Apr 27, 2018
In this interview, we take a look at imitation learning in robotics
through the eyes of Michael Laskey, a PhD student at UC Berkley.
Laskey explains how the technology is being applied to advance
robotic manipulation to create better and safer machines.
Manipulation is one of the hardest things for robots to do. One of
the biggest challenges is obtaining demonstrations suitable for
learning. One of the ways deep learning can be applied to robotics
is through imitation learning where a human is showing the robot
what to do. The concept of imitation learning is to provide the
robot with prior information about its atmosphere by mirroring
human actions. For example, the robot learns an action and mimics
the behavior, such as folding bed sheets.
Laskey does also address some of the shortcomings of imitation
learning, including the need for some supervision as well as data.
Additionally, when robots make mistakes, they can’t recover and
errors can multiple. This is a must-listen as we do a deep-dive
into the fascinating world of imitation learning in the field of
robotics.