If you worked on making a plugin to the Unreal Engine for this, you would get tons and tons more traffic on this. Real-time water physics for small-scale games would quite literally revolutionize a lot of game-styles. I've looked at FLIP and SPH methods, but your MLS-MPM approach seems much, much faster. Pretty much all games use flat-plane geometry with oscillations of the surface to represent water-interaction. With the push into the "metaverse" idea, players have been wanting to interact with their games more and more (voxel games for example). Flat-plane water techniques fall apart fairly fast in a non-static world where the players can modify the terrain. The currently available techniques are limited to very small volumes or very slow run-times (examples below): SPH method ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-7ojILF1Dmro.html FLIP method ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-uW1z2taixQo.html
Though I do not really know about Taichi, It is really cool and so many efforts behind it. Now I am wondering if it can combining with neural rendering techniques...
Hello! This looks amazing, how could one implement this code in practice? and would it be possible to export the end product to a popular CAD modelling software in formats such as STL or OBJ? Thanks.
Github is tremendously popular in China, the second most starred Github repo is of Chinese origin. That being said, Yuanming Hu is affiliated with MIT which, last time I checked, isn't based in China.
Hello, @Yuanming Hu On what hardware was this simulation calculated? And how much time and memory did the miscalculation take? And how long can it take to render 100 million particles of water or snow or sand? As far as I understood, Elastic simulation is the least time consuming. Or the simulation of Sand.
No no no no!! This whole video gives you the impression that Taichi is a high-performance *programming language* when it reality it's just a *Python library* ... Even one Python function call per element is far too expensive to call this a high-performance ANYTHING, and I believe this video is extremely misleading. Shame on you for false marketing.
Please take your time watching the video - the Python function calls are just for constructing the AST, and the AST will be later compiled into high-performance kernels.
@@ethanymh I never finished watching the video, because it seemed interesting enough for me to check out the project on GitHub, before I found out about the Python thing. You also shared source code in the presentation that didn't look anything like Python. Would have loved to see a standalone language for this. The Python element really killed it for me.
@@LoganDark4357 The python frontend is added after this work and makes it easier to use by more programmers. You should really dig deeper before drawing very assertive conclusion. Spending 5 mins and concluding something would otherwise require hours of investigation is not a good strategy to become more informed. Question for @Yuanming Hu though, I really like taichi would like to use it at some point. However, having a company behind it in China makes me nervous. Will the developer/user community become exclusively Chinese and shuts off to the rest of English speaking world? Also, will the best talents to attracted to taichi if the community becomes overly Chinese-speaking?
@@justachannel6234 how many years of indoctrination have you been through lmao getting nervous and scared just because normal people from china are involved is not normal
I saw lots of MPM simulations for soft jelly-like objects, would MPM be suitable for simulating the fracture of a hard object (crystal) with a high Young's modulus? (Maybe a much smaller time step is needed?) Thank you!