Collision detection doesn't seem to work when a particle's velocity exceeds the object size / check interval
cyber_kinetist 5 days ago [-]
There's another paper called "Offset Geometric Contact" to be presented in the same conference, which solves the penetration issues with the VBD solver (and thus compatible with this paper): https://www.youtube.com/watch?v=xxyniqSLJik
The new GPU collision detection method guarantees no penetrations throughout integration for VBD-like solvers (although you can also use it with Newton-based solvers with less optimal results). Over the past few years the direction for graphics research have mainly been looking at accelerating the IPC solver (which theoretically guaranteed penetration-free simulation but was too slow to use for real-time simulation), but seems researchers have figured out an alternative path that doesn't have the same scale-up issues on the GPU. Although I think you would still need an IPC-like second-order solver for very accurate simulations (ex. for mechanical engineering), for games and VFX this is certainly good enough.
stephc_int13 5 days ago [-]
Is there any source code available somewhere for the GPU demo or the 2D web demo?
Gbox4 5 days ago [-]
This is cool, are there any game engines doing this?
yorwba 5 days ago [-]
This is the demo page for a paper that will officially be published in August.
RicoElectrico 5 days ago [-]
If ever, expect Roblox to implement this, per one of the author's affiliation.
maikjustus 4 days ago [-]
[dead]
Rendered at 08:47:24 GMT+0000 (Coordinated Universal Time) with Vercel.
The new GPU collision detection method guarantees no penetrations throughout integration for VBD-like solvers (although you can also use it with Newton-based solvers with less optimal results). Over the past few years the direction for graphics research have mainly been looking at accelerating the IPC solver (which theoretically guaranteed penetration-free simulation but was too slow to use for real-time simulation), but seems researchers have figured out an alternative path that doesn't have the same scale-up issues on the GPU. Although I think you would still need an IPC-like second-order solver for very accurate simulations (ex. for mechanical engineering), for games and VFX this is certainly good enough.