Ph.D Thesis Proposal: Rowland O'Flaherty

10:00 AM-12:00 PM on November 10, 2014

Location: TSRB 423

Title: A Control Theoretic Perspective on Learning in Robotics

Rowland O'Flaherty
Robotics PhD Student
School of Electrical and Computer Engineering
College of Engineering
Georgia Institute of Technology

Dr. Magnus Egerstedt (Advisor), School of Electrical and Computer
Engineering, Georgia Tech
Dr. Ayanna Howard, School of Electrical and Computer Engineering, Georgia
Dr. Patricio Vela, School of Electrical and Computer Engineering, Georgia
Dr. Jonathan Rogers, School of Mechanical Engineering, Georgia Tech
Dr. Charles Isbell, School of Interactive Computing, Georgia Tech

For robotic systems to continue to move towards ubiquity, robots must
become more interconnected, intricate, and large-scale. The required
complexity of robotic systems comes with a cost. This cost is an increased
difficulty in design, particularly in control design. A way forward in
dealing with this difficulty is to use machine learning with control
theory. Control theory and machine learning are highly related fields; both
are used for making decisions in cyber-physical and robotic systems but
each is implemented at different levels of abstraction and at different
time scales. Control theory makes low-level decisions at high rates, while
machine learning makes high-level decision at low rates. For complex
systems it is necessary to use control theory and machine learning for
making decisions at any abstraction level and at any time scale. The
objective of the proposed research is to integrate tools from machine
leaning and control theory to solve higher dimensional, complex problems,
and to optimize the decision making process between exploration and
exploitation. In doing so, learning becomes part of the a control algorithm
and control becomes part of the leaning process.