The future is here: Meet Tesla Bot

[noko]

The first series used rubber skin, they were easy to spot...

Hate to have bought a $100,000 robot to clean up motel rooms to end up in the pool electrocuting the patrons swimming or just shorting out. I can see military robots using H2 fuel cells for power, run rather fast and take a beating, the 1 million dollar+ version.

 

[Xrave]

Considering how much trouble Boston Dynamics had just getting robots to walk and the decades and billions required to do so Elon will either have to steal the technology or he's so far out of the game as to have effectively no chance.

Although I also have my doubts about the robot, you could replace that analogy with one referencing NASA and SpaceX.

 

[Wat]

A robotic arm in an industrial setting costs between 25k and 500k for just the arm. No software. No maint. No training. No computer control. Just the ARM.
You can assume 2.5%/ Cost or 10k/year for a roundabout maint figure for the ARM itself.
You can assume operational lifespan before refurb/replace of 100k operational hours.
Honestly, in real life, you don't assume you have 100% uptime either.

If the cost of the unit is less than 350k it won't be able to walk a dog let alone do a profitable amount of work.

Yes, but you are ignoring the magic of elon. Look at the cost of a traditional NASA rocket launch.. falcon9 has been disruptive to that cost structure.

Why not the same for robots?
Don't dismiss it so quickly.

Personally, I am just relieved that these robots don't use solar power. That should keep us from darkening the skies and ending up working as a battery.

 

[michalrz]

Ahh, Elon. The go-to pipe dreamer for the clueless, average folk who like news about 'science and technology'.

 

[Dayaks]

Can a robot press charges if someone molested it?

Asking for a friend.....

It might choke you to death.

 

[GoodBoy]

You have some serious engineering misconceptions here...
Firstly those "tricks" are part of mapping movement. If the robot can't move it cant operate. The first unexpected terrain element and it fails thus ending whatever job it was assigned to doing and requiring a human to intervene(non-optimal for a work replacement design).

Second, a purely human-shaped robot for labour is a gigantic mark of "I don't have a clue" from musk. Most of the difficulty in labour is due to the shape of the human body. Ever wondering why carrying heavy loads hurts your feet? Just because you CAN doesn't mean it's DESIGNED to do that job. Nature is the universe's best engineer but she designs for reliability and adaptability. Designing a robot to match nature is stupid. Nature is better. Designing a robot to selectively do things we can do better(like lifting) is sane and has useful applications.

If musk wanted to make an actual labour bot it would be something that packs into a tiny space, would have far more than 2 legs+2 arms, and would look more like a spider or ant than anything human-like. The "It can lift 150lbs" thing is also just horseshit. It's a USELESS metric. If you are using drones to do labour you have the capacity to use SWARM DRONES and have UNLIMITED LIFT. Only Musk would be dumb enough to think "labour/construction bot" and go "single platform human-like".

Human-like bots have a purpose... but it is NOT labour related.(or exclusively sex-related... get your mind out of the gutter )

Wow. It's amazing that we have a robotics expert in the forums with us!! /s

You are contradicting yourself if saying that Nature is the best designer and we shouldn't copy that. By your logic we SHOULD be copying natures' designs.

If anyone can bring a humanoid robot to life, it's Musk. He has the resources, a lot of very smart people working for him, and most importantly, the vision and bravery to try it. I would bet on him over some random forum self-appointed expert every time. What are your credentials again?

Ok fine. This question could be answered with a few quick googles but I will bite. There are already humanoid robots that do this. They suck. They are not used except by an idiot or for bad PR.

First, let's go full education mode just to make sure there are absolutely no miscommunications. Firm foundations yield knowledge transfer potential.

Define manual labour? In this do we mean, as you suggest, just moving boxes? Shelf picking/packing? Truck loading? Manual labour at a job site for construction such as but not limited to lifting bags of product to put into machines like concrete? Moving completed frames into position for workers to install?

You really think a lot of yourself, don't you?

Unfortunately, all of these definitions are partially incorrect by being grossly incomplete. Manual labor is defined as work done with physical(human) work. This is moving an object or using mechanical force to do work. In short by definition manual labour is both the stock boys' job and the ironworkers' job. These two things are completely different tasks that are all done with the same 180lb average human frame and are called "manual labour".

Human robotics are nowhere near capable of doing that breadth of work. Period. Dot. End. The material science, battery technology, and computational technology are not there yet to do the entire scope of work.

You are assuming that the robot has to have a self-contained intelligence. It doesn't. AI is progressing pretty rapidly.. give it 10 years, a reasonable time to bring to fruition Musks' vision, it will need reexamined.

So what do we do? We specialize robots to do part of the task and then focus the design exclusively on that singular job.

...snip - lots of pictures and descriptions of specialized robots...

The design tesla wants to use(and others have already done) is AWEFUL for efficiency.
Work per hour.
Work per dollar.
Work per lb.
Doesn't matter how you slice it it's a terrible design for manual labour.

Yup, Nature totally failed when it designed bipeds.

What are humanoid bots good for?
Human interactions. A hospital orderly done in human bots might be slower... but the psychological effects of seeing human people is a known net positive to health care. Murder cockroaches changing bed linens wouldn't be good here even if they would be more efficient.

I know there is this big desire for N5 type robots but in reality, they are terrible and will always be terrible. By the time they can do everything humans can do and have the processing ability to handle multiple jobs with enough variation to not be a burden rather than a boon our material science will have advanced far enough to remove the need for building them. They are a want. Nothing about them is "good" for robotics.
(note this does not include PARTS of human or animal functions)

Let's see what Musk can do. Behind every successful invention there were 10 experts saying that "it can't be done".

 

[Valnar]

Yeah I'm looking forward to the third-party unapproved add-ons.

 

[longblock454]

Look what Musk did with SpaceX, how much cheaper? And rockets are one of the hardest things on the planet.

Perhaps he can work similar magic with robots.

 

[Nafensoriel]

Wow. It's amazing that we have a robotics expert in the forums with us!! /s

You are contradicting yourself if saying that Nature is the best designer and we shouldn't copy that. By your logic we SHOULD be copying natures' designs.

If anyone can bring a humanoid robot to life, it's Musk. He has the resources, a lot of very smart people working for him, and most importantly, the vision and bravery to try it. I would bet on him over some random forum self-appointed expert every time. What are your credentials again?



You really think a lot of yourself, don't you?


You are assuming that the robot has to have a self-contained intelligence. It doesn't. AI is progressing pretty rapidly.. give it 10 years, a reasonable time to bring to fruition Musks' vision, it will need reexamined.


Yup, Nature totally failed when it designed bipeds.



Let's see what Musk can do. Behind every successful invention there were 10 experts saying that "it can't be done".

An industrial scientist who has been around for a very long time.

Nature is the absolute best designer. She creates machines from a single cell that unpack into self-maintaining self-autonomous robots that self-learn and operate for (in some cases) hundreds of years.
If you are hubristic enough to think we can even COPY that... you are not remotely educated enough. We can copy pieces. Making a hand close? Jamie Hyneman on Mythbusters did that with scrap metal. Making a hand close with variable precision like an actual human hand? We currently have the most rudimentary facsimile.

Musk is well known in all engineering circles for being an absolute jackass moron who has never once actually developed anything in his life. The only gift he has is a cult of personality. For the purposes of being a CEO, this happens to work for him. Every project tesla has created to date has been garbage except SpaceX and even then the claims of cost reduction are highly debatable. It is entirely arguable that the reduction in safety focus has reduced costs. Being cheaper than NASA isn't hard. Every other rocket team on the planet has done this. Shockingly SpaceX is comparable to them too.

If you think baselining a discussion's main talking point is me personally being full of myself you have never entered into a scientific debate. Baselining the main theme so both sides can address the issue with understanding is normal. If you read anything more into my posts than clinical removal of the subject matter you are biasing yourself.

AI is not progressing rapidly. AI is progressing at a snail's pace. AGI or any form of above insect level thinking isn't in the realm of possibility for 50-100 years and that is only IF we get a breakthrough in quantum computing during that time. Self-contained intelligence is required for complex tasks. Radio? If you suggest this you have no understanding of how amazingly complex it is to run hundreds of complex robots with radio. Complex=Cost. Hierarchy of labor = reduced costs. Right now our absolute best processing systems can do a very very select few things. Programing even a fixed arm to do thousands of things isn't really possible yet and you believe you can make a 4 limbed self-balancing robot do thousands of things?

Nature also invented the brain. A massively paralleled computer loaded with software we can only guess at. The computing power of a brain and the resulting organic sensors make bipedalism possible. Boston Dynamics spent DECADES trying to get a robot to walk down stairs and every single baby on the planet comes preloaded with software to learn this in days on its own.
Knowing a thing is possible is the first step. We knew cellphones were possible in the 1930s. It took 50 years to make it a common reality. The human form is also not optimized for work. For a working robot, it is actually terrible. Insects pretty much demolish our ability to do work. Horses do as well. It's why humans make tools and domesticated animals... because WE personally are not specialized for work beyond agile tool making.

The issue with Musk is that he doesn't actually invent anything and he's a prolific liar and cheat. REAL scientists damn well know there are 10 people saying it can't be done. Those are our COWORKERS pointing out flaws we missed. We WELCOME the pointed educated criticism because it eliminates things we have to test ourselves. How about we stop obsessing over the human robot and actually design things that are less expensive than just hiring a thousand cheap laborers in china? How about we finally accept that eliminating cheap jobs is the first step to improving everyone's quality of life?
------------
Let's switch gears. Look at this video:

Spoiler

}
This is exactly what you suggested earlier in the post. They've removed the processing entirely as it is exclusively telepresence.
Notice how the robot isn't even remotely close to what a human can do? Notice how rough the movements are for something controlled by a human doing all the calculations? The lag due to remote operation? How do you expect to overcome physics here and fix these issues?

Spoiler

Agility robotics. Arguably one of the top guys for a humanoid bot. This is digit. I like digit. I follow his development because it's an actual cheap robot doing a decent job.
Do you see anything in this video that would make anyone ever want to replace a human being? Notice all the huge problems they have with this simple task?
Notice how they stripped a ton of "human" out of the design to make it work and ignored things like hands? Almost a decade of work here.

Spoiler

The final example. A robot where they designed it entirely around "safe operating space". They had absolutely no reason to worry about humans being in the robots operating space so it can swing and move as fast as it wants outside the exclusion zones. Yet they still have to do a ton of work away from the robot to get it to make hundreds of drinks and its throughput is still comparable to a human bartender.

When all the facts tell you Elon has very little chance to overcome the physics and engineering issues everyone else(with decades of experience) hasn't solved yet... you do the scientific thing and look upon his claims with an extreme dose of cynicism.

 

[NickM]

An industrial scientist who has been around for a very long time.

Nature is the absolute best designer. She creates machines from a single cell that unpack into self-maintaining self-autonomous robots that self-learn and operate for (in some cases) hundreds of years.
If you are hubristic enough to think we can even COPY that... you are not remotely educated enough. We can copy pieces. Making a hand close? Jamie Hyneman on Mythbusters did that with scrap metal. Making a hand close with variable precision like an actual human hand? We currently have the most rudimentary facsimile.

Musk is well known in all engineering circles for being an absolute jackass moron who has never once actually developed anything in his life. The only gift he has is a cult of personality. For the purposes of being a CEO, this happens to work for him. Every project tesla has created to date has been garbage except SpaceX and even then the claims of cost reduction are highly debatable. It is entirely arguable that the reduction in safety focus has reduced costs. Being cheaper than NASA isn't hard. Every other rocket team on the planet has done this. Shockingly SpaceX is comparable to them too.

If you think baselining a discussion's main talking point is me personally being full of myself you have never entered into a scientific debate. Baselining the main theme so both sides can address the issue with understanding is normal. If you read anything more into my posts than clinical removal of the subject matter you are biasing yourself.

AI is not progressing rapidly. AI is progressing at a snail's pace. AGI or any form of above insect level thinking isn't in the realm of possibility for 50-100 years and that is only IF we get a breakthrough in quantum computing during that time. Self-contained intelligence is required for complex tasks. Radio? If you suggest this you have no understanding of how amazingly complex it is to run hundreds of complex robots with radio. Complex=Cost. Hierarchy of labor = reduced costs. Right now our absolute best processing systems can do a very very select few things. Programing even a fixed arm to do thousands of things isn't really possible yet and you believe you can make a 4 limbed self-balancing robot do thousands of things?

Nature also invented the brain. A massively paralleled computer loaded with software we can only guess at. The computing power of a brain and the resulting organic sensors make bipedalism possible. Boston Dynamics spent DECADES trying to get a robot to walk down stairs and every single baby on the planet comes preloaded with software to learn this in days on its own.
Knowing a thing is possible is the first step. We knew cellphones were possible in the 1930s. It took 50 years to make it a common reality. The human form is also not optimized for work. For a working robot, it is actually terrible. Insects pretty much demolish our ability to do work. Horses do as well. It's why humans make tools and domesticated animals... because WE personally are not specialized for work beyond agile tool making.

The issue with Musk is that he doesn't actually invent anything and he's a prolific liar and cheat. REAL scientists damn well know there are 10 people saying it can't be done. Those are our COWORKERS pointing out flaws we missed. We WELCOME the pointed educated criticism because it eliminates things we have to test ourselves. How about we stop obsessing over the human robot and actually design things that are less expensive than just hiring a thousand cheap laborers in china? How about we finally accept that eliminating cheap jobs is the first step to improving everyone's quality of life?
------------
Let's switch gears. Look at this video:

Spoiler

}
This is exactly what you suggested earlier in the post. They've removed the processing entirely as it is exclusively telepresence.
Notice how the robot isn't even remotely close to what a human can do? Notice how rough the movements are for something controlled by a human doing all the calculations? The lag due to remote operation? How do you expect to overcome physics here and fix these issues?

Spoiler

Agility robotics. Arguably one of the top guys for a humanoid bot. This is digit. I like digit. I follow his development because it's an actual cheap robot doing a decent job.
Do you see anything in this video that would make anyone ever want to replace a human being? Notice all the huge problems they have with this simple task?
Notice how they stripped a ton of "human" out of the design to make it work and ignored things like hands? Almost a decade of work here.

Spoiler

The final example. A robot where they designed it entirely around "safe operating space". They had absolutely no reason to worry about humans being in the robots operating space so it can swing and move as fast as it wants outside the exclusion zones. Yet they still have to do a ton of work away from the robot to get it to make hundreds of drinks and its throughput is still comparable to a human bartender.

When all the facts tell you Elon has very little chance to overcome the physics and engineering issues everyone else(with decades of experience) hasn't solved yet... you do the scientific thing and look upon his claims with an extreme dose of cynicism.

You really have no clue, do you?

 

[Wierdo]

I thought the DOJO system that they're going to use to train their AI solutions was the star of the show personally. You're gonna need some serious power to accelerate that kinda machine learning, and it looks like they got a very specific system in the works for it, the details of the setup are pretty impressive:



1.1 Exaflops for NN workloads, 10^18 operations, in a fully integrated package (cpu, cooling, comm, etc). That's a pretty cool design.

 

[Nafensoriel]

You really have no clue, do you?

Tesla cars=Well known for being overly complicated(otherwise known as shit) engineering-wise beyond the battery. It seems their engineers only recently discovered unibody designs.
Autopilot= Musk himself says it's terrible. Has never delivered on promises. Isn't even close to delivering on promises.
SpaceX= Despite investment being nearly as much as their 10 nearest competitors(ULA/Blue/Rocketlabs/etc) combined they are not actually significantly cheaper than each other. This is his absolute best engineering team as well. They've actually made real advances in rocketry. They are nowhere near revolutionary when compared with peers however beyond actually having a budget.
Solar roofs=Late, nowhere near advertised specs, way over budget, no cost savings to traditional EV panels.
Neuralink=Ok. Actual good stuff here. No actual musk touching the project though.
Tesla Batterys=He sold making a cell 50% larger getting 50% more capacity and people thought it was revolutionary.

Musks actual patents:
https://patft.uspto.gov/netacgi/nph...=Refine Search&Query= in/(musk) and in/(elon)

Tesla's actual motor patents:
https://patft.uspto.gov/netacgi/nph...efine+Search&Query=an/(tesla)+and+an/(motors)

Don't get me wrong Musk actually has talent.... in software(zip2? First ebank?). All his greatest successes have been in code and in that realm he's pretty good even by today's standards. Physical engineering though? No. He's not all that great.

 

[Nafensoriel]

I thought the DOJO system that they're going to use to train their AI solutions was the star of the show personally. You're gonna need some serious power to accelerate that kinda machine learning, and it looks like they got a very specific system in the works for it, the details of the setup are pretty impressive:



1.1 Exaflops for NN workloads, 10^18 operations, in a fully integrated package (cpu, cooling, comm, etc). That's a pretty cool design.

This part...
Is actually cool though not as revolutionary as they are claiming.
The system is about what.. 3.5/4.5x? a Fugaku and they effectively just launched. If they aren't BSing and this actually is ready to go next year... Props it's a good improvement.

 

[Hagrid]

Tesla cars=Well known for being overly complicated(otherwise known as shit) engineering-wise beyond the battery. It seems their engineers only recently discovered unibody designs.
Autopilot= Musk himself says it's terrible. Has never delivered on promises. Isn't even close to delivering on promises.
SpaceX= Despite investment being nearly as much as their 10 nearest competitors(ULA/Blue/Rocketlabs/etc) combined they are not actually significantly cheaper than each other. This is his absolute best engineering team as well. They've actually made real advances in rocketry. They are nowhere near revolutionary when compared with peers however beyond actually having a budget.
Solar roofs=Late, nowhere near advertised specs, way over budget, no cost savings to traditional EV panels.
Neuralink=Ok. Actual good stuff here. No actual musk touching the project though.
Tesla Batterys=He sold making a cell 50% larger getting 50% more capacity and people thought it was revolutionary.

Musks actual patents:
https://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=/netahtml/PTO/search-adv.htm&r=0&f=S&l=50&d=PTXT&RS=IN/musk&Refine=Refine Search&Query= in/(musk) and in/(elon)

Tesla's actual motor patents:
https://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=/netahtml/PTO/search-adv.htm&r=0&f=S&l=50&d=PTXT&RS=(AN/tesla+AND+AN/motors)&Refine=Refine+Search&Query=an/(tesla)+and+an/(motors)

Don't get me wrong Musk actually has talent.... in software(zip2? First ebank?). All his greatest successes have been in code and in that realm he's pretty good even by today's standards. Physical engineering though? No. He's not all that great.

But at least he is pushing the limits and that is needed. Then it will push other companies/people to do the same, hopefully.
Remember that he may not be delivering everything he says, but that is also part of the learning process. Knowing what does not work.
We need more people like him!

 

[Wierdo]

This part...
Is actually cool though not as revolutionary as they are claiming.
The system is about what.. 3.5/4.5x? a Fugaku and they effectively just launched. If they aren't BSing and this actually is ready to go next year... Props it's a good improvement.

It's dependent on context. The Fugaku setup is more general purpose, so it'll likely win in scientific computing loads and such, for example.

The DOJO system is very specialized, its primary purpose is NN, that's the overriding design goal. So will likely dominate strictly for NN applications:
https://semianalysis.com/tesla-dojo...gnitude-advantage-over-competing-ai-hardware/

Someone estimated that DOJO (v1) may still place in the top five regardless, but yeah not a one size fits all computing solution relative to others on that list. Just something to keep in mind.

 

[Wat]

1-0-0-1-0-0-1
S.O.S.
1-0-0-1-0-0-1
In distress

 

[AceGoober]

If that robot can take out the trash, clean the house, walk the dog and play a good game of chess, I'd buy one. It'd allow me more time to further my education and pursue projects I've had on the back burner for quite some time.

 

[pendragon1]

 

[Nasty_Savage]

And some people wonder why I think we should have high powered weapons...

I'll take a phased plasma rifle in the 40-watt range.

Bah, some 60 lb test fishing line will do the trick. Then you can charge your cell phone for the next 20 years

 

[mischief08a]

If that robot can take out the trash, clean the house, walk the dog and play a good game of chess, I'd buy one. It'd allow me more time to further my education and pursue projects I've had on the back burner for quite some time.

I think thats my role as a hubby in my house now.

 

[kamxam]

Maybe they'll make this one next?