Title: Machine Learning for Agile Robotic Control

Nolan Wagener

Robotics PhD Candidate

School of Interactive Computing

Georgia Institute of Technology

Date: Thursday, November 16, 2023

Time: 10:30am–12:30pm EST

In-Person Location: Coda C1215 Midtown

Zoom Link: https://washington.zoom.us/j/99765863400

Committee:

Dr. Byron Boots (Advisor), School of Computer Science and Engineering, University of Washington

Dr. Panagiotis Tsiotras (Co-Advisor), School of Aerospace Engineering, Georgia Institute of Technology

Dr. Sehoon Ha, School of Interactive Computing, Georgia Institute of Technology

Dr. Seth Hutchinson, School of Interactive Computing, Georgia Institute of Technology

Dr. Andreas Krause, Department of Computer Science, ETH Zürich

Abstract:

Robotics benefits heavily from structure. By exploiting that structure, such as by modeling the mechanics of a system, roboticists can quickly generate solutions for a given task. However, this structure can limit flexibility and require practitioners to reason about challenging phenomena, such as contacts in mechanics. Data, on the other hand, provides much more flexibility and, when combined with deep neural networks, has given rise to powerful models in vision and language, all with little hand-engineered structure involved. While it is tempting to fully eschew structure in favor of learning-based methods for robotics, we show how data and learning can be gracefully incorporated in a structured way. In particular, we focus on the control setting, and we demonstrate that robotic control offers a variety of modes that data can be utilized. This includes: modeling off-road vehicles with neural networks for aggressive driving, framing model predictive control as an online learning process, using safety interventions as a learning signal, and leveraging human motions to ground learned robot motions.