Garrett Johnson

The short experiments in #261 are showing slower speeds so tabling this for now.

This paper discusses accelerated collision detection: http://www.cs.cmu.edu/~djames/pbmis/etc/jgt98deform_AABB.pdf - recommends removing some of the SAT axis tests to speed up AABB-OBB tests because it doesn't need to be perfect ("SAT Lite")

Another paper on the topic: https://hal.inria.fr/inria-00539916/document - recommends minimizing the volume for collision detection

#316 added a basic `MeshBVH.bvhcast` based on shapecast but so far walking down both BVHs and iteratively compare node intersections has not turned out to be a win. See this...

If you're referring to this line: ```js geom.setAttribute('position', new Float32BufferAttribute(primitive.getAttribute("POSITION").getArray(), 3)); ``` It might be superfluous at this point. I don't think there should be any reason to transform the...

> Ouch that's news to me... any quantized or meshopt-compressed glTF models written by gltf-transform will have normalized positions, we should fix something there I think. gltfpack doesn't normalize positions...

Okay I see what you're saying now -- one of the purposes of having the the BVH plugin always available is to ensure that even after geometry transformations the BVH...

> Doesn't hurt to embed those construction parameters in the file, but for official extensions I also don't want to assume that the extension was authored by glTF-Transform itself. Right....

> To confirm – three-mesh-bvh would normally reorder the vertices itself? With the change in #294 it would no longer need to reorder vertices, but still needs to know what...