Date of Award

Winter 2023

Project Type

Dissertation

Program or Major

Computer Science

Degree Name

Doctor of Philosophy

First Advisor

Momotaz Begum

Second Advisor

Laura Dietz

Third Advisor

Marek Petrik

Abstract

Learning from Demonstration (LfD) is a powerful approach that enables users to program robots by simply demonstrating how to perform tasks. Rather than just mimicking the task, the robot should capture key characteristics of the task to adapt and generalize in various situations. Given that there is no explicit objective for each demonstrated task, optimal control methods cannot be directly applied. Two primary paradigms have emerged to overcome the limitation: 1) inverse learning and 2) imitation learning. In this dissertation, I formulate several time-invariant and stable dynamic system-based LfD models that leverage both inverse learning and imitation learning in the continuous control setting. I evaluate the models on a variety of tasks, such as therapeutic exercises and household activities, demonstrating their potential for improving robot performance, adaptability, and generalization in real-world scenarios. The primary contribution of this dissertation is the advancement of LfD methods for continuous control tasks.

Share

COinS