Description
- Abstract:
- Humans are able to solve complex problems by distilling their knowledge of the world into simplified task-relevant representations and creating plans to achieve their goals. In addition, central to effective human-human collaboration is the ability to teach these concise models of the world to situated partners with ease. Motivated by these properties, this thesis develops methods that enable mobile manipulators to learn action and state abstractions for task planning, and to effectively communicate and learn relevant abstractions from humans via Mixed Reality (MR) communication channels. First, we focus on autonomously learning action abstractions. We describe a novel policy class for efficiently learning sustained-contact manipulation skills, and a method for bootstrapping learning of dynamic motor skills with motion planning. Next, we focus on autonomously learning state abstractions. We describe research on learning symbolic representations for navigation to support task planning on a mobile manipulator platform. Lastly, we describe research on learning action and state abstractions from end-users via MR. Our MR system enables humans to easily teach robots how to manipulate objects as well as label scene information to support planning. Collectively, these works lay the groundwork for enabling mobile manipulators to solve tasks in complex environments by learning state and action abstractions from interacting with the world or human teachers.
- Notes:
- Thesis (Ph. D.)--Brown University, 2023
Citation
Rosen, Eric,
"Abstraction for Autonomous Human-Robot Interaction"
(2023).
Computer Science Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:mvwuprkx/
Relations
Collection:
-
Computer Science Theses and Dissertations
Theses and Dissertations for the Computer Science department....