Robots at Play

(robot whirring) – The name of this robot is Vestri. – Apparently a mythical
gnome from Norse mythology. – A lot of times when people see a robot screw on a cap on a bottle they’re like, okay it knows how to screw
that cap onto that bottle, so it must also be able to
screw this cap onto this bottle. And that’s a very easy assumption to make, because if you watch a human do that that it’s obvious they should be able to figure out how to do that, but for these robots it’s
actually not at all obvious. – A person can learn
new things very quickly, but they do have their lifetime of experience to draw on.
– Yep. – So we have a childhood like, this is the childhood. (robot whirring) – The goal is to enable these
systems to learn autonomously, to learn on their own. What if it’s just a video
of a human doing the task with their own arm, can the robot just learn
from that raw video by essentially watching
the human do the task? One of the big questions here is, can these learn-to-learn
techniques scale towards not just a robot learning to interact with objects in this lab, but actually pushing it
closer to the real world where the robot isn’t just in one room, but it’s in many different environments and interacting with not
like a hundred objects but with a thousand objects. – The term that we often use when we talk about these kinds of
technologies is generalization. Generalization means that
the system can do something in a setting that is not the
same as what it saw before. Can they examine the world in one setting and then still be effective when the setting changes in some subtle way. I wouldn’t say it’s the
learning that’s the milestone, the milestone needs to be the capability, because that’s something
that we can measure. – Things like individual tasks, like screwing a cap onto a bottle, or picking up an object
and placing it somewhere, those are those are individual tasks and not really more high level capabilities, and if you try to get a
robot to do those things you’re not necessarily making progress towards intelligent systems. – It’s kind of very tempting to imagine sort of the science fiction future where you have a robot in your home that does all the things that
you want the robot to do. That’s actually a very difficult thing, because your home is
extremely complicated. I think much sooner than that we’ll start to see more areas for
specific applications. We already see companies
that have for example, robots in hotels that do
basic things like delivery, some of you might have
seen robots on the street doing kind of the prototypes
for delivery systems. The other factors of this technology, more so than the technology
we’ve seen in the past, builds on both technical
developments done by engineers, and experience collected
by the system itself So, for that reason, it actually makes a lot of sense that there would be this
kind of gradual deployment where you first start off in somewhat cooperative environments, not just because we need to do more coding to engineer the system
to work in your home, but because it needs to
collect more experience. (robot whirring) But it’s definitely at the stage where it needs to play more. And the reason that it’s
running here right now, by itself without anyone attending to it is because we need to run it continuously, we need it to keep
collecting more experience and building up it’s understanding. Well the great thing
about robots too is they, you know unlike people, they can communicate over the internet instantaneously with each other. – [Dasari] Like okay, we’re thinking of clamping it– – Clamping it here, here, and here, right? And making it so that like
it’s really rough here. – Once that actually becomes practical, once we can put these
things in peoples homes, offices, hospitals, they can collect experience, and get better together. – And we won’t need to
run an entire lifetime, we can actually parallelize
across all these robots. – Even in this lab, part of why we have multiple robots is because that way we
can prototype this idea, at least in small scale initially. – [CALIFORNIA] Do you feel, do you actually have like, get an emotional attachment to the work or do you give them names or? – They’re certainly
temperamental sometimes. (robot whirring)

, , , , , , , , , ,

Post navigation

Leave a Reply

Your email address will not be published. Required fields are marked *