[Narrator] Say hello the Clayo, the robotic cat.

It's, well, not particularly coordinated.

But roboticists here

at the University of Southern California

have big plans for Clayo.

They're thinking about robot intelligence

in a fundamentally different way, by looking to biology.

The idea is to use computer chips that mimic how neurons

work in real animals.

Because maybe the trick to developing advanced robots

isn't actually to start by making them start.

Maybe, it's about first making them kind of dumb.

Okay, so Clayo clearly isn't as nimble as a real cat,

but that could change one day.

Right now, it's remote controlled,

but Clayo's overlords are building a system

that works like a network of neurons,

those are the cells that receive and transmit

information throughout your nervous system.

They're gathering information from the world

using simulations of the sensors that we have

in our own muscles.

And they're producing actions

simulating the properties of muscles.

[Narrator] So right now,

if you want to teach a robot to walk,

the best way to do it is to teach it a bunch of motions.

Each step is a separate part in a process,

and sorting through all of them takes a lot of brain power.

These researchers at USC are trying to simplify that.

You see a lot of animals that are fantastically versatile,

squirrels, and mice, and they run around,

and they can survive very well,

they're very agile and versatile,

and they don't carry this big brain around.

It'll be interesting to understand from the ground up,

how is it that the nervous system and the body interact

to produce behavior.

And the first stop is the spinal chord.

[Narrator] Scientists know generally

how neurons form into networks in an animal's spinal chord.

The spinal chord has it's own low-level circuits

that do a lot of the micromanagement.

Like when you go to the pediatrician,

and then they tap your knee,

and you have this knee-jerk reaction.

So, those kinds of reflexes are housed in the spinal chord.

[Narrator] So they plan to program Clayo

with the simple goal of moving forward,

and use the robotic equivalent of neurons

to make that happen.

The neurons would fire randomly,

and at some point, a combination of them

would produce a movement in a leg.

The system would then strengthen the connections

between those neurons, and demote others,

teaching it to get better at walking.

This is known as reinforcement learning.

What would happen is that you could teach it tricks.

And it would learn them,

not because it has an algorithm or anything,

but because you're reinforcing a good behavior.

[Narrator] It could be a more efficient way

to go about things,

than programming line after line of code.

Across campus, biomedical engineer Terry Singer,

is experimenting with his own kind of neuron-powered robot,

programmed in this case, to like all things blue.

All it knows, is when it sees blue things,

it wants to go towards blue things,

and avoid everything else.

That's the only thing it knows, and yet,

it's able to do that.

And this is something that is fundamental

not only to humans, but to animals everywhere.

Everything that moves.

Every worms, every invertebrate, every fish.

Everything has the ability to go toward the things it wants,

and to avoid risk.

[Narrator] We're talking about

extremely primal behaviors, here.

Hell, you do it all the time without even knowing it.

The way we walk through the world

is not by estimating the contact forces with our feet,

or trying to identify every single thing

in the field of view.

We don't do that.

We just see, and we feel, and we move.

[Narrator] Maybe it's this kind of simple,

instinctual behavior that we want to give robots,

so they can navigate an uncertain world

without thinking too hard.

That's right, Clayo, one day, this could be you.

Just hang in there.