Before humans can get robots to fold our clothes, we've got to do it ourselves over and over and over again.

From American Public Media, this is Marketplace Tech. I'm Megan McCarty Carrino. Tech firms are racing to develop robot assistance that can take over our dreaded household chores.

But teaching machines to perform these deceptively simple tasks is tedious.

They need to observe the actions thousands, sometimes millions of

times. And there's a cottage industry springing up to provide this training. Ayanna Howard

joins us to explain. She's a roboticist and dean of Ohio State University's College of Engineering.

I as a human, I perform a task. I do it in different steps. I have some type of Engineering. I as a human, I perform a task.

I do it in different steps.

I have some type of manipulation.

And a robot observes what those tasks are and those steps.

And there's a translation of our human joints, our human steps to the robot space.

I will say that this next level where we now add that to the lower mobility of

walking and navigating, that is where I would say the innovation still needs to occur and still

needs to happen. But it's basically looking, taking what we're doing, and then extrapolating.

If I make a bed, there's a couple of steps,

there's forces. What if I go to another bed that's a queen versus a king? Pills are fluffier.

Can I translate the initial information that I've had to this sort of new domain, but it's still

the same kind of steps? When they're learning from demonstration,

is it pattern recognition in the way that we think about, you know, an LLM learns from all the

text on the internet? It spits out the most probable answer. Like, convert this into kind of what

we understand or what we've come to understand about generative

AI and how that applies to robotics.

...