Lawson Wong - High-Level Guidance for Generalizable Reinforcement Learning

Title: High-level guidance for generalizable reinforcement learning

Abstract: Reinforcement learning (RL) is a compelling framework for
robotics and embodied intelligence when the environment/task is not
fully known. However, it is difficult to make RL work. My thesis is that
RL is difficult because it is too general. We need to, and often can, provide RL a helping hand by providing a modicum of task-relevant high-level information. In this talk, I will discuss various thrusts in my research group on this theme: (1) Using symmetry to quickly learn to plan and navigate; (2) Following a single high-level trajectory such as a path on a coarse map; (3) Integrating a wider range of guidance into the RL loop.

Bio: Lawson L.S. Wong is an assistant professor in the Khoury College of
Computer Sciences at Northeastern University. At Northeastern, he leads
the Generalizable Robotics and Artificial Intelligence Laboratory
(GRAIL). The group's research focuses on learning, representing,
estimating, and using knowledge about the world that an autonomous robot
may find useful. His research agenda is to identify and learn
intermediate state representations that enable effective robot learning
and planning, and therefore enable robot generalization. Prior to
Northeastern, Lawson was a postdoctoral fellow at Brown University,
working with Stefanie Tellex. He completed his PhD at the Massachusetts
Institute of Technology, advised by Leslie Pack Kaelbling and Tomás Lozano-Pérez.

Видео Lawson Wong - High-Level Guidance for Generalizable Reinforcement Learning канала MIT Embodied Intelligence