These New Robots Do Previously Impossible Tasks!
Vložit
- čas přidán 11. 05. 2024
- ❤️ Check out Microsoft Azure AI and try it out for free:
azure.microsoft.com/en-us/sol...
📝 The paper "DrEureka: Language Model Guided Sim-To-Real Transfer" is available here:
eureka-research.github.io/dr-...
📝 My paper on simulations that look almost like reality is available for free here:
rdcu.be/cWPfD
Or this is the orig. Nature Physics link with clickable citations:
www.nature.com/articles/s4156...
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Károly Zsolnai-Fehér's research works: cg.tuwien.ac.at/~zsolnai/
Twitter: / twominutepapers
#nvidia - Věda a technologie
Kudos to NVidia for making the source code for this one open source. Most of the times we see these amazing papers, only to see hidden away in expensive functionality. It is nice to see research that will help others make new and interesting things!
They make money by selling hardware, not services. Similarly, Meta makes money by running LLM's so it can release Llama for free and let everyone else write software for it for them. These are the incentives, but I am glad it's open source!
@@nathanbanks2354 I do agree with that I observation! CUDA is mostly without an equal, and I will bet that these algorithms work best on a GPU. I would love to see Intel or AMD come out with a compatible or competitive framework to Nvidia. I think we all would benefit.
@@nathanbanks2354 😑 Still they kept Linux driver source code proprietary.... and refused to work with them, when they sell millions of chips to android and server markets that relay on this code
Agree , this will help alot when it comes to legged robots
Why did they make it open source?
If Telsa wishes to be worthy of Two minute Papers, they should produce a peer reviewed work!
I would respectfully suggest that the standard of your excellent chanel isn't lowered just to cover a company that already has an outsized coverage in the media.
Got it. Thank you for the feedback!
This. 👆
@@mLichy911 Peer review is noticeably absent at his brain chip company Neuralink too.
THIS
@JamesRosier
Here before... you know what? Enough has been said...
I'm more impressed with the robot's quick motors than the AI itself
Ye I was thinking of that as well
Yeah I think it's more a question of the robot's reaction time. If you watch industrial robots you know the motors can be very fast, but most AI robots are made to move slowly for safety. Even the tesla robot sorting batteries. I was like "Come on... you can do it much quicker than that!"
@@peetiegonzalez1845
-So you agree it's both? 😶-
@@ivoryas1696 yeah I guess I do. I’m not surprised by the speed though. I guess not everyone has watched industrial robots.
@@peetiegonzalez1845
Yeah... actually nvm. That makes sense...
Simple. "Robodoggy, Floor is lava"
Awesome video. Very hyped to read the research paper!
The tesla showcase: To be honest. I don't trust the tesla showcase. There is no proof of how it works. I want a paper and reproduceable results cause this looks cool =(
There will be no paper, it’s a proprietary product.
@@zachb1706 I know that is why I am sad.
Noted, thank you!
I'm holding so tight to my papers!!!
My robot dog ate my papers💀
Tesla Optimus is 10 years behind what other private modern robotics companies can do.
I guess the point of Tesla Optimus is to be cheaper... But knowing how "cheap" Cybertruck ended up being I wouldn't put my hopes up.
Examples besides Boston dynamics?
@@The_Nigga_Chin Unitree, NVIDIA, there are thousands of consumer robotics companies
@@MRboomchongo Nvidia doesn't make robotics, they provide a software platform (GROOT) which all other robotics companies uses to train their androids on. Tesla is about the same level as figure robot despite not using the GROOT platform.
@@The_Nigga_Chin You can't make a robot without chips can you?
It sure looks like that handler is helping quite a bit with stability.
Only if its lifting it directly up and centered. If you notice most of the footage they're holding onto it very loosely. That being said I wouldn't be surprised if these footage were all cherry picked among many failures.
If he didn't pull, it wouldn't move forward
That's why you can see slack in the leash and he's standing behind it. Its not faked.
@@deadlynightshade7867 He's pulling it from behind?
@functionatthejunction think, the dog is just trying to balance on top of the ball, and so if the man didn't pull, the ball and the dog would just stay in one spot once the dog has fully balanced on top of the ball
It's amazing to see the simulation application in real world examples now. You've been showing us plenty of papers, mostly by nvidia', that show how simulation can be used to program robots and it's great to have new terms like domain randomisation. Thank you always.
Very interesting to see the dog exchanging "style" for "robustness" by lowering it's CG as much as it could.
When we hear about balancing on a ball we don't typically think "crawling around it", I hope some day it will be able to walk on it elegantly like accrobats do !
And a speculative video on Tesla's humanoÏd robot would be a great exception that I'd love to watch !
It's nice to know about exceptional new capabilities even if when they're not in a peer reviewed paper. It sometimes makes me wonder how something works and how these improvements compare to real papers which describe the process. I'm glad proprietary AI stuff seems to be a step or two ahead instead of 10-20 steps ahead.
They made lucid dreams to make the robots work better
like the movie Inception
Thank you for speaking about the particular techniques used for training. It's really helpful to understand how these achievements were accomplished and how they relate to previous approaches.
As interested as i am in the development of androids such as the tesla robot, we should not support them releasing non-conclusive results.
However, as soon as they release something official to look at, i'm gonna be here waiting for you to release a video about it :)
good video!
regarding speculative video possibility, I prefer focusing only on peer reviewed papers!
Noted, thank you!
Actually, the robot dog was walking the man!
Clever robot - and yet the handler is using a leash. Let the thing go, I'll be more impressed.
The leash is pulling the robot off balance forcing it to keep on top of the ball. This is how the direction the robot travels in is chosen. Otherwise it would either stay in place or move randomly as it overbalances itself. The robot reacts automatically to the lead pulling to keep the ball under it.
It’s quite something to realise that robots have more than The Laws of Physics to work with. What a time to be alive! 😅
Seeing the dog balancing on a ball made me think that these could be fun to train on skateboards or skates or such.
Its always better then the thumbnail and title implies
It works because the dog knows where it is at all times. It knows this because it knows where it isn't, by subtracting where it is, from where it isn't, or where it isn't, from where it is, whichever is greater, it obtains a difference, or deviation. The guidance sub-system uses deviations to generate corrective commands to drive the dog from a position where it is, to a position where it isn't, and arriving at a position where it wasn't, it now is. Consequently, the position where it is, is now the position that it wasn't, and it follows that the position where it was, is now the position that it isn't. In the event of the position that it is in is not the position that it wasn't, the system has required a variation. The variation being the difference between where the dog is, and where it wasn't. If variation is considered to be a significant factor, it too, may be corrected by the GEA. However, the dog must also know where it was. The dog's guidance computance scenario works as follows: Because a variation has modified some of the information the dog has obtained, it is not sure just where it is, however it is sure where it isn't, within reason, and it knows where it was. It now subtracts where it should be, from where it wasn't, or vice versa. By differentiating this from the algebraic sum of where it shouldn't be, and where it was. It is able to obtain a deviation, and a variation, which is called "the ball"
I didn't see the dog balance on the ball on it's own a single time
Just ignore the guy with the leash and harness. 😁It is still pretty impressive though IMHO.
Soon enough we will
It did , the leash is there to save it if it falls down. Yes it does look a bit tight as if the human is helping it, but think about it this way , if it has no engines , how long could the guy keep it balanced on a ball with just a leash? Idont think the leash does much tbh, in fact , to the degree that its being pulled, it actually adds another factor for the robot to counter.
Then you didnt watch the video.
I think this channel is in a growth trajectory and more and more people who have little information will be watching. I remember in a chatGPT video someone said they didn’t know of it in the comments. So I think it’s important for this channel to cover these important areas of AI even if they don’t have papers
I'd actually like to see AI deployed to hard code these balancing capabilities into the robots, so that the code can actually be reviewed by humans. That would certainly be the pinacle of AI-Technology.
Yeah it would be great. AI is booming at the moment and a lot of it is proprietary. I’d like a deeper dive into what techniques these companies might be using and the process they’re making from someone with good knowledge in the space.
What a level of simulation! The robot dog is humping the ball!
More stable than before
This whole robotic revolution seems like a common effort. The year of 2023 and 2024 'give birth' to a lot of new Robotics companies and their new robots products. Moreover now that NVIDIA give more attention to it, and also with the accelerated growth of open-source communities. And yes, it all happens from the grassroots to the giant companies such as the TeslaBot Optimus.
What a time to be alive ... in a robot.
let's enjoy being alive until we still can
I have never felt more anxiety for a robot
I think it's best to stick to papers rather than get caught up in speculation. We should think of Tesla's video as nothing more than marketing materials.
(sees robot dog on ball crossing road) This is not how I expected Terminator to start
Not too sound like a hater but I feel like that tether was doing a lot of heavy lifting with the dog...
No if he wasn't there it would just balance in place and not move
@@LandHooman couldn't you (as a human intervening) apply force to the ball? or coax the robot into applying torque?
@@AZTECMAN Yes you could, but that's not what we see in the videos. In the videos it looks a lot more like the guy is intentionally pulling the robotdog off balance to make it move.
@@timwatz2330 Not sure if you are agreeing or disagreeing with the original reply from Uriah.
From my perspective, it would be alot more impressive without the tether, but it also would be more risk to damaging the robot - which is probably the reason they keep it tethered (in my opinion).
That's why it was slack, right?
People in VR goggles and gloves mixing with training robots at 4:44 were pretty creepy.
This Robodog here should jump on that Ball, roll through an Obstacle Course and then jump back down on the Ground without a Human holding that short Line.
That would be impressive.
You hve no idea how impressive what you are actually looking at is.
@@functionatthejunction oh, trust me i do... but it just looks boring.
And the fact he holds that short Rope so tight kinda looks suspicious...
what a time to be AI!
Yes, speculate on the Tesla Optimus bot progress!
One of those Tesla clips shows a number of people in VR gear apparently controlling the robots. Probably part of the training, but Musk does have a reputation for making claims that later turned out to be overstated
It clearly states in the video when it’s working autonomously and when it’s being controlled for training.
@@zachb1706 sure. And I stand by my comment
@@zachb1706 Autonomously doesn't mean playing back prerecorded movements
Great video, always a good day when 2mp drops!! I think you should have fun with the difference between peer reviewed and mostly publicity stunt. have a funny AI generated bumper and alert when talking about something that isn't cut and dry.
Next step - Robo Circus
Step after that - Both animals and robots enviroment (circus or park lol)
Dude is holding the dog in place by the leash. The leash is taut like 80% of the time.
0:38 That's probably super distracting for the drivers
It feels like the leash needs to be a little longer to tell how much it's contributing.
Its not fake. Dont even start.
@@functionatthejunction That's interesting, because I was careful not to say it was fake. I said you can't tell from the video.
Really cool!
WW3 will have robot dogs balancing on yoga balls with weapon systems attached, fuck
If I understand correctly the OpenAI robot is far, far past the Tesla robot.
Wow, the guy with the leash does an amazing job balancing the robot on the ball.
This dog on a leash looks very funny
Video from 2099: Look, this human can balance on a ball, they are so funny, too bad they are rare species.
"4D Gaussian Splatting" when?
Personally i think you should Stick to the Papers. There are already way too many news channels showing off unreviewed work
Papers? We don't need no stinking papers.
I think the papers are a significant part of the wonderful nature of this channel. It reviews the concrete results of science and research, and not speculation or advertisement.
He’s holding it by a leash. You can see him balancing it
The leash is to prevent falls if anything, that is why You see that even with the leash the robot fails to grip properly to the ball sometimes , meaning that the leash does some of the work but not all , if the leash did all of the work then the robot shouldn't fail as long as it has the leash
looks like LLMs are the new operating sytems of ai applications
I think the robot would fall without the leash?
I'd pay real money to see this number in a circus.
Holding the leash so tight it’s very hard to see how much the human is helping the robot dog not fall off the ball..
THe eash is literally slack and behind it. Its not helping.
As for non-paper stuff, I think it's covered plenty in other places. Please stick with papers, fellow scholar.
I'd be more impressed if the leash wasn't so tense.
Does the leash need be under such tension?
1:44 did it invent gopak while trying to learn to walk?
I read this paper, and a follow-up paper.
Yes please to your suggested video !
Heyyy. I saw a similar paper at ICRA toda- hold on... is it the same one?
The probability of us being in a simulation just keeps growing.
How so?
An AI being simulated would mean theres a risk it cannot stop or account for, so a rouge AI would likely make attempts to stop any chance of the universe being simulated
I would welcome some speculative analysis by you, even if a paper is not available. Unfortunately, as it's clear a vast amount of money is involed, we will see less and less publicly available AI research and more proprietary technology. I like your channel not just for the excellent analysis of a paper, but it keeps me appraised of what is going on in AI development in a grounded way.
Please what are the inputs in the simulation and in real world. Still confused!
The robot on a ball looks so freaky, I cannot help but keep looking for generated video signs. Insane
Did you go to college for forensics?
Looks like the leash is helping it stay up?
what the dawg doin 💀💀
How long in simulation is equivalent to real life training?
there is too much tension in the rope, as he appears to be stabilizing the robot.
Did you reupload i remember seeing this video already
like to see optomus cordinate with two hands.
cant wait to see this on a moon or mars robber
all this 'kicking the robot' footage needs to be gone when it becomes sentient 😂
Terminator lied to me. I will be scared of balls now.
Shaky chihuahua robotics
"Fellow scholars," are we not dear to you anymore good Sir? ;-;
You know what that means boyz
One step closer to robogirls
Balancing on balls no doubt.
Futurama warned us about Marylin Monrobot.
Is this in the US or in Asia?
Speculative video about Tesla Optimus would be nice with a disclaimer. Also if they lied about something big then it would be really nice if you corrected it in your future video like you did with Devin. Also great video. :)
Noted. Thank you! 🙏
The robot constantly would fall to the front of it if the human would not hold it with a robe.
cool !
The dog would be falling off if there wasn't a human there keeping its lead taught. Its cool but its not like its really balancing by itself lol. You can even see like at 1:04 how much it is fumbling and how much the human is pulling it back to the right spot 🙄
"Working remarkably well"
During all of the videos, rope holding dog up is straight. It's not falling because they are always holding it.
Its slack and behind it...
THE FLOOR IS LAVA
"with *responsible* AI built in"
hrhrhr sure thats why they immediatly connected it to the internet ^^
WATTBA!!!
Optimus video would be cool
why are all videos in 5fps?
Robot wants to learn to use a Gun! What move should the robot forget to learn to use a Gun!?
you didn t talk about the use of gpt4 in the training of the robot.
Why are they holding the leash so tight?
They just need to improve it enough for it not to look like the dog is humping the ball
now things are heating up
scary stuff
holy mother of papers XD
Good, good, now make it balanced itself on a ball on top of another ball 😈
Are you following GPT-4o?
Tesla's Optimus has really failed to show anything innovative that was not already possible. Them taking out batteries and putting it in a box could be done far quicker and cheaper, and probably more reliably by a simple robot arm. It walking around is absolutely nothing compared to what Boston Dynamics has already been able to demonstrate with Atlas for a very long time. Add to that that Tesla has done plenty of faking stuff before to push sales.