Crazy 50,000,000 Point Bouncy Jelly Simulation!
ฝัง
- เผยแพร่เมื่อ 21 พ.ค. 2024
- ❤️ Check out Weights & Biases and sign up for a free demo here: wandb.me/papers
📝 The paper "Vertex Block Descent" is available here:
ankachan.github.io/Projects/V...
Notes from the authors:
“The range you mentioned (100x to 1000x) is appropriate. However, please keep in mind that, as we noted in the paper, this figure is specific to the simulation settings we selected. For relatively tame experiments with less stretching and motion, and especially for highly stiff and high-resolution simulations that are much more expensive to simulate, we would expect Newton’s method to eventually overtake all alternatives beyond a certain level of convergence.”
Fluid control:
users.cg.tuwien.ac.at/zsolnai...
users.cg.tuwien.ac.at/zsolnai...
📝 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 - วิทยาศาสตร์และเทคโนโลยี
I've seen many questions about the platform it runs on. As the first author of the paper, let me clarify:
All the demos run on a consumer PC with an RTX 4090 and a Ryzen 5950X.
Additionally, the method supports both rigid body dynamics and fluid dynamics.
You are an author? Congrats, this was awesome!
Your work is amazing, and still just a PhD candidate? Can't wait to see your future papers!
This is truly amazing work.
Massive respect 💫
👏👏👏👏👏👏👏
What an elastic time to be bounced alive! 🎉
These simulations are getting out of hand!
🤨🤨
What a time to be jiggly
@@Mertiven 😂
hold on to your armadillo
SIR IM HOLDING ONTO MY PAPERS VERY HARD
SIR I AM SQUEEZING THEM
This jiggle physics will have good applications, of course, for educational purposes 🍑
thicc squishy jelly a.. 😍
@@Roberto-nb5cbbigger please 😂
In Stellar Blade 2 hopefully :)
I was searching the comments for project ideas. Looks like we found a winner.
Obviously strictly for research purposes 🎂
*Men of culture, we meet again.*
For those not familiar with the topic, things took DECADES to get to this point where elastic simulations of this kind now take seconds per frame, even when involving millions of collision calculations. The typical physics solver would rather explode than to even finish the simulation, and potentially take HOURS to calculate it. That's why this is impressive, so it's important to first put things in perspective. 🙂
pls forgive my ignorance, but when this will reach ... humm ... 60/120 fps? and when it will be implemented in offline singleplayer games?
Wow impossible to guess it might happen within 5 years😊
@@el-_-grando-_-_-scabandri Maybe 2-3 papers down the line? But looking at the existing physics solvers, a couple of seconds per frame with millions of collisions is totally unheard of. Just by itself, this is extraordinary. Like I said above, we have to put this in perspective. ^^
@@el-_-grando-_-_-scabandri It can already reach that, just not with the millions they are showing in the video, if you were to scale it down to tens or hundreds of thousands, then it would probably take less time
@@el-_-grando-_-_-scabandrifor sims with smaller scales (I mean with 100k vertices, it's already in real time)
Blender desperately need to integrate this
just what was on my mind all the video
who going to give them the money?
Blender needs 10 million in American fiat currency or 1400 Megawatt hours of pure energy to complete this task in a reasonable time. Feed the machine!
I was thinking BeamNG, IDK but I doubt it would be impossible to add plasticity into this
BeamNG's physics has a very low mesh resolution, and unstable enough that it really sounds like a pain to work with
I'm undereducated on how software works.. Do they need an AI to do this? And if so, wouldn't there be a way to p2p train it through the open source network? And if so, then why tf do we need these manufacturers? (I know physical systems somewhat well so if you can compare to physical stuff that would be amazing!)
I love it when it's not reliant on AI, feels like we actually discovered a new technique instead of using a very very cool hammer to solve all our problems
Lol, love the analogy.
It's *almost* how machine learning works. 😄
You still haven't learned the Bitter Lesson, I see.
Yes, that is exactly how i felt.
Your enthusiasm is incredibly contagious!
Would love to see the new paper by Anthropic, it's really interesting: "Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet".
Károly: "Let's flatten this poor little armadillo"
The little armadillo: Yes?
Two Minute Papers: "They learned how to pack 1 million people into a tiny teapot"
Blackrock: "Write that down! Write that down!"
"University of Utah & Roblox, USA" - why roblox? haha in 7:18
Probably also works at Roblox
Learned more about programming from Roblox than the Uni.
It turns out (suddenly) the Roblox engine does not write itself
roblox is interested in world domination
I know you're the two minute papers guy, but would you ever consider doing an overview video on the current best/workhorse simulation methods? There's so many, running on such similar looking benchmark tasks, that I feel lost every time a new one comes out. I just want to know what's out there, haha.
Amazing the new speed, I think it's important whenever you talk about speed (seconds per frame, for example), to post the specifications of the hardware used to achieve those results mentioned in the paper, so I've checked and it is: AMD Ryzen 5950X CPU, 64GB DDR3 RAM, and an NVIDIA RTX 4090 GPU.
the hydraulic press channel would love to work with these squishies
Keep the videos coming! Thank you!
genius papper wowww loved it just imagine what could be possible in two more papper
No AI here? Wow! I almost lost my papers not holding onto them enough!
thank you for the video. Always excited for ingeneous hand-crafted techniques!
It's nice to see TMP get back to roots. AI is amazing but this is the good stuff we've been missing.
Amazing quality and speed! Stunning!
Now that's what I'm talking about!! Restir and cutting edge physics simulation videos one right after the other, and it's all hand crafted with no AI! This is what I am here for!
This is awesome I've been pretty down and sick lately but this made it a bit better 🙂
I love it when you cover papers like this
Wow! Hats off to the researchers.
Dr. Cem Yuksel continues to be part on amazing research projects lol. Props to the team for doing this job!
Finally some old school 2 minute papers content!
The occasional graphics/simulations video, eh? :^)
Glad to have you back, if only for 8 minutes every once in half a year.
Soon we will be able to design a whole universe, where the creatures who live in it will think that they are real and will not have a clue about their origins.
Its already happened.. we are the creatures 😂
Has a Houdini vellum enthusiast myself, I love this video
You theorized a working set of physical laws as a thesis? Impressive!
nice seing some physics simulation again, always a highlight for me
Balls teehee
Wonderful, I love really squishy balls!
Hey Karoly, how fun it must be to go to work each day just to do simulations of this caliber!🎉😅😊
Hah I was watching distractedly, I though to myself "well OK there are some things where AI people are really useful" and then you said "no AI is used here". What a time to be alive!
when he said "HOLY MOTHER OF PAPERS!"
I felt that
Thanks!
This is absolutely amazing. A nice break from the AI stuff to talk about an amazing paper like this.
What a time to bounced for alive !
Cool. I knew one of the authors, Cem, from grad school. Small world. Thanks for reporting, Károly. Connections like these emphasize the "human" in human ingenuity.
It is truly refreshing to hear "No AI was used here"
Imagine showing this to someone from the 90s or early 2000s, and saying, this is what's coming up, but you don't won't get it for 30 years.
Ok, this looks like it can actually be used in games now. Not in a specialized, either optional or highly focused way but just as a general gameplay feature. So cool!
(Ok, I didnt watch the whole video; seconds per frame obviously requires 2 more papers for usability in videogames :D)
Wow this is amazing
This looks promising 🤔 gotta apply that to some specific body parts for more stability. I love science ❤
Dr. Papers, this was *thrilling!* (That's as best I can translate my last seven minutes of swearing and inchoate gibbering).
Wait if all nodes are independent of each other could this run on a gpu ?
HOLY MOTHER OF PAPERS
i love this channel
As a regular user of FEA and CFD, this is astonishing! Can't wait to see the jump in productivity in the coming years. This is what AI should be used for.
It's great to hear again about simulations and not just AI.
Woo simulation content
This is insane.
The smell of those spiky ball toys would have been unimaginably bad irl.
How to use this in blender or unreal engine or whatever ?? Like what the skills I need to learn to be able to transfer research papers to real applications ??
Yay, back to squishy satisfying physics papers 😌 I've been avoiding GenAI papers, but thats meant not watching Dr Károly
Could this also simulate more rigid structures correctly? Then it could work for large scale simulations e.g. of earthquake scenarios. Shake up the whole city.
Speed and 'how fast' it is seems to be mentioned all the time... on what computer? A super computer and/or a gaming computer?
If it is a supercomputer, these advancements are going to help us develop the same level of processing power on more accessible devices.
When we look back 20 years to what a household computer could do and compare it to the standard computer today, the advancements have been huge. The speed of advancement is only growing so I'd guess that within 5 years, devices with the computing power to complete the simulations seen in the video today will be well within the public's reach, likely on devices such as or as small as our cellphones.
This is however, just my assumptions.
I grabbed my papers so hard with this one that they managed bounced back somehow
Is it calculating with air pressure? All the elastic bodies in the glass jar would cause suction, changing the dynamics.
(No, going by the unchanged falling forms before hitting other objects)
Is this the mother of all oversights?
Game developers jumping on this paper in 3, 2, 1...
AWESOME ❤
That is the first step to functional virtual muscles, right?
I'm sure the Corn industry is going to put a lot of money into these simulations.
Damn this first one looks incredible. Soon they can simulate everything.
I wonder if you wouldn't mind giving us some real practical applications where these are currently used or if they aren't used yet then where EXACTLY they could most likely be to be used? For each video.
two weeks in CORN-flix
1:49 This could be very useful 😂
Now to do the the same but with tearing on top of elasticity (in the same simulation I mean). Next paper perhaps?
I'm an fx artist , and I know how painful it is to simulate them. BUT this is just a miracle.
As a 3D hobbyist, at last we arrived there! I spent days simulating fluids, softbody and hardbody simulations that took several hours to simulate a few seconds. Will it come to our favorite 3D packages soon?
How does this behave against explicit dynamic FE-simulations ? Is it „just“ creating nice pictures or is this actually generating realistic numbers ?
Do they sell tose algorythms to 3d software companies? Are they open source?
Any idea?
This is amazing - no AI and we're at just a few seconds?
That probably means we can do this in real time with AI.
Hardware kind of matters when talking about frames per second. Did I miss it? 3.6s/frame on consumer PC? Supercomputer?
A high end consumer PC with RTX 4090 and Ryzen 5950
seconds per frame when gaming: 👎
seconds per frame when simulating: 👍
This combined with VR is going to be crazy immersive
Damn it was crazy when he showed the semi logarithmic scale
OOOOhhh SO nice 👌to see some non-ai papers here again!! Thank you :)
Right!! Hope he does more on occasion. Especially in the graphics space or other practical applications
any estimates for when this can be implemented into games without too much difficulty?
Trying to fix the chair while you sitting on it :D
don't get me wrong, generative AI is cool and all but i definitely missed me some Classic TMP
It's pretty amazing they can do this but TH-cam video compression algorithm gets pixelated when showing so many different things moving around on the screen.
Yes this is probably very usefull, but I need to know where I can get 2 hours worth of these mesmerising simulations in 4k and some popcorn!
Loving it but you forgot to explain the magic sauce how they made it
Airpot 1 mill bodies bump into each other💀
4:22 well… this gives me a n idea for a video I can’t post on TH-cam…
0:29 Imagine, an airport with one million people bumping into each other!🤣🤣
4:56 Now, imagine, that all of these people are packed into a tiny teapot!
Cooooooool
I wonder how long it will take to get these technologies implemented into blender
Kinda glossed over it but the tear sim is super impressive
nice! where is the 88-line code version? ;D
i remember we ised to play with balls like this :D taking one string and spinnig the ball hahaha
FINALLY, NOT AN AI PAPER
Why the little cube was tossed away?
In real world I understand but in a simulation, with perfect positioning, shouldn't be the lack of a lateral vector? Or it was some little wave in the bigger cube that pushed a little and it took it the perfect positioning?
I would guess some parm in the sim included noise, or there is a little noise inherent to the technique.
1:50 Now i see how this will be useful...
R34 3D artists are drooling right now.
Good
nice
Woah!