Assistive robots have the potential to support motor recovery after an injury as well as to alleviate the burden of physical tasks for both impaired and able-bodied individuals. Over the last few decades assistive robots have become increasingly more capable. Many can now provide task-specific assistance and can be safe enough to physically interact with people, but they are still incapable of versatile collaboration. While effective assistance requires robots to adapt to and infer intent from their human partners, adaptation and intent inference are particularly challenging when the tasks are novel and/or dynamic. As a result, existing robots have difficulty engaging in activities that involve dynamic motion, such as catching a falling object, as well as tasks that are unfamiliar to the robot. One of the main contributing factors are the limited communication capabilities available to the human-robot pair. Unlike a pair of people, the human-robot pair does not have the ability to communicate flexibly to coordinate interaction, so the robot often relies on passively inferring intent from human movement or interpreting inputs from a joystick-like interface. There is a need for flexible and comprehensive languages for human-machine communication. In this thesis, I present mathematical formulations and algorithmic tools that enable the robot to interpret dynamic motion as a communication signal in the context of a task. I begin by characterizing metrics of dynamic performance and show how they can be used to study post-stroke impairments. I then introduce methods that enable human-robot teams to co-create communication protocols from interaction. Lastly, I design algorithms that parse demonstrations of dynamic tasks to learn task definitions that enable the robot to recreate a new dynamic skill. In each case, I show how the proposed algorithmic tools can improve robotic assistance by testing them on robots interacting with people. I conclude this thesis with a discussion of future research directions that will help facilitate flexible human-machine communication for versatile assistance and rehabilitation.

Creator

• Kalinowska, Aleksandra Ola

DOI

• https://doi.org/10.21985/n2-k6ze-r019

Subject

• Physical therapy
• Robotics
• Communication

Language

• en

Alternate Identifier

• etdadmin_upload_1014307
• http://dissertations.umi.com/northwestern:16780

Keyword

• stroke rehabilitation
• human-machine communication
• human-in-the-loop RL
• metrics of motion
• imitation learning
• assistive robotics

Date created

• 2023-01-01

Resource type

• Dissertation

Rights statement

• In Copyright