Title:
Robot Manipulation Alongside and in Collaboration with People
Robot Manipulation Alongside and in Collaboration with People
Author(s)
Kent, David
Advisor(s)
Chernova, Sonia
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
Autonomous robot manipulation in unstructured environments is a required behavior for many robotics applications, from day-to-day household tasks to remote exploration of dangerous environments. To effectively deploy such systems to solve real-world problems requires approaches and representations that are compatible with people, who can be direct collaborators, subjects of assistance, or independent agents sharing the robot’s environment. The objective of this work is to improve the autonomy of robot manipulators in unstructured environments while addressing the unique challenges of working with and around humans. To address such challenges, we posit that autonomous manipulators must be both easily adjustable by system designers, and adaptive to humans in the environment, which we achieve through the use of transparent representations, human-in-the-loop systems, and learning from demonstration, across both skill- and task-level manipulation.
This thesis seeks to investigate the hypothesis that improved autonomy, adjustability of behavior, and adaptiveness to people lead to greater robot efficiency and effectiveness in manipulation tasks when operating alongside and in collaboration with people. To support this claim, this thesis contributes: (1) novel approaches for human-in-the-loop grasp pose specification for teleoperation that leverage depth data and robot autonomy to balance responsibilities between the operator and the robot; (2) efficient skill-level learning by means of a pairwise ranking formulation of autonomous grasp calculation that enables robust mobile manipulation and supports interaction-efficient training and adaptability; (3) efficient task-level learning by means of a novel unsupervised learning approach for hierarchical task models with action execution preferences that enable human-robot collaboration; (4) a novel algorithm for adaptive and collaborative task planning that builds on our learned hierarchical task representation; and (5) a formulation and exploration of autonomous human observation that utilizes manipulation-enabled free-flying robots to unobtrusively support humans during non-collaborative tasks in remote environments.
Sponsor
Date Issued
2021-01-22
Extent
Resource Type
Text
Resource Subtype
Dissertation