Dissertation/Thesis Abstract

Leveraging Structure for Learning Robot Control and Reactive Planning
by Sutanto, Giovanni, Ph.D., University of Southern California, 2020, 151; 28086666
Abstract (Summary)

Traditionally, models for control and motion planning were derived from physical properties of the system. While such a classical approach provides mathematical performance guarantees, modeling complex systems is not always feasible. On the other hand, recent advances in machine learning allow for the acquisition of models automatically from data. However, naive empirical methods do not provide performance guarantees, may be slow to train, and often generalize poorly to new situations.

In this dissertation, we present a combination of both approaches -- infusing prior knowledge by incorporating structure into learning methods. We show the benefits of this combined approach in three robotics settings. First, we show that incorporating prior knowledge -- in combination with a manifold learning technique -- applied to learning the latent space dynamics for a tactile servoing testbed, can help simplify the control problem representation such that the solution can be derived analytically from the learned model. Second, we show that infusing physical properties into learning inverse dynamics with a differentiable Newton-Euler algorithm speeds-up the learning process and improves the generalization capability of the model. Finally, we show that structure can be incorporated into the learning framework for feedback models of reactive behaviors, facilitating guarantees on desirable system properties like goal-convergence.

Supplemental Files

Some files may require a special program or browser plug-in. More Information

Indexing (document details)
Advisor: Sukhatme, Gaurav
Commitee: Culbertson, Heather, Finley, James, Meier, Franziska, Itti, Laurent, Lim, Joseph J.
School: University of Southern California
Department: Computer Science
School Location: United States -- California
Source: DAI-B 82/2(E), Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Robotics, Artificial intelligence, Computer science
Keywords: Differentiable programming, Machine learning, Manifold learning, Reactive behavior planning, Robot control, Structure-infused learning
Publication Number: 28086666
ISBN: 9798664734997
Copyright © 2020 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest