Robotics Colloquium Series
Abstraction in robotics
Dr. Lawson Wong
Friday, November 8, 2019
2:00 pm – 3:00 pm
Gateway Park 60 Prescott St.| Rm. 1002
Abstract: Robotics is a big data problem. To make sense of the physical world, perform tasks well, and generalize across environments, robots need to represent and understand the world at the "correct" level of
abstraction. What "correct" should mean remains to be seen.
In this talk, I will describe two lines of work that attempt to answer this question from concrete and theoretical perspectives respectively. I will first discuss work on using natural language as an abstraction of
robot trajectories and behavior. This is an important problem for robotics, since we envision users using natural language to instruct and communicate with robots to perform a wide variety of tasks. To use an abstraction effectively, we need to both decode from and encode into the abstraction. Applied to natural language, I will present our approach for grounding natural language instructions to robot behavior, via intermediate representations such as linear temporal logic (LTL). Next, I will describe preliminary work on generating natural language descriptions from robot trajectories.
In the second half, I will discuss ongoing work on the theoretical foundations of state abstraction in reinforcement learning, a common framework used in robot learning problems. In particular, we view state abstraction as data compression, and apply results in information theory (rate-distortion theory) to the behavior cloning setting. Given an expert policy, we can automatically synthesize state abstractions in a principled fashion. There are still many challenges to extend this effectively to the full reinforcement learning setting, and I will close by describing some threads that my group is currently pursuing at Northeastern.
Bio: Lawson L.S. Wong is an assistant professor in the Khoury College of Computer Sciences at Northeastern University. His research focuses on learning, representing, and estimating knowledge about the world that an autonomous robot may find useful. Prior to Northeastern, Lawson was a postdoctoral fellow at Brown University. He completed his PhD at the Massachusetts Institute of Technology.