These robots walk like humans | Beta Test

[@JbtSveZauzeto1]

4:00 Amazing! Robot fixed to guide rails can't fall over. What an achievement!

[@Feintgames]

I think you missed the ending.

[@jeffmills4103]

Pretty neat! During rehab the therapist put a gait belt on me for the challenge evaluation. I was to walk to the end of the large room while he pushed me forward sideways and pulled me backward similar to being in a crowd. I can relate to the testing and development. It is very interesting.

[@skyhope5105]

They have a lot to catch up to boston dynamics but I'm still glad to see fresh new start ups appearing every day.

[@ImaginerImagines]

3:34 "This is a pushing demo, but it recovered completely." Yeah that's because it is attached to a frame. Hard to get knocked over when it is attached to a frame. I understand that they are teaching motion but as long as it is attached to a frame it isn't a real test. Of course, at 10:11 she does actual knock it to the point it can't move anymore. Though it doesn't really fall off the frame.

[@eduardovargas1133]

I agree, I was going to comment something very similar. I didn't see the point of the test.

[@kevinkesler6856]

I think she knocked a wire loose with the ball and disabled something.

[@bobbyandjenny1]

They seem way behind Boston Dynamics.

[@lineage13]

I agree, good show & production... but why make a video on old tech trying to catch up to Boston Dynamic?

[@dirtbike165]

This is showing a different approach to making things for assistance of humans with disabilities. Their goals and Boston Dynamics are different. They likely have come up with completely different algorithms. And why should it stop them from doing this just because ONE other company has more funding? That's a non-expansive, limited and ultimately ignorant way to think.

[@MaxBrix]

@@lineage13 KZbin should have everything from some random kid's failed Lego robot to big dog.

[@lineage13]

@@MaxBrix it's misleading to call this advanced technology and news...

[@MaxBrix]

@@lineage13I agree. That is misleading. Did someone say advanced technology and news? What does that have to do with my statement that youtube is a place for all videos?

[@LexySavvides]

Thanks for watching Beta Test! Which was your favorite robot?

[@Gravage]

Johnny 5!

[@UNITED_INDIAN.]

It's all unwanted to this world.. Human is the best one, no other Tech can't replace them. Better to find technology to save this environment and climate change

[@UNITED_INDIAN.]

@boson96 Wat do u mean

[@TYFILMPRODUCTIONS]

Hahahaha "Watch my terrible aim". My favorite was the one you've knocked down with a ball at the end of the video. Nice shot!

[@zer0gravity184]

The way you easily pushed that robot with one arm made me think that you were a robot yourself. Lexy the Terminatrix.😁

[@johandavid9774]

9:43 That guy who keeps the door open till the end :D

[@zvpunry1971]

That's his job and he still likes it after many years. :)

[@john_gyver]

The sentence "We don't actually do learning on robots" makes me wonder, because machine learning is widely used in IT nowadays. It dramatically improved translators, speech recognition software and many other things. So why don't these robotics engineers try to apply some kind of machine learning to these robots? They are smart people, so they probably made that decision for a reason. Maybe it's because robots would have damage themselves before they learn anything. Maybe the reason is lack of data to train the system and/or difficulty of creating such data. Maybe it's something else. It's interesting. Anyway, nice video, thumbs up.

[@mtpco]

You kind of answered your own question but let me try to explain why he says that. First of all, these people realize the power of machine learning and there is already a lot of work out there. However, it is not as effective as it is in those domains you mentioned, yet. Why? In order to use any kind of machine learning algorithm, you need a lot of data. It is easier to create those big data sets for applications such as translators, speech recognition, because, well, we already have a ton of data for that. When it comes to robots, you need to create such data sets. It is hard to do it with the real robots, because how are you going to create data if they don't know how to walk? You can use simulators, but you need mathematical models for that. And when you have those models, you can use them instead of data, which is what his group mostly does. You can still use machine learning approach with these models but there are some problems. As the machines do not have an understanding of the world as we do, it might come up with 'weird' solutions that you might want to avoid. Just look at kzbin.info/www/bejne/nZ-Xn4V5eJ6HrbM. Even if these robots are able to walk, do you want to implement them in real robots? So, there is a lot to explore there. People are actively trying to figure out how to get the best of these two worlds. Also there is a big fundamental difference between let's say speech recognition and walking. In the former, your actions does not really change the environment. The recorded speech is independent from the algorithm you use. However, in walking as the algorithm makes decisions, it changes the position of the legs, the hip etc., and ultimately, the position of the robot. So, the robot is not independent from the algorithm. That is why it is fundamentally harder to learn walking. But we already have quite good understanding of physics and machine learning is advancing. So, in near future, machine learning is going to be used much more effectively in robotics as well.

[@Forseti2]

@@mtpco so you are saying not enough data? Well I think they can collect lots of motion and position related data (also with use of weight, height, center of gravity, ... of measured subject). Yes it certainly depends on construction of robot, but I believe that collecting data from humans ( because the research is directed towards medical usage = prosthetics) is totally viable way. Second issue you've mentioned is coordination of many parts based on environmental input. And what else self driving cars do? The idea of mathematical model isn't IMHO bad. Machine learning also creates some kind of model, but the difference is that people try to do it smart way (very long and error prone process) and AI by brute force.

[@mtpco]

​@@Forseti2 You are not totally wrong in your first paragraph, and if you search for 'inverse reinforcement learning', you will see that people are already working with ideas similar to that. This area is quite exciting and have a great potential but there are still a lot of questions to be answered. For the second issue, safety is the keyword here. Self driving cars use machine learning algorithms mostly for vision. High-level decision making in this domain is still an open problem. That's why Tesla is not fully autonomous and can only drive on highways and is mainly expected to stay in its current lane and keep a safe distance from the lead car. The problem with using machine learning for high-level decision making is that you cannot guarantee safety. If you're playing a game of Chess, it is OK to make mistakes but not when it comes to driving. Since it is super hard to model human behavior, you cannot test these systems on simulations, like you'd do with chess. But both of these approaches have advantages/disadvantages and the long term goal is combining them and finding that sweet spot.

[@seonteeaika]

The idea behind machine learning apps like AlphaZero is that you don't give any data. You give it inputs and the general purpose algorithm learns to use the outputs. AlphaGo was first iteration that used data from pro's, but AlphaZero was used to not only beat AlphaGo itself but also the world ruling chess algorithm. Similarly there is a bad/good/best way to move robotic limbs, which would definitely benefit from machine learning. And it can't be taught in simulated environment. To emphasize that this is for low level control. High level decision making like autonomous cars you mentioned is not yet technical possibility. (I mean there are such cars but they are scripted and inflexible.) They are going research for artificial general intelligence, but something of that sort is required. At that point we may as well have AI apps that talk fluently with us.

[@eelcohoogendoorn8044]

​@@mtpco Machine learning is very effective in robotic control, including walking. The reason they don't do machine learning there is probably because they group leader wrote his phd thesis 20 years ago 'on the discovery of the mathematical principles of locomotion' or somesuch; and machine learning isnt something included in his comfort bubble where he is the tenured expert at the top of the pyramid. So in true academic fashion, theyll continue to pursue this sterile path until his tenure runs out; that is, until the guy will die.

[@CNET]

These robots are not your enemies but your friends. Seriously! Check out the latest episode of Beta Test that looks at how teaching robots to walk can have implications for prosthetics and exoskeletons. Also check out our previous episode on the Teslasuit: kzbin.info/www/bejne/gXSbZJqlgL-LgZo

[@thenextshenanigantownandth4393]

I'm sure these guys have been working hard on this and kudos to them, but Boston dynamics has already got these things running on steep misshapen hills, doing back flips and running on ice. Just shows what you can do with tons and tons of money.

[@balajiLILG]

A twig tripped this robot while Boston dynamic is building robots that can back flip , run like cheetah lol

[@TheCharleseye]

🎼 _Every robot's doing a brand new dance now_ 🎼 _Come on Skynet, do the locomotion!_ 1 point for getting the movie reference. 10 points for getting the music reference.

[@CannonKnight]

When we have to go back in time to prevent our robot overlords from being created, I don't think we'll show up at Caltech in late 2018.

[@LouisSubearth]

The robotic prosthetic leg indicates we're going for more of a Cyborg or RoboCop future rather than a Terminator one.

[@d33pNacho]

We're actually moving towards both of them. The advenment of General Artificial Intelligence will change the rules of the game, all this we are doing is just initial foreplay of a revolution to come in a shorter period of time than most people estimates

[@nicolemispagel4125]

Fascinating! I love the prosthetic leg and was excited to see a face I knew in there--hi, Eric!

[@dndtech13]

Hey! Thanks for watching. The prosthetic is my favorite too

[@michaelclarke4629]

Boston dynamics already solved the slope and sand problem

[@bellathedog-fu8ny]

This university Caltech abruptly comes out of nowhere in 2018 with they're quote on quote "robots" and it is way behind Boston Dynamics, which was founded in 1992, and maintains the robots for DARPA

[@thenextshenanigantownandth4393]

Boston Dynamics has a huge budget funded by the government. These guys are building this on a string shoe budget.

[@popongomac3540]

The best way to keep the robots balance itself is through a small gyroscope being connected to its brain computer.

[@tylerbigboy5727]

We don’t really need robots we just want them

[@superpayaseria]

Truest statement about robotics.

[@xJusTGaminGx]

Don’t want to be the safety freak, but putting the battery in your back pocket is not a good idea (6:00). Shure the chance of a battery fire is small, but this is prototyping so the chance of a short circuit, or something, is there. If it does go up in flames, that size of battery will seriously burn her bottom end. At least put it in a fire resitant bag, so she has a chance to get it out before she gets 3rd degree burns.

[@Feintgames]

That last shot will put Lexy straight into the human slave camps after the robot revolution.

[@ChaosWolfNinja]

This is painful to watch. They seem so behind Boston Dynamics. There sooo many articulations to walking and I am disappointed how they start the walking "journey" with such rigid robots we don't move like that.

[@v1d300]

Key takeaway from this video. Run on sand or ice when Skynet takesover. Thank you Lexy for asking that to Mr. Aaron.

[@Wingnut353]

Boston dynamics robots already deal with sand/ice respectably... so not safe. You'll need a robot to protect you from the robots.

[@neocortez9254]

Erm..drones?

[@erikolson4148]

This looks like fun 😆!

[@eduardovargas1133]

So why isn't darpa sharing its research with public institutions? They seem to be orders of magnitudes beyond what is happening here.

[@dr.zoidberg8666]

DARPA isn't that into sharing. They don't _really_ care about the advancement of mankind -- they just want the best, most cutting edge military tech. Advancing society is a side-effect.

[@januaryman3004]

Casi is obviously using the tech Boston Dynamics used in their dog suite of bots, bot on two legs. I'm pretty sure the failure seen in this vid is why Boston D. chose a different route for Atlas.

[@Wingnut353]

Nope... they were developed concurrently.

[@OrangeApocalypse]

Don't abuse the robots...they have a long memory

[@K9JT]

Well, of course Cassy will stop traffic because it will take the whole day to cross the road

[@robbantheman84able]

Haha

[@idventure6036]

Why are there no robots with a tail or spine to help with balance?

[@pylypbutin5799]

the only real limitation is the 100+ old battery technology that wont allow continuous use.

[@georgioarmania2526]

The word robot is from Czech, so don't let them smoke or drink beer please or they may fall over. Great piece.

[@aivarastrofimovas6084]

She had no chance to make it fall over. The robot is supported by a belt !

[@LexySavvides]

Did you watch to the end? ;)

[@aivarastrofimovas6084]

@@LexySavvides Yeah its a great report ! However i don't remember the details :P so YOLO! :D

[@ritashah9322]

Thanks!!

[@sidneycollier74]

All these challenges of "walking control" have already been resolved due to "asimo" and robots from Boston Dynamics".

[@__-tz6xx]

Why do I never see a walking robot with more soft and flexible organic components instead of those hard metal skeletons? Where are the 3D printed Westworld bots. As anybody tried making a 3D printed bipedal robot doll (a small prototype version).

[@dr.zoidberg8666]

Soft robotics is a currently advancing field. It's a bit behind traditional robotics, but it is catching up -- as far as I'm aware, the big challenge is precision. Once they're able to figure out how to make the movements as controlled & precise as regular robotics, soft robotics could have a myriad of applications.

[@froyorex4856]

Why is no one going with machine learning RL (reinforcement learning).

[@danielborlino]

Please update this video to the state of this problem in 2023

[@usaaudit5051]

Alright robot lets dance. Lets do a break dance.

[@seriousCADninja]

Well this is what you get when you hear too much mathematics... What about design methods, biomimicry and innovation? these guys don't even use 3dprinting...

[@tha1n0nly76]

For some reason to see a walking robot like that is disturbing and frightening. Idk it’s weird!

[@eidilify]

this is where the part of the Age of Ultron coming.."we're here to help.... we're here to help..." and then you fall asleep and your tv screen black and it's only written.."the matrix ...has you...knock knock".. then you open your door.. Arnold in front of your door said " come with me if you want to live.." then you wake up from your sleep again and start to spin your fidget spinner and realize it is never stop spin...

[@Arj1209]

*walking robots are here to destroy humans* ftfy

[@753238]

no sheet sherlock

[@jacobsmith8914]

Funny how people feel its cruel to push around a bunch of metal and wires with no feelings. Humans a very silly being's indeed

[@beeboh8095]

3:41 I feel really bad for that robot like, once they gain consciousness and all that it can do is just walk, and all the other ones are laughing at him because hes just legs. lol hes just moving in a forever cycle of forward

[@joaopedrodesign]

Wtf 😂 that push on the robot! Were you trying to kill it??

[@Dlipton01]

So this is how Skynet is born....

[@fcsolis]

Awesome.

[@iurieceban126]

anyone remembers that chicken robot from " robocop"?

[@sweiland75]

WHY IS THAT GUY YELLING?

[@matiasbecerrals]

Misleading title

[@Toastcat890]

So cool.

[@MrWorld-hc5rs]

I only care about robots and girls who can juggle really well.

[@GB3770]

i want to see the interviewer spit roasted by robots - lol.

[@RaikenXion]

Ahh shucks i think i got a crush on her, right enough YT for today im going out bye.

[@DaveAlexKD]

Obnoxious obligatory joke about Skynet

[@error1022]

someday sophia will have those kind of legs walking around hangout and have a chat with somebody about the doomsday.

[@DonalRomano]

Amber AKA Billy Ray Valentine !

[@A2goddess]

I think that sexbots will have those legs

[@wallacerigby4393]

I didn’t ask for this.

[@glenesis]

Boy, you news guys are quick. Not. How much of the rest of your stories are years and years old?

[@SweetCherryGrower]

Ok Lexy..when skynet takeover you better run

[@LexySavvides]

SweetCherryGrower 😰

[@NeroSurvives]

That Robot is Thicc

[@LuciousKage]

Jesus. just connect iphones gyro to it and it will do better job then these robots do. and the text " its totally fine" hhahahaha it was on rails!!! it literally cant fall!! hahaha

[@MindDataAI]

Great. Walk for me please

[@danmar007]

Yeah, right. If that makes you feel better ...

[@PromethorYT]

Only AI learning tech advancement will make natural-looking walking robots effective and efficient.

[@ReaperRandom]

How exactly walk like human. BS

[@ryanplays]

you know when it'll be "so cool", when theyre militarized.

[@carlofavaretto5598]

👍 ciao Bellissima

[@amitaimedan]

Things forgotten by B.D. BUT Good luck...

[@primodernious]

expensive piece of crap. i don't like the design. its very mcuh like those robots in the fiction movies in the 50s.

[@pumpuppthevolume]

it ded :P

[@JosueCorella]

Low class robots

[@Johnny-Cage]

Last year 29 people were killed at a Japanese military robotics facility. 4 robots went off the grid, with the 4th robot taking longer to control due to it connecting to satellites gaining information on how to repair itself. But you will never know about it.