Stanford Seminar - The Next Generation of Robot Learning

For robots to be successful in unconstrained environments, they must be able to perform tasks in a wide variety of situations — they must be able to generalize. We’ve seen impressive results from machine learning systems that generalize to broad real-world datasets for a range of problems. Hence, machine learning provides a powerful tool for robots to do the same. However, in sharp contrast, machine learning methods for robotics often generalize narrowly within a single laboratory environment. Why the mismatch? In this talk, I’ll discuss the challenges that face robots, in contrast to standard machine learning problem settings, and how we can rethink both our robot learning algorithms and our data sources in a way that enables robots to generalize broadly across tasks, across environments, and even across robot platforms.

Introduction.
Behind the scenes....
Robot reinforcement learning.
Can we learn something more general than a policy?.
Has meta-learning accomplished our goal of making adaptation fast?.
Prior literature on multi-task learning.
Hypothesis 1: Gradients from different tasks often conflict.
What does our data look like?.
Can we accumulate and reuse broad datasets across labs?.
Can we use other data too?.
Learning from Observation and Interaction.
Can the model leverage the observation data to improve?.
Goals intelligent behavior in open-world environments.
If you're interested in learning more....
Students & Collaborators.