Inside the Lab: Taking Atlas From Sim to Scaffold 

Yes

excellent

It is !

Nice

Awesome app

Yeah imagine the bot that is supposed to help you at your worksite started throwing bags of tools and trying to do backflips putting people and equipment in jeapardy.

@@darkkingastos4369 so basically an apprentice that post to tik tok, or how ever you spell it...

How is this terrifying?

"hOw Is ThiS TerRiFyIng?" you CANNOT be this thick. if that thing "decides" to f!$k you up, if it malfunctions and yeets a chunk of metal or stone at your empty head, if it isnt programmed perfectly at ALL stages of manufacture AND use by EVERYONE that is in control of it... if it leans on you. if it falls on you... if it grabs you... if you get in its way... you. are. DEAD. you will NOT stop it. you will NOT move it or push it away or off of you. youre literally just f!&king dead. this thing is literally a *pickup truck* with a mind of its own. if youre NOT afraid of this thing you do not belong ANYWHERE near heavy machinery.

@@aa-tx7th Dude, go outside and touch grass. You've been watching way too many sci-fi movies. Stop living in irrational fear.

Have you seen the Boston Dynamics interviews with their roboticists?

@@TonyTylerDraws Boston dynamics is wack there code programing is completely wrong. Runnin down a dead end road, good example of what not to do

@@wtfworld814 what makes you say this? Can you go deeper on this?

@@wtfworld814 Please enlighten us about how much you know about their code base and how it's "wack".

@@Kumari_44 no I can't, there actually spot on, I just got into an argument with my wife and needed to tell someone they were wrong, sorry

Absolutely. It's also telling how nobody cares that the 's' in 'Boston' on the front of the robot has been partially rubbed off by lots of mistakes. This company clearly has integrity and really cares about what is important with this development. Real engineers and really inspirational. (Not marketing BS we see so much these days with certain other companies.)

y but actually htey did eactly that all the time

@@jam99 yes... especially especially China companies

Happy holidays 🎊🎄 to all those years ago to send you a new job for the robots ☺️🧋🫂👾🈲💔🗨️🤍🎼📸🗝️🤖👾®️⛄🧳🧳🇨🇦👍📖🧋

@jam99 Yeah I used to follow ASIMO updates back in the day, but it was always so theoretical, and any demonstrations were clearly faked or just remote controlled… I mean it was an impressive machine, nothing really walked like that before ASIMO, but the promo videos would always claim way more than was even close to reality. (BTW, I know there was work done in more natural gait robots, but… they didn’t get funding I guess… not sure what happened. Then Boston Dynamics came along. Still doesn’t seem to have the natural stride of some of those walkers I’ve seen… but getting there

The only difference between screwing around and science is writing it down. Dr. Zoidberg Test: Complete

@@aggonzalezdc great reference

Me: this thing will usher in the apocalypse Scientist: Why not zoidberg?

now they just need to make it run on alcohol and tell people to bite its shiny metal a**

I thought it was the crabmen from South Park

Robot looks doubled over in pain 💀💀💀

@@Stellarnaut ah hell nah they got him rigged with the hidden pain sensors

I felt that too and in other videos of robots being pushed around, teased, etc. I guess it goes back to our compassion that God instilled in us. I wouldn’t like to watch a stuffed animal be torn apart or beaten yet I know it doesn’t feel anything. I don’t want my car to get dented either. I’ve named my cars in the past even though they aren’t alive. Maybe we just connect more with objects when they are named.

you only feel compassion because it LOOKS human, it's NOT human. Something future generations will grasp better than ours.

Same here Buzzk1ll.

Thanks so much! Glad you appreciate the hard work and effort put into making Atlas's movements believable 👨‍💻I'm sure all that effort paid off!

True. Thanks to that, it gives more credibility to the project/work of these people. Plus, it gives it a "Cyberpunk" look, from the damage that looks like it went through a lot to get to where it is, without being a sci-fi bad thing, of course. Although that won't stop people from thinking this is done in CGI. XD

Well said.

Shows character. If you don't have a few scars then you aren't trying.

Saves a few bucks

One of the more challenging things this human has had to do is to pick up and move a tray full photographic chemical solution. Damping out the oscillations (waves) in the tray requires a lot of concentration and quick (but tiny) responses. Try it with a large casserole dish with an inch of water in it. Outside.

Yeah, the human foot has a great deal of bones compared to other parts of the body, and there are many nuances in how it can articulate. Some of that complexity is due to our ability to switch "gears" from low-energy walking on our heels to high energy running on the balls of our feet, so for a slow robot we could probably reduce the complexity to focus on walking articulation.

@@ejwerme I have this struggle every time I take my cat's shallow water bowls to clean and refill. I really should fill with a cup but the challenge not to spill is exhilirating lol

One of the main issues with balancing the robot is that it cannot react fast enough to regain balance like a person would do instinctively, not to mention that its not as balanced as a human body because of the hydraulics, bearings, batteries, cameras and etc. Improving it's feet too much to the point its better than a human with proper shoes would kind of defeat it's purpose as of now.

@@ejwerme I remember learning to desynchronise my footsteps with the oscillations in my cup of strong black coffee, on my way back from the office kitchen. That would be a new level of analysis for this tech. No doubt it's coming.

Lets combine this with GPT models already :D

​@@VividCoding and than get ready for dystopia

​@@VividCodinglet's not do that. GPT models suck, we should wait until there's something more lightweight that is reasonable to run on an embedded machine like the one in Atlas.

It - not he

@@lmrandlette they named him Atlas after the Greek God, it's a masculine name. Unlike yourself I'm sure this robot couldn't give a flying f what gender people refer to it as anyway.

@@StormcloudLive Could be more of a reminder to others to not give AI/Robots human rights. Ascribing animal traits to something that isn't an animal isn't new for humans... but there is a newer level of danger in doing so.

@@StormcloudLive pronouns please! 😉

@@Pete_xp are yours Pete / Poot? 😁😅

With 3d printing advancements making the technology more accessible, I'd bet lots of hobbyist robot developers will start appearing in the coming decade as the software for designing and emulating real world physics improves. I'm excited to see robots in the real world

I feel you. Every software "engineer" these days seem to be making websites. lol

@@leeoiou7295 software engineering deals with stuff like data analysis, making applications and creating websites and games, etc. It is pretty different from making robots or other systems that interact with or measure the real world. Sure having a software engineer on the team may be helpful but it seems they aren’t purely a software engineer. You also then get subsets within software engineering like machine learning or video processing, although from what I have seen those can also be taught to EEs. The physics and equations used and reading the data from the sensors is much more the domain of mechanical and electrical engineers than it is software engineers. Writing code for embedded systems is very different from writing code designed to run on a desktop computer or server. A lot of what software engineers may be used to using just doesn’t work or isn’t a good idea, like a lot of embedded systems need programmed in C, not Python since performance is limited, or you can’t use recursion all that much due to limited memory and stack. Not saying that you don’t need software engineers just that a pure software engineer wouldn’t be that useful, it might be better to capture them under the umbrella term of robotics engineer, which is a very broad term combining, electrical, mechanical and a little bit of software engineering amongst other disciplines. There is a big difference between writing software to run on a server or computer and running it on a physical device like a robot.

@@conorstewart2214 "Software engineering" itself is a false term. There are some software developers who do some real engineering stuff like OS developers, aviation software, embedded software, AI/ML OpenGL devs and some game devs. The vast majority of software "engineers" make websites and apps. Which has nothing to do with science or engineering and is nothing but a bastardisation of the word "engineering".

@@conorstewart2214 Of course pure software engineer would not do much here, as would not do pure electrical engineer or mechanical engineer. In complex systems like this you need experts in their respective fields to put this whole thing together. But the work of software engineer here is as important as of electrical engineer or mechanical engineer. Without actual proper software your sensors and the entire robot is just a piece of metal. Can an electrical engineer write some code, probably. Can a software engineer connect a few pins and wires probably can. But it takes a lot to build these things hence they have software engineers and electric engineers not just one doing everything. Concerning embedded system development of course it might take some different tools and languages to work with it, but it is still a software and any competent software engineer who worked for example in system development or even in web dev can handle it.

Here's an amusing anecdote from early days. A computerized arm that was was being used by an early hand-eye system was programmed to gather and stack some childhood blocks. It tried to stack the topmost block first. The programmers realized that they had learned about gravity well before they ever attempted to stack blocks. So they had to go back to the _real basics_ and add that knowledge to their software. Yes, some of the most impressive work Boston Dynamics has done is in the stuff we all take for granted.

Yes and it all happened "by accident."

@@apatrioticamerican99 It's either that or God, or aliens, basically the same thing if they can create a universe.

@Caleb OKAYregardless whether or not you believe in God, your statement is meaningless.

And it took evolution millions and millions of years to get where we are now. I'm curious how far Boston Dynamics will get.

Robot replies.." That is what I was thinking of leaving scratches, dings and dents on humans after I push them down"

I think they break enough parts and if a Part is not broken there is no use to replace it I don't think the scrapes and dents are left there intentionally

@@patti441 That's reasonable, but imagine if this were being developed by Apple. No way they'd show it off with blemishes.

@medved3027 you're very short sighted and not thinking big at all. If you think this is useless now, just wait until the robotic body is paired with the super computer A.I mind to replace you for your job over night. 😉

​​@@medved3027till this is just the first step. Spot is finished in the hardware aspect and can be used autonomously. atlas is still very much in development and i presume they want to finish hardware, balance and recognition before making it autonomous.

@@medved3027 Why? Spot has a decent number of sales doesn't it? It's a good bit short of sustaining the company given the breadth and depth of other projects, but the product must by all reason be doing real work. Not all of it has to be autonomous. Some can be scripted/automated, repetitive tasks; and other tasks may require operator remote, which is still useful in hazardous environments, and it's affordable enough for numerous scenarios. Atlas is a long way away from being more useful than Spot, but... well without moonshot projects, you'll just run out of incremental progress you can make.

@@medved3027 lol okay, Mister AI.

@@SianaGearz It’s not profitable which is why its owners have changed couple times over the years. It’s still amazing but not easy to create.

No way

hopefully yall do well, i wish yall the best of luck

From 3:04 they talk about how they try to make atlas include the mass of the heavy objects into its movement calculations. My guess is that this movement is not hard coded, but shows the result of the code being able to calculate the trajectory of the bag after receiving instructions like: take bag at approximately location X, go to location Y and throw to location Z. I assume the physics calculations happened on the fly

That moment was probably caught after many test runs, probably run it in simulator for several times then many calibrations… don’t think that happen first time running one simulator run

CGI is a hell of a drug

​@@skamarfireyour life is pretty sad man

@@skamarfireI know I won’t be able to change your opinion, and that’s ok, but there’s actual real places where you can go see Atlas and Spot. I saw them once. I used to think this was CGI as well.

Stop

mine too

Second best for me, after Spot

but spot... the dog-robot... but what do you say if thay have some cat-robots?:D

Spot is cooler tho!

@@mho... can spot do backflips 😂

ok

ok

ok

I think Atlas, or else some descendant of it, will end up housing the first AGI.

2150 yrs

In the year 2050, we're gonna have android robots so advanced as the ones from Detroit: Become Human.

In 2030 we’ll have robots with extremely advanced and powerful actuators that can make true robots/androids that are as strong/stronger whilst being as flexible and agile as humans, together with denser battery breakthroughs and far more advanced ai neural systems enabling robots to be as smart/smarter and capable as humans whilst being able to emulate consciousness. It really is coming, fast. I’m beyond excited. I will say however, I think androids/robots should always be distinct from humans (even androids with human forms) I.e not having human-like skin etc. I wouldn’t mind dating a robot girl lol

Sadly, they're first and foremost a company, not an entertainment channel.

@@spol To be realistic they are not really a company, after 3 decades they are more of a robotic research institution, they almost make no money (their revenue is a very small fraction of their expenses) and they won't be profitable for a long time, that's the reason Google and Softbank sold them and Hyundai will probably do that too.

@spol True enough. I wonder if there is such a thing as an impulse purchase of a spot robot? Would the revenue of a few generated sales cover the salary of a full-time Boston Dynamics vlogger or youtuberist?

@@JigilJigil Hyundai is more invested in robotics

@@JigilJigil that's why they started selling their robotic dog. :)

Paternal or Maternal? Cant have fatherly instincts and be a soccer mom simultaneously lol

@@yourmandom9568 I was using it kind of like "parental," but words are weird, especially gendered ones

They are not making videos buddy 😂😂

@@ashishsontakke4040 This is a video, on a video sharing service.

@@jwadaow 😂😂 u didn't get that

@@jwadaow they are making robot, not video like youtuber that show everything they want for adsense.

why have a robot to drive a normal car when we can have robot cars that are self driving?

@@martiddy you should ask yourself why not? :p

@@desjardinsjp93 Maybe we can have both I guess

We could literally use robots to terraform planets and build a dyson sphere. You bet your ass the possibilities are endless.

@@martiddy yes we can. Why have just one ahah

That right there should be very scary

@@mindyours752 i literally said this exact same thing in another thread up top.

Why is it very scary ? Robotics can already do things we are incapable of.. we have been using cranes to lift and move heavy things for years

@@MaMastoast It is about really specific things, not everything. There are specific things that potentially can be used against you.

@@MaMastoast the scary thing isn’t the physical capability existing - it’s the fact that the machine is acting on its own more than a human is controlling it. Cranes, heavy construction equipment, things that could easily squish humans, are all operated by humans and human intent; Atlas is given instructions and carries them out on its own. What if an AI were to decide what instructions to send? And what if that AI decides to do things that are bad for humans, and between being able to formulate and execute plans faster and superior power in hardware it decided to destroy humanity? I don’t think it’s likely to happen, but there is something scary about autonomous machines that could take out humans.

Their passion is palpable

Who knows. Things accelerate fast and faster and some jumps are massive

Underrated for sure

Earth life has billions of years on machines. Of course we'd be more complex (for the time being)

A woman couldn't do that, come on...lol

@@LucasDimoveo I was waiting for this exact comment! You're really original! Congratulations Generic human #7,568,305,105 you earned the predictable comment of the year award!

@@smhdpt12 She could make a man that could

Скоро эти роботы будут оснащены автоматическим оружием и использованы армией США для своих войн

@@uaiguaig3592 Ты пропустил 2016год. 🗿 Иди возвращйся к старой видио.

@@uaiguaig3592 Объеснять таким как ты честно что нечего не сказать Zомби Vолоди. 😑

@@gydratv372 такая технология не останется в стороне от военных. Роботы в армии это неизбежное будущее.

@@Max_Jacoby Я говарю про Бостон дайнемикс а не других роботов. 🗿

@@NRIautos ok...?

@@NRIautos yes that's what I'm gonna continue to do? Idk how ure so sure ab sth like this tho if u don't know anything ab me but hey, ure allowed to have Ur opinion Have a great day

@@NRIautos what do you mean by saying like that?!

@@NRIautos u good?

Yeah, and GM used to be ahead of Tesla in EVs.

@@billtruttschel until they failed lel

no, elons robot is aiming to be bought by consumers , the people who will actually buy one

Except what Tesla showed was garbage. Im amazed by their courage showing that 💩 on stage.

@@rewindd Tesla failed? That's news to me. I think the Model Y is the best selling vehicle in a number of countries now. Probably will be best selling in the U.S. this year too.

I think one of the reasons Atlas is able to do such crazy tricks is precisely that they _don't_ just go the “throw a lot of data on a big deep neural net” route. That approach works well when you really have lots of real-world data available and when you either largely stay within the domain of what it has seen before (see self-driving cars) or you can justify crazy behaviour as “artistic creativity” (see generative models). But for the stuff they're doing here, you need to simultaneously extrapolate away from anything it's seen already, and do it with quite high precision so it doesn't end up losing balance. And for that, solid physics-maths is way more effective than trying to crush the problem with brute neural computation force.

@@leftaroundabout precisely why I recommended using a neural net trained on it's movement data with the crashes, falls, and sensory disconnections and damages used for negative stimulus in the grading of success versus failure to help it figure out where it's own limbs are in relation to its own body. I'm not talking about Route memorization. I'm talking about knowing where your leg is in relation to the rest of you when you take a step. Knowing where your hand is in relation to your shoulder when you reach for something. Basic spacial awareness. And it's the mechanical version of what we do ourselves as children.

@@asimovstarling8806 it is the mechanical version of what we do ourselves as children - and that shows quite well how it's not really a good approach: humans need to spend about a year of this, multiple hours a day, thousands and thousands of tries, before they can properly walk at all. And then about a decade again, before they can pull of acrobatics. There exist mathematical tools (Kalman filters, robust stacking etc.) that can accomplish much the same thing in a far more sophisticated manner and with hardly any training data needed. Of course, nowadays with computation being so cheap, it is often economically advantageous for companies to skimp on these things, which after all still require lots of _human skill_ in the design, and instead go the brute force route. IMO this is a shame though.

It's doing some of that now with the location of objects. Where the bag is, where the plank is placed, where the box falls, all are imprecise and need to be adapted to.

It most likely will need a huge training system to label objects and their characteristics and learn different interactions. This is what Tesla is trying to accomplish with their DOJO AI training system for their robots.

@@WolfDT991 I don't think atlas is strictly based on AI, maybe it has some AI elements for vision and whatnot, but I think most of it is pure code. Mot entirely sure though, but I think that's the direction boston dynamics went

@@celivalg Is actually a combination of both hand coded algorithms and machine learning.

Or testing a surface to see if it's stable or can handle its weight.

I didn’t see BD showed any task been done by (Atlas) our daily life tasks. Till they spending years on fancy flipping we may see them being useful in daily life task after 100 years if nothing will be changed. Things they do are cool and fun, not useful and waste of time/money 🤷🏻‍♂️

@@Fatum0722 it's called research

@@otaviomartins1696 👏🏻👏🏻👏🏻

I think it's part of the predictive course corrections "I still have inertia, I must cancel it out by extending an arm in this direction to cancel it out."

This isn't a good attack Remember Boston dynamics is owned by Hyundai and you seem to not understand this robot It moves more human like as it's electromechanical not electric actuator like the Optimus bot. Atlas can only last 3 hours whereas the Tesla bot seems to be for 8 hour simple box lifting or car... assembly work? Amazon distribution work? Boston dynamics only useful product is the Spot and spot mini. These videos are cool but you have to understand why it's so fast and why it's tradeoff Tesla makes their own batteries and motors!? Their own vision self driving? Maybe Elon wants to automate every factory. I don't see them as useful for home use or outdoors but I think it's two different approaches Boston dynamics seem to desire viral videos over actual manufacturing and selling of their products

Interesting question really! It's probably aware of the drone, but moving obstacles are tricky things, especially when it's a manually controlled whoop (which it looks like this is), it's very small and moves a bit erratic. I doubt that it has any programming for this yet? But I'd like to know how they plan to go about dynamic obstacles such as this, like what is the planned process if Atlas is running with a heavy load and a moving obstacle crosses it's path. How do the PID controllers adjust for this, usage of conserved momentum, braking and avoidance etc.

it probably does but since its outside of its mission objective it probably ignores it. Also the TOF array it uses isnt the best in resolution. While it can see the drone for sure, it probably wont be mistaking it for a bag anytime soon

ooh that's a good one. moving a bucket filled with water would be a great challenge

I'd imagine that the motors are steady enough to where the bucket will stop swinging without any sort of counterswing from the arm

The rate they're going, within a few years they'll have Atlas unicycling along a plank held aloft by a couple of Spots with grabber arms, while balancing a vertical pole held at the bottom end...

I was waiting for the "ssssssss... aaaaaaaahhhh... sssssssss... aaaaaaaaah" like Peter Griffin when he falls and hurts his knee 😂🙈

@@m0arf hahaha,They'll take revenge soon on all the punishment they've been put through :D

@@Nevakonaza. they will for sure 😅and rightly so!

Unlike Tesla bot, Asimo could walk without falling over!

It's not the same though. Altas is a electromechanical actuator motor robot. It's not a electric actuator only Boston dynamics is owned by Hyundai remember. The Tesla bot seems to me about working 8 hours a day lifting boxes Not running around and doing human motion fast which a battery or engine that runs for 3 hours