Traditional robots work in very constrained and specific
environments right now, and what I'm trying to do with
this research is have the robot think about different ways
it can fail before it even starts doing something and then
adjust its environment to cope with those failures.
So for example when you're trying to place an object that
tips over very easily because it's got a very small base, or
it is very tall, you want to be able to use your other hand
to help you place the object. Our algorithm can identify the
fact that this tower, that we see here, will tip over and use
the other hand to help guide dropping that tower. So you
can imagine any robot that's working in a household
environment needs to be able to deal with objects that are
weirdly shaped, or that it has grasped weirdly. And how my
algorithm might be useful, is that you can now reliably place
these objects instead of having a failure.