A Repulsion Simulation! But Why? 🐰
Вставка
- Опубліковано 19 чер 2024
- ❤️ Train a neural network and track your experiments with Weights & Biases here: wandb.me/paperintro
📝 The "Repulsive Surfaces" and "Repulsive Curves" papers are available here:
www.cs.cmu.edu/~kmcrane/Proje...
www.cs.cmu.edu/~kmcrane/Projec...
❤️ Watch these videos in early access on our Patreon page or join us here on UA-cam:
- / twominutepapers
- / @twominutepapers
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bryan Learn, Christian Ahlin, Eric Martel, Gordon Child, Ivo Galic, Jace O'Brien, Javier Bustamante, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Michael Albrecht, Michael Tedder, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, 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
Meet and discuss your ideas with other Fellow Scholars on the Two Minute Papers Discord: discordapp.com/invite/hbcTJu2
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.at/~zsolnai/ - Наука та технологія
I love everything coming out of Keenan Crane's group and this was no exception
Kram! One of our early Fellow Scholars - so happy to see you! I love their work too. So many amazing geometry papers are coming from there!
@@TwoMinutePapers watching every single one of your videos :)
@@Kram1032 I was thinking about you the other day, wondering what you are up to. Such an honor, thank you! 🙏
@@TwoMinutePapers The honor is all mine, thanks for introducing the world to so many phenomenal works
For sure!
I'm sure topologists will be going nuts over this stuff.
Litterally what came to mind as soon as I saw it lol. This is a topologists wet dream.
I'm not a topologist, but I'm completely shocked by this. I mean, every mathematician must be.
There would be a huge progress on the knot theory.
I'd love to see this as a tool in Blender in s few years, I could loose hours tinkering with this
"And, if we feel like it, we can also start with a small piece of noodle inside a bunny and start growing it. ... over time it starts to look like intestines."
Yep, that's a pretty repulsive paper after all, lol.
Bingo. 🙈
That's legitimately what I thought the video was about, but I was pleasantly surprised by math I guess.
@@nyanSynxPHOENIX same here. An AI that had an understanding of what humans fond repulsive.
@@ts4gv same lol
This is simply amazing, specially if it doesn't break the bounds of the mesh, this could be the next cloths in games, forget about clipping through objects, this could save it by repulsing one of them or both, imagine this with grass and a foot messing around with it
Finally the shrubs I wander though could deform properly instead of clipping! Outstanding!
Check out this channels recent videos. There are some methods that are well suited to those kinds of problems.
My thoughts exactly. The navigation example also sound great for crowd avoidance. Only question I have, is it fast enough to be used in actual games or is only just fast enough for a showcase scene.
Days Gone has the most realistic facial animation I have ever seen in a video game
Real-time 3D rendering with equally real-time anti-clipping?
I'd love to see this applied to brain growth and wrinkles. Some people think brain wrinkles are adaptive, while others think that the typical brain folds and wrinkles are primarily due to the forces squishing the brain as it grows and the skull solidifies. Would be interesting to model brain growth in this and see if it produces characteristic physical human brain patterns.
+1
not a brain researcher or anything, but the presence of wrinkles suggests maximizing surface area, or conversely minimizing mass to volume. koalas are considered one of the least intelligent animals (compared to mass) on the planet, ant their brains have no wrinkles. there is a correlation, and I bet a causation between these factors.
@@devilofether6185 : not disagreeing, but I've personally never read a good explanation as to why more surface area is helpful.
@@NirvanaFan5000 more surface area means you can cram more cells in less space
think of it like shoving deflated balloons into a box, vs having to inflate the baloons. more surface fits in total due to the creases using space more effectively
our lungs and intestines actually pull a similar trick!
the lungs have alveolae that increase the surface area of the lung so we can absorb more oxygen
the intestines have folds all along the inside of the tube for more cells that allow the absorption of nutrients and water along the surface area of the intestine itself
well, for the brain it simply allows more individual braincells to be crammed in the limited space of the braincase, thus making us have more brainpower and a higher potential for intelligence than koalas and their smooth brains
i hope this helped you understand why folds are useful!
@@triangularlizard I don't understand why you think wrinkles would allow for more cells. if you can fit a 3lb dough into a bucket, you can't fit more dough in by giving it wrinkles.
"At first you had my curiosity,"
*See's Pathfinding* "Now you have my attention."
a couple uses off the top of my head based on my own experience.
separating curves in a hair/fur groom for improved volume conservation. we can already do this at sim time, but that adds a lot of computational overhead. if this is cheaper it can simply apply the detangle/deintersect post-sim frame and keep things iterable.
deintersecting non-trivial crowd agents (quadrupeds especially) as a sim postprocess would also be a benefit. again this usually has to be done at sim time and is incredibly prone to errors and rapidly inflating frame times.
These are two really practical applications of this tech which are easy to imagine being used frequently in a real world scenario. Thanks for this comment
Could be used, in general, for any dynamics sim as a startup state, yes
The most practical application I heard so far is for untangling tangled earbuds
I love that this insightful comment comes from someone with a he man profile picture
0:40 I find it fascinating it made a Hilbert curve without being programmed to
Applications in knot theory? Topology? What a time to be alive!
This has to be one of the best channels on this site. Keep it up
I totally agree. No other channel blows my mind with every new video the same way this one does. What a time to be alive!
eggwuh
I have not missed a video from you for a year now. Thank you for this thought-provoking and inspirational content. To the right person watching this, there are so many ideas you can derive from these papers.
As an early career researcher in Molecular Biology, I hope in due time we will see content like this for every scientific field. Someone has to share those papers with a wider audience, not just hold on to them! Thank you for putting so much effort in that Károly!
Anton Petrov does this for science papers, mostly astrophysics on his channel ua-cam.com/channels/ciQ8wFcVoIIMi-lfu8-cjQ.html
Applying this to protein folding could be interesting.
I believe that could revolutionize the field, actually.
I remember the game Foldit that came out a while ago that attempted to crowdsource certain folding problems
@@btat16 Indeed. Although it was hard to qualify it as "a game', but it was a rather curious piece of code in its own right.
It would need heavy tweaking to account for all the molecular forces involved but yeah, it would be interesting indeed.
Bingo!
Thank you for your service :)
Thank you so much for your generous support Danny! 🙏
If the path planning works well, could it be used to guide air traffic so as to both optimise and make safe the paths of aircraft in controlled airspaces?
I was expecting an animation of how to turn a sphere inside out with this technique.
Depending on how fast it is, it'd be very nice for crowd simulations. I'm also curious if there's a trivial way to constrain the points/curves to a mesh
that it is quite fast is a big point of the paper
Would this be a good paper to apply to protein folding? I'm no biochemist, but as I understand it all structures on the macro scale are defined by the atomic linking of matter and the shape it takes (folding). There is a whole decentralized process to compute folding experiments called Folding@Home that might benefit from this technique.
Well, I’m not a biochemist either, but I believe it would help.
However, the nuances are numerous enough as to make it just that: help. protein folding is governed by more than just repulsive forces.
I can't say conclusively, but as a biochemist who is using AlphaFold2 in his current work and used Rosetta in a previous project, I'd say tentatively probably not... Rosetta is the closest piece of software to what you're suggesting; it applies individual forces to molecules (as opposed to AlphaFold2 which looks at all previously characterised proteins and uses deep learning to approximate what new sequences would look like). Rosetta is already reasonably intensive for large proteins and that's after already taking several shortcuts by subgrouping amino acids into trimers and heptamers and folding those first before having the subgroups interact with one another. If this method could be applied, I can only imagine it'd be more intense than Rosetta. And Rosetta is not always close to perfect (depends on the protein). But then again I'm not actually a bioengineer; I just use the software and don't actually do any coding. So I can't confirm for sure!
Exactly what I thought! That would be amazing!
@@markarts404 bioengineers use the software then?
@@davidmattes199 Bioengineers make the software, biochemists use it - imagine Photoshop vs Photoshop users (except with Rosetta it's best you at least have an understanding of the fundamental way the software works so you can interpret the results)
When I saw the hilbert curve happen in the intro I literally jumped out of my seat with excitement.
This is oddly the most satisfying video to date, could watch these simulations all day
Thank you for making these videos. Im always so excited for the next video!
You are too kind, thank you so much! 🙏
@@TwoMinutePapers no problem! I've been watching for at least a year now... It's so interesting seeing the progress!
I wonder if this can be applied in knot theory topology 🤔
I usually have a sense of how the implementation of ideas would actually work, at least in general. But this, I don't understand how this can be done. This is exactly the thing graphics designers have needed since literally the beginning of computer graphics being a thing, and the applications are unending, but just...how?!
They summoned the computer spirits and learned their black sorcery?
In pretty much every video on this channel, you always seem to find the paper way more excited than I am, even with exciting videos, but in this case, I spent the entire video wondering why you weren't reacting to just how truly amazing and useful this one is! My mind is now buzzing with all sorts of interesting applications for this! I finally understand why you're so excited in every video: it's to strike a happy medium between the attitudes of your viewers. Thanks for the amazing content!
This could definitely be used to improve clothing in games with character editors allowing diverse body types.
0:50 I immediately recall the "How to turn a sphere inside out" video. "Remember, you mustn't tear or crease it" 😂
Wow, this would certainly help with shrink-wrap applications, currently most shrink-wrap operations/modifiers produce lots of artifacts and overlapping faces. This is very exciting stuff 🙂
This is your best video to date. I was NOT able to hold onto my papers for this one.
For those interested in this work, I can highly recommend the online youtube courses from Keenan Crane on discrete differential geometry (he is a co-author of this work). They are one of the best lectures I've seen thus far.
I knew his name was familiar!
This tool seems like a topologist's best friend.
You should show more of the path finding using this technique, very interesting!
Finally I am educated on the anatomy of Stanford Bunny!!
I'd use it to upgrade graphviz. It's a tool that's used to visualise graphs, like that social media graph example. This paper would produce much more readable graphs. So many things in computing can be represented as graphs, so being able to visualise them in a readable way is really useful for debugging.
Wow that manifold optimization was really good, computational techniques are getting really good
6:13
I finally pull the trigger on buying wireless earbuds, and then they invent a way to keep the wires untangled...
I love this paper. It really shows how repulsive curves can effectively fill a space with no intersections, like in our brain or digestive tract.
This would also be perfect for plant growth simulations, since plants adapt to their surroundings. an example of this is crown shyness, where trees will avoid stealing light from other similar trees.
exactly what i needed, i was thinking of nodes in editor softwares that can get very messy, with this you're good
This one folks-this channell is top of the notch. Thank you for yor videos. At least we know what are the possibilities for future. WHAT A TIME TO BE ALIVE👍👍👍
Károly, what are you doing to me?! I should be working and now I will be reading this paper on fascinating stuff I didn't know I needed in my life just 10minutes ago. And that's not the first incident of this sort. No, really thanks for your channel and work you're doing.
I don't know what any of this is but your commentary and explanations are just so interesting to listen to
Very Interesting concept I have not thougt of before! Also im gonna try to model and 3D print those last shapes at the end. They looked cool! :)
A video game based on this tech would be amazing. I have no idea how it would work!
That moment when you're to lazy to untangle your headphones so you build a whole Repulsion Simulation to untangle them for you
Oo, I was planning on building a new tower defense game that brings back the old td strategy of mazing. I'd love to be able to use this to help build the paths a.i. use to travel along your maze, but also maybe use it to build a.i. that can plan out mazes too
Excellent presentation as always! Would love to see this and fluid simulations used to simulate the physical properties of such things as soup with noodles and other stringy or flexible foods in full-immersion VR! I know it's still a ways off, but your videos make me excited for the day we no longer need traditional rasterized graphics and clunky headsets in order to enjoy virtual worlds!
thanks again for showing another great paper!
my immediate thought is that it can put to good use for FAA flight (drone or airplane) path planning simulation or self driving car!
I once saw a knot theory explanation and they said it was incomplete because once you generate a knot you can’t untangle it and find what knot it is, or if it’s a new one
Movement path planning is a neat application for it, but how quickly does it compute those paths?
this is an important piece, a discreet but fundamental advance in physical modelling, the question is what wouldn't i use it for, so glad to see many commenters see the potential also
i love the little among us at 1:04
With a background in psychology and neuroscience, of course my first thought is cortical sulci and gyri. I imagine other structures of the nervous system would be well modeled this way, as well. Very interesting, thanks.
Wow! That part where the video described pathfinding makes me think it can be combined with smart-car driving AI and logistics. Most people think about having cars communicate with each other but this makes me think about a server with this repulsion tool to map out the routes of thousands of cars at a time, using repulsion to stop any collisions. If you can add obstacles while it's mapping then you can also use it for cargo ships too so they can go around accidents
Excellent Paper!! I loving it!
I really liked how the second "trivial implementation" (@ 5:30) looked like some kind of alien being growing or something.
woah, topology is a weird field!
Amazing they're working out all this stuff!
I was interested in the collision avoidance application, surely this has great value.
"Is this it? Is this turning a sphere inside out?"
I'm not a scholar computer science, simulations etc but I just find it fascinating to watch and learn how ingenious AI and algorithms are solving problems and also the potential creative uses that emerge
This could be useful for designing printed circuit boards. Take your electrical schematic and component dimensions. Confine the movement of the components to a plane and use the repulsion simulation to figure out where to have the connections go, maybe adding constraints, such as minimising the size or fixing the locations of certain parts. As an example, if you're a retrocomputing enthusiast and you're designing a new ISA card, you'd fix the location of the pads for the ISA connector, limit the external connectors to the right edge of the board (but allow them to be anywhere along it), impose a maximum length and height and ask the program to try and minimise the board size.
Two minute papers, your skills in discovering and publishing entertaining and highly educational videos about advanced algorithms is nothing less than amazing. 😎😎😎👍
I stumbled upon this knowing nothing about neural networks or digital design, but this was neat to watch
I've been watching these videos for so long that I started making sure I have a paper handy so that I can hold on to it ans squeeze it as needed.
The ideas that are simple and have many applications are always the biggest breakthroughs.
Ive been working on problems related to building path finding graphs in 3d spaces with an aim of achieving near real time computation. The problem of filling an unknown 3d space with points while avoiding surfaces is incredibly difficult, and even more difficult to do quickly. I truly need to understand this paper.
A very neat little technique with many useful applications
Awesome 😎
This feels like it might have some use for simulating protein folding, but I don't know enough about that to be sure about it. Still cool as hell though.
‘I know how to take cuff of you’ *melts cuffs* magic wow it’s not like anyone gonna be walking around with a blow torch to do a repulsion method
Keenan crane is one of the best we have, highly recommend checking out anything he’s done…and for sure one of the most beautiful thesis papers I’ve come across!
Love Keenan Crane's entire body of work so much!
now if we reverse this repulsive force with attractive force without overlapping, we can simulate protein folding. also by adding a goal shape and train a neural network to achieve certain shape for the active region of the protein we have general protein solving model.
Shrink wrapping of point clouds could be amazingly useful for land surveying - volume calculations and contouring.
protein folding would be a cool application
It can be used for Fleet of Drones flying path planning. To create crazy shapes and an amazing show.
I just saw Belle by Studio Chizu. There is a meta verse equivalent in the movie called U. I think it could be used for their movies as they seem to investigate stuff like this, I think it could even be used for their visuals.
Assuming this technique can be refined to the point where it can work in real time, the non-intersecting curved paths part mentioned at 3:57 suggests an application in a self-driving cars network. It could almost eliminate the need for stopping and waiting at intersections, as speeds and trajectories would be pre-calculated not to collide. Entire traffic light systems could be replaced with an intersection computer that communicates with and directs autonomous vehicles in and around the intersection. Idk that’s just what stood out to me
It would be cool to see npcs walking through a town/village and using repulsion as both a deterrent for intersection as well as a measurement of distance before an npc needs to stagger or come to a complete stop before running into another npc, or even use repulsion to define a distance between npcs and players where npcs will path around or nod at/greet characters to create a realistic and immersive environment. Really cool stuff!
great video. i always fascinates me when a global problem can be solved locally. it reminds me of some research one of mine professeurs did, where instead of air traffic controllers the only rule basically was that any manoeuvres had to be away from the closest encounter. it was crazy effective, they could get multiple times the aircraft density of the busiest european airspaces, and pilots still complaining that there weren't enough aircraft and it wouldn't work if it's was busy (while being 2-3 busier than that).
The point cloud solution is pretty cool
im really excited for the future of shrinkwrap just one click sw would be so satisfying and time saving
The untangling the handcuffs part makes me think "Ah yes, very practical way of untangling some metal handcuffs... Melting them down and reforming them into a different pair of handcuffs with the same material in a different position."
I suppose for stuff like video games and min maxing things like shrink wrap efficiency it has uses.
I wonder if the shrinking method could be used to create more efficient Bounding Volume Hierarchies. There's a cost to compute the shrunken shape up front, but for complicated geometries - where the approximation of a bounding box or sphere doesn't give a good fit - this would probably be worth it.
It could maybe even be used to help with collision detection of deformable objects. Or maybe it could help with raytracing by minimizing the surface area of bounding volumes: start with your AABB, then run the shrink algorithm to give a better fit around your shapes, and do that for each node in the tree.
This reminds me a lot about that video called outside in. They try to invert a sphere without having infinitely small intersections
When seeing this idea I thought about Lagrangian liquid simulations (SPH for example). This kind of simulation can suffer from particle collapses or volume inconsistency which is sometimes fixed by the expensive computing of solving of the Poisson equation.
Maybe the idea of this article can find applications in Lagrangian liquid simulations to conserve volume.
the non-collision path stuff seems like it would be super useful for automated robots that move things around like in warehouses or self driving cars
1:22 make me remember about video titled "outside in" video describing about how a sphere turning inside out, very funny and fascinating educational video xD
i'd use this for generative art, it looks similar to differential line growth
I love this so much~! TY :)
this would definitely be awesome for pathfinding in games
That bunny with the growing tapeworm inside it is nightmare fuel.😨😅
6:06 I would try to turn hypercubes into single cubes to extrapolate how the time dimension forms our world. I have a feeling it's just swirling around until it looks like 1 cube. It might even manage to visualize 7 cubes next to each other from 1 hypercube.
The shrinking surface bag is the dream for surfacing high detail machines.
Wrote a repulsive force algorithm to visualise arbitrary tree graphs back in undergrad, and I can confirm that getting them numerically stable over time is not as trivial as you would hope :)
one of the applications can be imagine congestion-free continuous flow smart intersections where this can be applied to V2I and V2V connected vehicles
applications
Path finding is useful in conflict negotiations.
This would be perfect for path planning groups of units in RTS games! You could prevent the units paths from intersecting.
I just realized a potential utility for this algorithm being able to work on incomplete point clouds: sensory data! If we have cameras, rangefinders, sonar, etc. that are analyzing the obstacles around an object, this algorithm might provide an accurate and efficient way of creating a smooth mesh between all those points. What a time to be alive!
Can this be used to perform faster (or better) protein folding simulations?
Finally something I can understand, I'm an expert in repulsion, according to all the girls I've met in my entire life
I don't understand this at all, but I can't stop watching.
But as someone going into game design, this honestly intrigues me. Wish I has the brains to comprehend it though.
I work on how ants use regular meandering behavior to search an area efficiently. This simulation might be useful for search (drone/robot) algorithms in uncertain environments, like natural disaster areas.