Can the genereated single or multiple convexs fully wrapping the original model?

[binghui1992]

Since I wanna use V-HACD for collision detection in industrial robotics, so the generated convexs must be fully wrapping the original model, otherwise there would be potentially severe industrial production accidents. Is there any guarantees or parameters that we can adjust to ensure satisfication of such criteria of V-HACD? Thanks in advance.

[jratcliff63367]

Well, the algorithm is certainly designed to do that, within some small margin for error based on the voxel resolution.

[binghui1992]

@jratcliff63367 Please see the picture below, as you can see obviously that the convexes (3 convexes) does not fully wrap the original model (the blue semi-transparent one is convex, the white one is orginal model), I'm using v4.0, the algorithm's parameters showed as below as well:

[jratcliff63367]

Gotcha. Could you send me a copy of your source model in Wavefront OBJ format so I can mess around with it?

I'm retired now and I don't really code anymore. I'm kind of hoping to sunset this library.

I would recommend you check out this library: https://github.com/SarahWeiii/CoACD

The authors specifically set out to make something better than V-HACD and, as best I can tell, they succeeded.

[binghui1992]

@jratcliff63367 Many thanks to your reply. I've upload the obj file via Google drive and below is the link. I'd be very willing to checkout the repo you mentioned and have a try.

https://drive.google.com/file/d/1-4ynhcM7cbGL43VlAU2JOB-jSmPWQL_P/view?usp=drivesdk

[jratcliff63367]

Ok, I will dive into this deeper if I need to. But the first thing to try would be to just increase the overall precision and see if that helps.

The default voxel resolution for V-HACD is pretty low, just for a fast quick approximation.

Try changing the voxel resolution to something like a million, or even five million.

The other precision setting to try changing is the volume error percentage. It defaults to 1% but you can try dropping it to say 0.1

See if a combination of a higher voxel resolution and lower error percentage fixes your approximation problem.