iCub

So Robot in a Crib is the title.

And, and they put a little baby-looking robot called the iCub.

I've seen it before, like, "Oh, they got an iCub." They, and they put it in a crib and put a mobile above it, and the little mobile has little QR codes on it, okay?

[laughs]

So the camera can see this mobile, and what do they do? They do one of these experiments where they attach one of the limbs with a string to the mobile, and they equipped it with a very, with a very particular, their own br-br-- their own brand of machine learning algorithm, deep learning algorithm, to try to get this robot to learn how to predict the motion of this mobile.

Because

he heard the iCub when he heard you, Alex.

Yeah.

Now he's just ready for it.

So it's Robot in a Crib: How a Playing Robot Helps Us Understand Sensorimotor Contingency Learning.

So they actually put a child-sized iCub robot in a crib, and like a baby, you know, might use its arm to stir around the air and get a mobile hanging over the crib to move, they ask, "Can a robot do this, and can it learn that there's this link between your actions and their sensory effects?" So this is a way of training robots.

It is really funny that they use a baby in a crib, but it's a very interesting problem to investigate.

Okay, back to the main show.

Exactly.

The vast majority of the robots used in these 49 studies were models like SoftBank's NAO or Pepper or the iCub.

And those look very robotic.

They do.