Scientists are teaching robots how to walk, run, move, just by showing them a video like this.

It's a breakthrough designed to help people with mobility issues.

We leverage AI.

We want to use AI to harness the power of simulation.

Artificial intelligence.

Researchers at NC State have created a new kind of exoskeleton that already knows how humans move thanks to AI.

Exoskeletons have been around for a while.

Traditionally, the systems were bulky, heavy, and focused on the lower limbs.

Those older models also required each user to experiment with them, while lab technicians calibrated the machinery that ultimately helped them walk.

[MUSIC] The new system is lighter, smaller, smarter.

>> So overall, our exoskeleton is a personal mobility assistance device.

So everyone can use it, including able-bodied individuals and also people with a disability.

So it's because it's very lightweight and portable.

Many exoskeletons, they are very, very heavy.

Many exoskeletons, they are about 20 to 30 pounds.

Our exoskeleton is about five, six pounds.

So much lighter and much smaller.

>> Everybody uses their muscles in slightly different ways to do most things.

In fact, there are a few hundred muscles in the body you use together simultaneously.

Those older exoskeleton models focus just on muscle groups in the lower limbs, while ignoring others needed for an activity.

And that wasn't exactly helpful.

The AI teaching this exoskeleton changes everything.

These researchers have used it to develop a device a person can use virtually straight out of the box.

It's kind of like a Segway, remember those?

They came out years ago, and about after a minute of practice, you can just sort of go.

Sue envisions the device to be something like that, since the system already knows how humans move.

It just needs to be calibrated for the individual user.

>> So usually it takes one or two hours to do human experiments, to figure out how the robot controller, the software, can coordinate how to assist the human locomotion, like walking or running.

But in our case, we do everything in a computer simulation.

It doesn't require any human experiments, and the robot can be deployed immediately.

>> The aha moment for this new type of exoskeleton was to have AI analyze how an able-bodied person walked, ran, and climbed stairs.

Researchers then had the computer imitate the motion and muscle coordination, and use that data to train what they call a microcontroller on each side.

As a user moves, the exoskeleton adjusts and fine-tunes itself to the user's movements.

Importantly, it learns how much force is needed and when to apply it.

The AI calibrates itself to walk through a host of variables like timing and how much power to give the user.

And thanks to AI and machine learning, it has learned the minute workings of around 200 muscles in the body, top to bottom.

>> We don't want the robot to generate the assistance too early or too late, right?

Think about if a person are helping you to walk, the timing when to provide this assistance is crucial.

If it's too early, maybe the person can fall, or too late, the person can also fall, right?

After this AI part figures out the activity, then the AI part will generate the appropriate assistance with appropriate magnitude and appropriate timing.

[MUSIC] >> Patients who have suffered strokes or related neurological issues, or elderly individuals who may be at risk for falls, may be helped by the discovery.

Su says there are even applications for those with cerebral palsy.

Right now, it's just a one size fits all device, and the team hopes to make it available for more people in about a year.

So this is the second generation, is that right?

>> Yes, this is our second generation of it.

We try to make it as lightweight and more compact back than our first generation.

And it's also more adjustable to different types of body sizes as well.

>> So can you feel it moving your legs?

>> Yes, so whenever I try to lift up my thigh here, whenever I walk, it will give you a quick jolt, try to lift up your leg.

And whenever you're stepping back down, it will push your leg back down so it can straighten out your legs whenever you need it to be.

>> It's the action waiter and microcontroller that move the exoskeleton.

The user moves, the microcontroller senses that movement and adds however much power the device thinks the user needs.

It learns how humans put the best foot forward, as it were.

And what's different here is that this robot, this device can actually move your legs backwards as well.

>> Yes, exactly.

So a lot of passives, devices like with springs, it's only in one direction.

So these power ones are definitely very useful in both lifting up your leg and also lowering and straightening out and supporting your leg as well.

[MUSIC] >> But when a user puts the exoskeleton on, human and machine are in sync.

>> The more data you give it, the smarter it kinda gets.

So it's really trying to fit your gait profile, essentially.

So the more you walk, the more it kinda picks up that data and try to fit that curve on this kinda thing.

>> So Sue is optimistic about the future.

He says that by doing all the testing and simulation, his team has streamlined the process, which he says will make these exoskeletons more affordable and accessible.

And he says who knows, a device that helps people with paralysis walk might be just around the corner.