[MUSIC]
Hi.
So, when you think of walking robots, the first thing that you probably picture is something from Boston Dynamics, those robots that do backflips and that can run.
Run really fast, well I'm here at the Anvilap at Celtic in California.
Where they're building and designing a different type of walking robot.
The idea is that if you can teach them to run, walk and jump, just like us.
You can translate that into something that's really useful for humans, like prosthetics and exoskeletons.
So, I'm going to go inside, take a look at the lab and have a play with some of the robots.
[MUSIC]
How do you teach a robot to walk like a human?
Is it a process of we say, a lot of falling down, trial and error?
There is certainly a lot of trial and error involved But we try to actually minimize that and so the word teach is interesting because we don't actually do learning on the robots themselves.
So the idea is we're more we teaching ourselves.
We try to understand the basic mathematical principles of locomotion.
Right.
We go to physics, we go to the dinamics.
And we try to sort of prove theorems and have this mathematical representation of locomotion.
And once you have that you can distill that into algorithms, that ultimately go on the robot.
So the end goal of understanding locomotion is to help humans, correct?
That's right, that's right.
And that's a big part of it.
So.
By having this approach where we try to take the sort of fundamental, mathematical principle [INAUDIBLE] point on robotic walking is what we can take that knowledge and translate it to.
And again, the idea is every time we achieve the behavior, let's say walking on running, we'd like to put that on a prosthetic device, have a person walk around with it and ultimately an exoskeleton device.
For example, for a paraplegic, somebody that can't walk.
We would like to have them get up and walk with the same kind of algorithm, so we're using on our walking robots.
So that's the ultimate process that we're shooting for.
[MUSIC]
Before the robots walk, they have to hop.
This is a one dimensional robot hopping up and down on a track.
It's a collaboration between Cal Tech and Disney Research, with the final robot designed to live inside a ball and bounce around.
[MUSIC]
And you'll see it settle into that desired hot place.
It's getting into the groove now, it's getting into the groove.
So it's quite noisy.
It is quite noisy.
How are you gonna make it quiet when it's in a sphere?
So the sphere itself is gonna help a lot with that, we're not gonna have metal contacting the ground.
We're not going to have this whole frame.
Kind of fun.
Yeah.
Can feel it.
This is what we're hoping will happen in the park.
There's gonna be kids walking up and seeing this and start playing around with the robot.
Yay, [LAUGH]
[MUSIC]
Amba is a two-dimensional robot used to study locomotion.
It can go backwards, forwards and up or down.
We also designed this controller for perturbation, including kicking or pushing, push recovery, per se.
All right, that sounds scary, push recovery?
So if I literally go in and push it, it's gonna be able to recover.
It will recover unless you're really agressive.
Okay.
But I doubt it, you can
I challenge you to-
Really?
You challenge me?
Okay.
Yep.
Where should I try and push it.
Okay, so the safest place will be around the hip in this line, so just this bar.
So, using the bar, I should try and push it back?
Yeah, give it like a strong push.
All right, this feels really cool, this feels really cool.
I can't believe I'm trying to trip [UNKNOWN] [LAUGH] It's totally fine, wow.
That was, was that hard?
Was that hard?
Was that not hard enough?
That was more aggressive than what I normally do.
[LAUGH]
Really?
I was scared I went way too hard too soon, I'm sorry.
You know, I'm surprised.
[LAUGH]
But it's perfectly fine, it recovered completely.
Yep.
Besides shoving the robot, I also tried tripping it.
This feels so cruel.
I can't believe I'm gonna do this.
I'm sorry robot.
[BLANK_AUDIO]
It had a little stumble.
Yeah.
But it's fine.
I got me a little nervous but-
Really?
[LAUGH]
[MUSIC]
And if that That's not enough, I push it with a stick.
What if the robot uprising happens and the robots watch this, and they think that I'm cruel?
I'm not cruel.
No, you're trying to make them move better.
I'm trying to help you.
I'm making you all better, okay.
Yep, you're learning their limits.
Here we go.
I feel so cruel doing that, but it's recovered.
That's amazing.
All right.
So cool.
[LAUGH] So again why do we do, why do we try and trip the robot?
What's the point of that?
Because when you're designing walking controllers you have very particular scenario in mind, but in reality when you put a robot outside there are so many things that can happen.
We want to design a controller and consider it's robustness against some type of uncertainty.
So it can go outside and walk outside.
[MUSIC]
[MUSIC]
The overall aim is not to try and mirror how humans walk, but achieve what's called human-inspired control.
So they can translate robotic walking to prosthetics, like this one called Ampro.
Unfortunately, I can't try it on myself because I haven't been trained.
But Rachel gives me a quick overview of what it feels like to wear.
So how does it actually feel to give over control to this and walk?
Yeah.
It took an adjustment to be able to learn to balance with it.
And also I have to fully trust the robot to put all of my weight on it.
Because there's obviously times that I don't have my foot on the ground, and I'm only standing on the prosthesis.
So how heavy is it?
So it's slightly more than the weight of a human leg and so it's around 11 or 12 pounds, but it feels heavier because we still have the weight of our own leg.
It's all powered through this litho battery.
So you just hold on to it on the back there?
Put it on.
Yeah, so I just put it in my pocket.
And it can last for about three to four hours of continuous walking.
How fast do you think that you could go with it?
Is there a limit, or is it really just up to you?
It's just up to me, but the pace I'm walking here, that's about the normal pace that I walk when I'm with it.
Right, so you don't really wanna be doing sprints and stuff with this?
No, it's just made to do walking on flat ground, for right now.
And the actual mechanism of the foot.
So, it looks like it's just walking fairly flat but if you're walking uphill or stairs, how does it cope with that?
That's like something for future work is to design the motion to specifically be for walking down the slope or up the slope But it's robust enough, two slopes that I can still walk with it.
And so obviously in the lab, it's a very controlled environment.
Surfaces are flat.
There's no kind of weather incidents.
But how do you test the things that are edge cases like icy, slippery surfaces, sand and dirt, things like that.
That's right, and you really pointing to exactly one of the big problems is in the lab, we know the surface is hard.
We know exactly how to walk on that.
But if you take and put that robot on ice or sand it's gonna fall.
It's gonna fall hard.
I joke with people when they're worried that sort of robots are gonna take over the world, and Terminator is coming.
I said, just run onto some sandy slope.
It's gonna fall on its face and you're gonna be totally safe, because we haven't really solved that problem yet.
So
[MUSIC]
But we've made gains and that's really what we are working towards right now.
Meet Cassie made by Agility Robotics.
It's a robot that moves in three dementions and all the mass from it's movements is up top.
So it's carbon fiber legs have a full range of motion.
So before we actually put anything on the robot, what we do is Simulate it, so you actually will design our walking trajectories through the same optimization problem or through the same type of methods as we do for Amber robot prosthetic that you saw earlier.
So we're using a lot of the same concepts here when we're designing the controllers.
So this is the walking controller that you're going to see on the actual robot
Being simulated.
Unlike the other robots in the lab, Cassie is designed to go out in the real world.
It's semi autonomous, so Jake is using a controller to give it nudges in certain directions, but isn't telling it where to put its feet.
[MUSIC]
[MUSIC]
Cassie is doing an amazing job, walking on terrain that's got a lot of slopes, taking twigs with ease.
It's fantastic, and you've seen people that have just literally stopped in their tracks, watching, reacting.
There are kids in prams that are wide eyed And seeing this amazing feat for what it is, Cassie literally stops traffic.
[MUSIC]
But the real test for Cassie is getting out onto tricky terrain like dirt and grass, or in this case, bricks.
[MUSIC]
[MUSIC]
[SOUND]
[UNKNOWN]
That was amazing.
That's really-
That was really impressive, I was like shocked, you can see the [UNKNOWN] here.
Look how hard this is.
[UNKNOWN]
She did as [UNKNOWN] as well as I could have [UNKNOWN].
Yeah.
Which is the whole purpose of it, so.
Fantastic.
Yeah, If I'm gonna make it to here it's like Yeah.
You know an A on the exam so.
Well that was an amazing experience at the [UNKNOWN] lab I got to interact with robots, trip them over and prove distrubance testing.
And it's really fascinating to see how this is gonna translate to eventually helping humans.
I can't wait to see what happens next.
Thank you so much for watching the show and make sure to give it a thumbs up.
And subscribe and I'll see you next time.
[MUSIC]
Thats just mean, but okay, I'm going to do it anyway.
Watch my terrible aim.
Anywhere.
You got it.
I got it.
I got it.
[LAUGH]
See, you [UNKNOWN]
I can't believe it!
I broke it.
I'm so sorry, [UNKNOWN].
But this is all for research, right?
Yeah, yeah.
[LAUGH]
[MUSIC]
