718