Teslabot has WON The RACE Already! Humanoid Robots are Essentially a Solved Problem Now!! AI Day 2

It's time to take a step back and realize that Tesla has Already Won the race to a functional humanoid robot with flexible capabilities! How can what we saw during AI Day 2022 give me confidence that this is in fact the case? It comes down to hardware, software, and data. And I believe Tesla engineers have already pretty much solved each one of these major challenges.
The Florida TeslaCon event details can be found here: www.teslaownersflorida.org/ev...
**To become part of our Patreon team, help support the channel, and get awesome perks, check out our Patreon site here: / drknowitallknows . Thanks for your support!
**Want some awesome Dr. Know-it-all merch, including the AI STUDENT DRIVER Bumper Sticker? Check out our awesome Merch store: drknowitall.itemorder.com/sale
For a limited time, use the code "Knows2021" to get 20% off your entire order!
**Lift pucks for your Tesla: amzn.to/3vJF3iB
**Emergency tire fill and repair kit: amzn.to/3vMkL8d
**If you're interested in a CO2 Monitor, you can check it out here: amzn.to/3PsQRh2
**For 30% off the amazing OneCarStereo instrument display for Tesla Model 3/Y, use this link: bit.ly/3vd7NRv
**Check out Artimatic: www.artimatic.io
**Get 20% off an awesome Ekster wallet and/or wallet locator(!), check out my affiliate link here: shop.ekster.com/DrKnowitall
**Want to get in on WeBull's awesome Crypto and stock fun? Check out this link, and get started trading stock and Crypto!
a.webull.com/i/DrKnow-it-allK...
**If you are looking to purchase a new Tesla solar roof, solar tiles or PowerWall, just click this link to get up to $500 off! www.tesla.com/referral/john11286. Thank you!
**You can help support this channel with one click! We have an Amazon Affiliate link in several countries. If you click the link for your country, anything you buy from Amazon in the next several hours gives us a small commission, and costs you nothing. Thank you!
* USA: amzn.to/39n5mPH
* Germany: amzn.to/2XbdxJi
* United Kingdom: amzn.to/3hGlzTR
* France: amzn.to/2KRAwXh
* Spain: amzn.to/3hJYYFV
**What do we use to shoot our videos?
-Sony alpha a7 III: amzn.to/3czV2XJ
--and lens: amzn.to/3aujOqE
-Feelworld portable field monitor: amzn.to/38yf2ah
-Neewer compact desk tripod: amzn.to/3l8yrUk
-Glidegear teleprompter: amzn.to/3rJeFkP
-Neewer dimmable LED lights: amzn.to/3qAg3oF
-Rode Wireless Go II Lavalier microphones: amzn.to/3eC9jUZ
-Rode NTK Studio Microphone: amzn.to/3vjMomW
-Focusrite Scarlette 2i2 audio interface: amzn.to/3l8vqDu
-Studio soundproofing tiles: amzn.to/3rFUtQU
-Sony MDR-7506 Professional Headphones: amzn.to/2OoDdBd
-Apple M1 MacBook Pro: amzn.to/3wPYV1D
-Docking Station for MacBook: amzn.to/3yIhc1S
-Philips Brilliance 4K Docking Monitor: amzn.to/3xwSKAb
-Sabrent 8TB SSD drive: amzn.to/3rhSxQM
-DJI Mavic Mini Drone: amzn.to/2OnHCEw
-GoPro Hero 9 Black action camera: amzn.to/3vgVMrH
-Suction car mount for camera: amzn.to/3tcUfRK
-Extender Rod for car mount camera: amzn.to/3wHQXsw
**Here are a few products we've found really fun and/or useful:
* NeoCharge Dryer/EV charger splitter: amzn.to/39UcKWx
* The Humane Interface, by Jef Raskin: amzn.to/3ajM7YT
* AI Superpowers, by Kai-Fu Lee: amzn.to/3dgRSs3
* A camping mattress for your Tesla model S/3/X/Y: amzn.to/3chSRrQ
*And of course the amazing "Elon Musk: Tesla, SpaceX and the Quest for a Fantastic Future": amzn.to/36hqdmS
**Music by Zenlee. Check out his amazing music on instagram -@zenlee_music
or KZbin - / @zenlee_music
Tesla Stock: TSLA
**JOWUA:
*For an awesome wireless charging phone holder mount, check out: bit.ly/3tRMSQR
use the Promotion Code: DRKNOW5
*Tesla USB Hub: bit.ly/2PXHPit
use the Promotion Code: DRKNOW5
*Powerful wireless iPhone/Android charger for your Tesla Model 3 or Model Y, check this link: bit.ly/3b5fHSW
use the Promotion Code: DRKNOW5
**EVANNEX
Check out the Evannex web site: evannex.com/
If you use my discount code, KnowsEVs, you get $10 off any order over $100!
**For business inquiries, please email me here: [e-mail]
Twitter: / drknowitall16
Also on Twitter: @Tesla_UnPR: / tesla_un
Instagram: @drknowitallknows
**Want some outdoorsy videos? Check out Whole Nuts and Donuts: / @wholenutsanddonuts5741
Related videos of mine to check out:
Teslabot mechanics: • TeslaBot Hands REVEALE...
GOOGLEBOT vs TESLABOT!
• GOOGLEBOT vs TESLABOT!...
Occupancy Networks:
• Don’t Crash, EVER!! TE...
Tesla AI using PANOPTIC SEGMENTATION:
• Tesla AI using PANOPTI...

[@YouT-DJ]

I'm in agreement. On a manufacturing floor at a specific station or a mockup of a work station... once an Optimus has been trained to run that job - all Optimus can inherit the training, they are all trained! Around the world. Over the air updates. Build 1000 Optimus and set them to learning tasks. Say they are slow learners and learn one task a month. In one year you have 12,000 tasks learned that any Optimus unit can perform. My God this is going to happen fast.

[@sgfx]

As with humans, I don't believe there will ultimately be just one bot. I believe there will be stronger bots, faster bots, some with more dexterity, and so on. There may be a few that have all the advanced features. But I feel there will be bots that are designed or configured to do a given job type. There may even be bots that have interchangeable parts depending on task. They will all be based on the same main concept and technologies. And some may not have the human form at all. There should not be a limit to what form or shape they take. Being Humanoid is but only one option. In a way, like you see in Star Wars. Maybe not that extreme, but who really knows?

[@Nightthefirst]

I think they will have one bot, as they develop more and own more factory, they gonna upgrade the bot, like iPhone, Tesla bot 1, Tesla bot 2, Tesla bot 3, maybe Tesla bot 10 would be stronger faster

[@billyrocket62]

100% agree! I've been thinking the same thing. If the robot only has to move on a smooth factory floor, then why not on wheels? If it is going to be installing the same bolt over and over, replace one hand with a powered wrench, etc. The same eyes, software, battery, body, etc... For mass production, but customizable for specific jobs.

[@stanleyasmith]

I think you may be right. Not everyone is a weight lifter or an artist.or a sprinter. Having all robot be identically the same would be a waste. A robot whos job is to assemble small items don't need to run fast.

[@casperhansen826]

I think the point is to have one robot with the ability to use tools and power tools if needed, I don't think that they will have any individual special training

[@omnicurious2949]

You don't believe in the one true bot? Repent you blasphemer!

[@rowland5951]

Your jumping the gun here Johm early days still.

[@thorddespace2773]

I would have liked to be 20 years old now seeing all the possibilities.

[@davidmacphee3549]

Same here. I'm 68

[@joshgray1331]

Great stuff! Re. the hands: they are brilliantly simple! I have made more than a dozen robotic / animatronic hands for large companies and this strategy is by far the most simple. Brilliant use of a worm gear at the proximal driving joint. When you close the grip to hold something heavy or put substantial torque on that joint you can power down that motor! Worms are great for this self-locking attribute, and it saves Teslabot a lot of power doing heavy lifting. Also the design is only one DoF (proximal joint, no ADD/ABB). The second and third phalanx are free (torsion spring return, true). This is passive compliance. No sensors / programming is required to have those fingers wrap around a cylinder for example. This of the finger as more of a tentacle...the base engages the object first then the rest of the tentacle wraps around. Very, very smart, LEAN, and elegant design choices. Plus...it looks cool!! THX

[@BrianBellia]

I agree. I've been waiting for Optimus for over 50 years now and even though I had very, very high expectations for AI Day 2022, I'm impressed with what I've seen so far.

[@IronFreee]

What part of that robot impressed you? What innovation have you spotted?

[@BrianBellia]

I'm impressed with how much progress they've made in such a short time. Aren't you? I understand the point that you're making here, however. Optimus is just like all the others. Which is perfectly true. But he's the most human-like robot I've seen that can walk on two legs - and he's only 8 months old!

[@IronFreee]

​@@BrianBellia I expected something even more ridiculous, but I'm not impressed. There is no innovation. It's like if Steve job had introduced the IPhone by showing a prototype that does less than the competition. "But he's the most human-like robot I've seen" I don't think you've seen many robots, because there are plenty that look much more human like. But I don't think that's a criterion for being useful (unless you want to have an intimate relationship with it?). The reason there are no human-like robots in factories is that it's a stupid idea. If you need an arm to build or load stuff, you just build an arm. It just seems like a new idea that will never be viable like almost everything Musk promises...

[@BrianBellia]

@@IronFreee I have seen plenty of human-like robots, and they're little more than animated mannequins. Their faces and arms move, but other than that, they're immobile - and they have no brain! Optimus can actually perceive the world around him - that's a huge deal. And clearly, Tesla has more in mind for Optimus than just a mere factory helper - he's being designed for consumer use; for things like cars and homes where having human proportions and human agility is of paramount importance. But I guess we'll see.

[@IronFreee]

​@@BrianBellia Yes, most other human-looking robots are not designed to move around. But Tesla hasn't solved the problem of navigation for cars and it's much harder to do for a robot that is supposed to walk in all kinds of environments. Do you call computers "brains"? Or is that specific to this one :D "he's being designed for... things like cars and homes" I suggest you think about what you are saying... Having a car driven by a robot adds so much more difficulty to the task :D As for houses, I already have robots, one is a blender to make food, another cleans the floor. I even have some "brains" that I can give orders to like "make me a coffee", "start the washing machine now". One of the "brains" knows when I come home or leave and adjusts the lights, music, blinds... accordingly. I don't need another body taking up space for nothing.

[@nobrien1]

I see it very differently. In my career in tech, working on some projects that were leading edge at that time, the old corollary to the Pareto Principle proved to be very true. Specifically, that the first 80% of a project is accomplished with 20% of the effort and the last 20% of the project requires 80% of the effort (I would add that the last 10% might require 50% of the effort.) Example: Tesla FSD - the hardware has been done for 5 years but we still don’t have true self-driving and it is not really clear when we might. And, FSD is a relatively small subset of operating in the overall world which means that Optimus’s AI will need to be an order of magnitude ‘smarter’; perhaps even something approaching generalized AI to achieve Musk’s vision of a very capable personal assistant. Might it be able to do limited, repetitive tasks in the next couple years? Sure, but there are off the shelf robots that do that today. Look at Tesla’s manufacturing line as one example. To be clear, I am in favor of what Tesla is trying to accomplish and I applaud their effort. I just think it is highly optimistic to think that the competition is ‘over’ and that we will see something substantive any time soon. (We are operating on ‘Elon Time’ after all.) I’m guessing that Dave Lee will do a discussion with James Douma about AI Day and I will be interested to hear what James’s take on the timeline might be.

[@davidbeppler3032]

FSD is the last 20%.

[@Hyperlooper]

FSD has to be nearly perfect to be useful. The robots can afford to not be perfect because the dangers are generally a lot lower.

[@kkulkulkan5472]

Agreed this is just the beginning and the robot hardware is going to become commodity like EV cars over time, so it is all about the AI and software. Vision AI is a good starting point but need AGI over time which is orders of magnitude harder. Simple fixed tasks like collecting or sorting garbage (google’s AI prototype robot) or moving things around in a factory maybe doable within a few years, but I think we are more than a decade away from seeing one in our home. Maybe Tesla will buy OpenAI to accelerate the efforts.

[@davidbeppler3032]

@@Hyperlooper FSD only needs to be 1% better than the average driver. The average driver blows stop signs/lights, rear-ends stopped cars in front of them, drives into shops over parking blocks, and hits construction workers/police daily.

[@Hyperlooper]

@@davidbeppler3032 your point is completely reasonable, but the thing is, people will accept grandma running someone over because we understand that humans aren't perfect. People won't accept FSD doing the same. I think FSD will need to be two or three times better than a human driver, even though that is not really rational.

[@lwdp74]

When the robot recognized a watering can, picked it up, then recognized a flower, which it then watered! That blue me away. That’s when I remember a utube about character recognition for seeing about driving. Tesla’s been creating and as analyzing ai for over 20 years.

[@craigruchman7007]

You nailed it! Thanks for a great video to counter the FUD.

[@randreas69]

I would rather like to see a humanoid robot take on the manual shifted cars we're still used to here in Norway, with a hill start, and split icy-ness conditions. I was a postman for 11 years and it's hilly here.

[@biovmr]

I feel like I'm in the position for TSLA stock to pay for all the Tesla products I'll eventually want. Invested heavily in TSLA just before it took off (Jan 2020), and consistently added shares since. Today, I'm still driving Honda (ICE) Fit for another couple years, then it's time to upgrade to EV. Probably the same or next year will upgrade to solar/powerwall. Then 6 years after that (when I'm retired), get a Tesla robot to do chores I'm not physically able to do anymore (or simply do not WANT to do), and have the robot do jobs around neighborhood for reasonable price (whatever that is at the time, who knows?)

[@jasonlam5738]

I’m in a wheelchair. Things like this will dramatically improve my life. Excited for the future

[@garethrobinson2275]

I'm quite similar but a little ahead of you. I put my pension into Tesla in June 2019, have had a M3LR for two years, have taken early retirement, sold my house and am gradually using the funds to buy more Tesla. I was thinking I would get a Roadster 2 next but now I want a Tesla bot more! In part it's a pay back to Tesla for changing my life in such a fantastic way. Onward to the future!

[@jasonlam5738]

Cheers to the future and one day owning a Tesla bot 😄👏🏻

[@sgfx]

I like to see a smart "Teslabot" wheelchair that could assist in mobility and even have arms for those who need such help. Imagine a chair that could navigate tight areas safely and quickly, or one that could assist in feeding or reaching for doors and drinks or such, for a person who has limited functions. Adding in a self-driving Tesla van, could really give some people new freedom.

[@cfkelty]

Very well presented. I appreciate your insight and agree with your conclusions. The robot presented and aiday2 is truly monumental when considered in its entirety. Very exciting times we are living in

[@garyswift9347]

I love your videos, but have you ever been in a factory? I think the challenges are WAY bigger than you think. Thanks for another great video. Keep it up, and I'll keep watching. I taught Lean Mfg for many years, so I have a good grasp of the challenges here, and it's not trivial. I'm also a programmer, btw.

[@TechViewOpinions]

Agreed! Glad to see I'm not alone in the excitement of what is going on with Teslabot!

[@dannylemmon7123]

They could make it somewhat modular so if your use-case requires more hand dexterity, just pay for upgraded hands/arms.

[@martincossette7420]

More dexterity will come in time... as an OTA update ;-)

[@roryreddog3258]

Hey Dr KIA, appreciate your intelligent insights into so many general areas, your great economy with words and your high-energy delivery. Respect ✊ Don’t have time to watch 1 and 2 hour ramblings from others.

[@wholmes7177]

Love to hear your thoughts on SpaceX using FSD tech to land Starship and Super Heavy.

[@larslysdahl4586]

Great video thanks!

[@dareldavies7348]

I agree this is going to progress rapidly. That's why they said that they might move to monthly updates.

[@BrianBull]

YIKES! Big words. Can't wait to see your video in 365 days to see what comes about!

[@ArendJanV]

Love your take! Thanks

[@ebinrajmon5532]

Great video Doc! Thanks. Tesla is going to be literally creating the future right before our eyes!

[@thomasaquinas157]

Agree 100%. Great buying opportunity since most people don't seem to understand it, comparing Optimus to 'brainless' Boston Dynamics robots.

[@garethrobinson2275]

Brainless, handless and almost pointless. However their physical capabilities will be interesting when paired with the right brain and hands!

[@terrulian]

Brilliant. One of your best.

[@teslahype]

Love the optimism 🤖

[@wwrussell180]

Excellent stuff!

[@sandmehlig]

Decoupling varieties of the hardware and "higher" functions is going to be a major task - if they are going for a monolithic ANN and not a hybrid strategy.

[@MrDuncanBooth]

Thanks John

[@streetmoney21]

Great video

[@CrumResearch]

On point, spot on. Love the discussion of the robot hand, you are right. The Utah / M.I.T. Destrous Hand project was just a hand. You may have dismissed the possibility of playing piano too quickly. With a bit of evolution, Optimus will hopefully play piano as well as violin, I'd guess!

[@andreasklossek9252]

3:35 That would imply, that only special parts could make the job... which obvious can do more, off the shelf ^^

[@BrianBull]

11:50 But the cars on FSD doesn't have "On Device Learning". It seems the bot will need on device learning. The presentation at AI day 2 didn't seem that was there approach unless I misunderstood.

[@winstonmontgomery8211]

Boston Dynamics robots are all programmed to move and do flips where as I think the point you were trying to make is that Telsa Robots are fully autonomous and can move and sense the world around it in real time.

[@omnicurious2949]

Simulation is helpful in increasing the 'training dataset' exponentially (by varying colors, lighting angles, positions etc) BUT you need real world data to start with. A totally hand built simulation world will only make the robot fail in real world! My thinking/suggestion would be for Tesla to hire 50-100 folks whose job is for each one to stay at a different Airbnb/Vrbo each day....walk around the place with head-mounted-cams and manipulate objects using pressure/position sensing gloves (manipulating includes opening/cleaning windows, heating something in a microwave oven, making bed etc). Within months Tesla can have 10s of 1000s of unique household environments. Initially a HUMAN labelling team is also required to label the thousands of unique items. Eventually auto-labelling may be employed. Simulation will accelerate the training by mixing up objects/colors/lighting in various houses.

[@mjnelson8655]

The question/observation it should be is. How many of them have to be made, until they can start making their self? I expect it's around 50.

[@ralphacosta4726]

That was always my plan if i built a robot - make the first ones to be capable of building more.

[@Madchuck42]

nailed it... you get "IT"...

[@Myrslokstok]

Love the backdrop! But Optimus walked on a dark sceen where the light changed from the background, not like boston dynamic labb or outdoor stuff. Not to take anything from Boston they are ofcourse way ahead jet, lets wait two years and see.

[@MrAalestrup]

I was a little disappointed when I first saw it. I probably expected a lot more. But hopefully as I re-watched those clips again and again, I am far more fascinated by the product. Ex. the movements the robot makes are very similar to what we humans do! See how it keeps its balance when it moves its arms, which after all shifts the center of gravity and it compensates for this by rotating the hip joint. It is fascinating to watch. It must be endlessly exciting to be on that development team... I'm envious, but I have to admit that my abilities are not enough. I will look forward to your next presentation. Hope the speed is up when it goes. And that it might want to walk more upright instead of sinking to the knees.

[@michaelcoughlin8238]

"It must be endlessly exciting to be on that development team..." that was the point of the entire day, recruiting. Well said. Others focused too much on what the robot demonstrated on that particular day instead of the potential as this video stated. The physical part is over, now the software updates begin!

[@docforest4851]

Love your presentation. I’m not quite convinced, but I’m getting there..lol.

[@garycarson3128]

Great Episode John. You need to send this to Warren Redlick. He thinks the data gathering required for the Tesla Bot is similar to the real world data required by FSD but in reality, a very useful Bot can be designed with different versions designed for specific jobs - factoty Bot, roofing Bot, nursing home Bot, etc. Each of these Bots will have different well defined, controlled and relatively static micro environments. Each Bot doesn’t need to know about the whole world environment to be useful and profitable to Tesla and their customers. Years later these hundreds of micro environments can be combined into all Bots.

[@keitho9508]

I hit the thumbs up button because, as it says, "I like this" but I really wanted a 10x Like button! Great video as usual. Few understand or "get" NNs and FSD and fewer get Tesla Bot and how significant she is. You are un-fogging the lenses. ✴

[@mragendds]

Are there any gyroscopes in the bot?

[@jimknoll]

perspicacity wow you fancy. (thanks for my new word of the day)

[@markchambers5729]

I have a lot of comments in my head. One is the possibility that Tesla could not only advance humanoid robots, but this could parallel with advancements for prosthetics for humans. Consider coupling prosthetics with neural link tech (not implants into the brain, signals from the nerves translating to prosthetic movement). Imagine, the 6 million dollar man or the bionic woman could cost more like 10-50k.

[@dannygjk]

0:17 top right a flashback to a black and white movie I watched when I was a teenager. 👍🏼

[@hopper2716]

*looks at stock price* 😓

[@Mrbfgray]

RELAX...whatever gains don't happen sooner WILL happen later. Currently setting up for major upside in next 18 months or so, price action is just noise.

[@terryeasterday580]

Buy the dip... When the price goes down, look at it as a buying opportunity.

[@rogerstarkey5390]

Just remember there are lots of "people" who want the stock low. They will push hard for that, but they won't hold it down for ever

[@andreasklossek9252]

10:59 There many things, that can happen in a factory, but not in the real world :P

[@stcredzero]

If highly specialized neural nets can be developed for specific contexts: eg. Factory Work, domestic cleaning, walking on city sidewalks, etc... then wouldn't it make sense for the Tesla bot to have a large hard drive or SSD where it could store these specific neural nets?

[@macrumpton]

I wouldn't call the race yet, although considering the lead in AI, actuators, manufacturing and materials expertize and a built in market for the product, I am not sure who can match that.

[@Hyperlooper]

The thing is, no one else is even seriously trying to do what Tesla is trying to do with the robots.

[@ethangillese]

Hey John, great video as always, I have a couple questions for you if you’ll indulge me. 1.) when do you think we’ll see basic humanoid home robotics for consumers? (roughly) by that I mean, you download a clip from wiki how onto the robot via microSD to do repetitive basic tasks (like folding clothes, dishes etc) 2.) based on an alleged timeline for humanoid robots in the home and our daily lives, will the need for a robot to have legs become redundant? The UN and many countries around the world have commitments to make their respective countries fully accessible to people living with disabilities. So would it be worth the extra computing power to have a robot with two legs if a robot on wheels could perform just as well if not better provided it’s operational domain is fully accessible (to a wheelchair for example) Thanks!

[@Discoducky73]

While I'm not John, I'll give these a shot since they are great questions 1. I'm thinking it will be more like this. You demonstrate a task for the bot, it attempts it modestly, you provide it feedback or you ask it to try again or you repeat again, at some point in the iteration, it will make a decision to upload to the training computer, aggregate the taught task positives and negatives and attempt to update it's NNs to complete the task. 2. Legs are important as we will expect it to reach high and low and get into tight spaces. Also, walking isn't that much more, if any, power intensive as other forms of locomotion. The bot will be optimized to use its own weight to propel in a more natural and efficient motion.

[@meelis79]

But when robot has task to assist/help person in wheelchair, then legs are needed.

[@jasonlam5738]

The idea of a humanoid robot with legs is incredible. If you think about it, it could fit itself in a world that’s designed for humans. I’m in a wheelchair so it’s exciting to see how this bot can help with basic things.

[@richardgoldsmith7278]

Refinement. Yes, thats all thats left to be added. The fact that is will lead to AGI is beyond the race to the humanoid robot that is useful.

[@AudiTTQuattro2003]

...so which will come first, FSD of the robot? Robotaxi was promised years ago and still no sign it will be available any time soon.

[@richardgoldsmith7278]

@@AudiTTQuattro2003 correction: no sign that you have noticed.

[@navg42]

What matters is that the company with the deepest resources is actually trying to solve the problem, I could care less how long it takes, just that it happens. That takes action.

[@monikapatel5875]

"Its too late, always has been, always will be, too late." - Dr. Manhattan

[@krisvandermeulen253]

I can definitely see the need for a Tesla robot trainer, someone who know how to best tell or train the robot on site what it must do and how. Simply because it would decrease the learning time of the robot.

[@papercrowe8772]

The problem is that the majority of this is “it could do this in the future” right now this is an incredibly simple robot compared to what we already have. The biggest positive is that it *may be* using neural networks. But it seems unlikely that it wasn’t just a preprogrammed walk path. They only demonstrated its capabilities in a HEAVILY edited few second clip. (You can see in the background of the original that all of the things like people and computers are skipping around because it was filmed in a bunch of different small segments.

[@mzrts9333]

0:03 the time reference is incomplete without timezone

[@fredpsimas1874]

Tesla Not only using custom but electric actuators…electric much more precise than hydraulic!

[@macrumpton]

Much lower maintenance and weight as well.

[@maxhvargas]

You nail it! 🗿

[@kstaxman2]

Most don't understand that a functional robot is much different than a research robot. It's like looking at an Olympic gymnast and an everyday person. That gymnast has amazing abilities and does things impossible for all but a few carefully trained people to do . But the every day person can do everything needed to get by without having the skills of a gymnast. Those skills simply aren't needed to do normal tasks. Tesla realizes that and is very happy to let others make the gymnast robot that will be ready in the future. They are building the robot to do everyday things in the next year or so.

[@jwmc41]

Some were very dismissive of the watering can demo but is it true that it would still have worked if the can were moved?

[@andreasbjerreovergaard4641]

It was not a demonstration but a concept video. The clips that makes up the video are not related and mostly animation graphics. We have not seen it interact with anything yet.

[@johnleung9619]

How about a Teslabot sitting on driver's seat? Will it speed up the humanless driving ?

[@rwhirsch]

i hope the first optimus goes into the Smithsonian.

[@kevryfabroni]

agree!!!

[@andreasklossek9252]

2:52 Boston Dynamics? ^^

[@vuththiwattanathornkosithg5625]

Agree

[@Volte6]

Did he just drop "perspicacity" on us?

[@rogerstarkey5390]

Needed a Daffy Duck voice

[@avocade]

Agreed.

[@rlbf1967]

For what my thoughts are worth I’ll watch the development and see how it goes. Taking a look at how animals have developed over evolutionary time they all have converged on continuous real-time learning of one sort or another based on nervous system structure very different from the abstract models in common use throughout attempts at AI today. Therefore I imagine Tesla’s humanoid robot design will be upgraded with actual machine intelligence architecture as it is refined by external research. Additionally, extending that observation, tactile sensors embedded in new materials may be needed in home care applications as useful feedback when helping an infirm person. These things are only a couple out of many aspects of a functional humanoid robot, so although TeslaBot may be a few generations behind Atlas and Asimo, all of them have a very long way to go. The race isn’t finished, the starting pistol was fired back in 1970 with WABOT-1 at Waseda University, and the players are just finding their feet (and hands) at the same time as oscillating over whether to grow a brain or not ;-)

[@aikidoshi007]

You're right, from here everything is a refinement. That is one thing Tesla does really well. As with the cars though, now that all the other tech companies have seen it done, the competition will be full of enthusiasm and will find it easier to get money. If it turns into an arms race we're screwed.

[@christopherfry2844]

Perhaps a multiple like button - press once, 1 count (addition); press extra twice, 1 count; press extra 4 times, 1 count; ...

[@christopherfry2844]

I've learned to check my text. Now I have to learn to check my placement. This reply belongs with KeiTho's comment.

[@GBCobber]

I guess someone had to state the obvious. Good job.

[@mnml2006]

Very thoughtful and interesting, your point about the bot being "mostly done." Immediate next steps would probably improve the stability of the robot's posture, training it to move about on the factory floor, and handle boxes, parts, and maybe tools. And make more of them! It's stunning and exciting, how fast the development is going.

[@andreasklossek9252]

7:16 What i yet to hear about it... Do the hands have a feedback?

[@edwardgreer491]

How long did it take Boston Dynamic’s to teach Atlas to walk on its own ?

[@normvanduker9999]

I had an interesting conversation with my son (who is in an applied mathematics Ph.D. program at UCSC) recently about training strategies for repurposing neural net stacks to perform tasks different from but somewhat related to whatever it was that the original stack was intended to accomplish. The specific example he talked about was a hypothetical 10 layer stack of neural nets that required perhaps a data set of 1,000,000 images in order to learn how to recognize the image of a cat. Now suppose it is decided that recognizing cats is less useful than doing something like reading X-rays to make diagnostic decisions. Well, it turns out that it is possible, without even understanding exactly what each of the neural nets in the stack is doing toward solving the problem of recognizing cats, to do a mathematical analysis to discover which of those neural nets in the stack would be most useful in retraining the AI to evaluate X-rays. And the answer might typically be something like 8 of the existing neural nets in the stack are identified as capable of doing 95% of the work of reading X-rays, 2 would actually inhibit the process and should be discarded, and 3 new neural nets need to be added. But then here is the really astonishing finding: the math predicts that the newly reorganized stack should be able to master its new task of reading X-rays using a data training set perhaps 2 orders of magnitude smaller than what was required to learn the task of recognizing cats. In other words, it should be possible for the newly reorganized stack to master the task of reading X-rays better than a radiologist using a training data set of perhaps only 10,000 images as compared to the 1,000,000 images that were required to train the original stack just to recognize a cat. Of course the implications of all of this are absolutely staggering as we consider the task Tesla now seeks to undertake in training a vision system to navigate the entire world. Tesla has already developed a stack adept in the specialized task of "seeing", understanding, and navigating the complexities of everything that might be encountered along a roadway. According to the mathematical analysis my son describes, this means Tesla should be able to reorganize and modify its existing vision stack specialized for driving to create a new stack that it can quickly train to the task of seeing, understanding, and navigating the entire world.

[@trottingwolf]

That is super interesting and it totally makes sense that AI should be able to learn with less data. Us humans can learn a new task with very little training. If a doctor could view 10,000 images and actually get full use from all of them they would probably be on the top end of doctors in that field.

[@ekstrapolatoraproksymujacy412]

You just described in a quite convoluted way a mix of prunning and transfer learning, those are very basic concepts in deep learning, transfer learning in particular is literally the first thing everyone do when prototyping a new system.

[@normvanduker9999]

@@ekstrapolatoraproksymujacy412 You're right and my flaw is that I am a doctor (although I got my undergraduate degree in EE and worked for 3 years doing logic design prior to going to medical school), so AI isn't something I am very fluent talking about. That said, my son was actually discussing something that goes beyond the usual concepts of pruning and transfer learning. He was describing a set of formulas they have recently developed that are based upon some extremely sophisticated mathematical models (created within in his and other collaborative research institutes) that can quickly pinpoint with near surgical precision what needs to be pruned in even the most convoluted stacks in order to quickly re-task to a new function in a manner that circumvents much of the time consuming trial and error learning that requires massive data sets in order to refine/retrain an existing stack.

[@T0mmyAngel0]

Boston Dynamics is years ahead, their robot can do many useful things like jumping, running, dancing. Who needs hands right 😉

[@Hyperlooper]

That stuff is impressive but mostly chorography.

[@eclectice]

The era of Chobits will come

[@jaseman]

I hope we won't have to wait a whole year before we get any kind of glimpse at how Optimus has progressed.

[@cherubin7th]

FSD data is much more constrained than the data this robot needs.

[@jimjam36695]

I suspect you are correct. But how do you know the competition doesn't use these same techniques? Hard to believe Google Xaiomi, and Honda are not already on par with vision a.i., simulators, and a lot of data collection. We need a video comparing the current tech from the top level competitors side-by-side.

[@rogerstarkey5390]

He told you, the data collection is based on the FSD system. It's on the vertical section of the S Curve. It's going to learn quickly.

[@anypercentdeathless]

Hey, Big Wednesday, it's "et cetera."

[@jasongooden917]

I find it amazing that people can't see the potential, but I can't say I'm surprised.

[@handsanitizer2457]

It's because you're a Elon cuck who believes in him no matter what. He's years behind Boston dynamics and it's sad after he touted this as a masterful achievement.

[@antoniomontanez4124]

Tesla bot can be use as Bomber, Fighter Drone pilots, as well Test pilots to FSD to the other bots on the line. A deterant against any first strike to the freedom of the world.

[@andreasklossek9252]

1:41 I would at least have a decent amount of money, when i get a dollar, everytime Tesla shows something, one person start to think "Others have no chance!". Yet, i need to see one, that has achieved this.

[@davidradtke160]

I think Tesla gets a lot of credit because of space x. They started landing rockets a long time ago…no one else has really managed it and they absolutely dominate the launch market now because of it.

[@NicholasShanks]

The prototype is called "Bumblesee" not "Bumble-C" because it uses vision. So the next version won't be Bumble-D.

[@nth7273]

where is it called "see" I've always seen it spelled "c"

[@vicentemontero3615]

Good vid, but please put tags (topics) on your vid

[@markbryan9115]

I can see Tesla having the first all robotic assembly line. All human functions on the production line is highly repetitive, which is what Elon says is the main purpose of the Teslabot. What better way to test (real world training) and then profit after all human production work can be done by the Bots. Of course, robot technician positions will increase.

[@timbruns2597]

I wonder if Teslabot will one day be able to fit in the drivers seat of a car and drive it, thus making the tesla cars redundant... 🤔

[@manuelight]

lmao

[@michaelweber3836]

HW 4 will be one chip for All? Like a gigantic SoC, like Nvidia Thor?

[@JonathanRootD]

This is the same reason why you were wrong about FSD being basically a solved problem

[@FredPauling]

It's bold of you to call the race now, but I definitely agree with the sentiment that what we saw on AI day hides the many advantages Tesla enjoys from the starting gate.

[@Andrew-hb2cs]

Great video! Totally agree that Tesla will dominate humanoid robot and I believe AI in general. It also looks like Tesla has the ambition to dominate AI chips as well.

[@short-shorts]

Some days I am given the impression that I have been unfair and set the bar too high. There are then the two types of people I am presented with. There is little left to change in this game of snakes and ladders. This is my fuel, my inspiration, my fuel to induce inspiration.

[@andreasklossek9252]

4:46 Stepper motors... but i know, Tesla has invented a brand new type....

[@Ifyoudonttakeitucantfakeit]

Why design sciatica into the bot?

[@jimcallahan448]

Tesla has redefined the huminoid robot as an assembly-line worker / warehouse unloader. Under this definition the robotic hand only needs the dexterity of a "gloved" human hand because gloves are worn either to protect the product (white gloves) or to protect the human hand (rubber, leather or heavy cloth gloves). Try playing a musical instrument with gloves on! The gloves limit the dexterity of the human hand so a robot replacement only needs the dexterity of a gloved hand. I think the first application of Optimus outside of Tesla factories will probably be more warehouse than factory. Perhaps Pepsi will use optimus robots to help load/unload the Tesla semi trucks. I doubt optimus will be driving any forklifts for years to come. But optimus may use pallet jacks in addition to lifting boxes one by one.

[@fdelacou]

While I generally agree with the analysis provided, there is also a very clear unknown on what was hardcoded vs self determination from the bot. Seeing the bot walking shows that they have the mechanics and the software to support walking down. BUT... you have no way of knowing if the bot is analyzing the environment and reacting to it. Everything that the bot is doing on stage could be done by a blind person. This is not a matter of vision so we can't assess that aspect. Same with the decision making. There is very little information on how much was precoded vs self determination from the bot...

[@nickmcconnell1291]

As to using Teslabot in Tesla’s factories. Until they can use Teslabot everywhere on the production line, then the lines move as fast as the slowest human that is left on the line. So until they are able to replace all humans involved in building the cars, then Teslabot will only increase the bottom line as it is cheaper than hiring a human.

[@pooglechen3251]

Since we don't have enough training data for the bot and can't simulate the complexities of all the tasks, we need the bot to be able to "monkey see, monkey do" so that it can adapt to undefined scenarios

[@Ant3_14]

Repeating what human did would be cool. One bot learn new skill, all bots have learned. Actually that would be enough to do most/all worker jobs in factory

[@fredturk6447]

I agree about the mechanical side. Very well done, game changing. However, you need to get the AI right and that is critical. As yet I don’t see FSD is solved. There are issues in mid term planning and there is as far as I can tell no long term memory included. So every time you drive your FSD car somewhere the car has a new experience it does not learn on the job. It’s does not recognise and remember specific places and what to do so it’s going to take a generic not optimal approach to driving. I suspect task planning and mid to long term memory maybe important in a robot successfully operating in the real world. Tesla has achieved a lot engineering the daylights out of its FSD and that’s great. However, is that going to be enough? Is there something we don’t yet understand about building AI software systems that is critical? Can we achieve enough functionality with current hardware? I don’t think we know the answer to these questions yet. So what has been done is fantastic, but it is still very early days and the problems maybe more difficult than we think!

[@sebpatu]

So incredible to see the cult in action. So efficient even when lamely executed.