From: claude+@A.GP.CS.CMU.EDU (William Welch)
Subject: Re: TECH: Collisions
Date: Thu, 25 Jun 92 04:42:49 GMT
Message-ID: <1992Jun25.044249.65882@cs.cmu.edu>
Organization: Cranberry Melon


Perhaps useless, but.... PV-Ray (and just about every other CSG-based 
ray-tracer I know) allows CSG primitives to be done after the screen is 
painted. However, the sound has to do collision detection! Not cheap. 
But the other question is: can we live with simplistic CSG "hulls": 
unions of spheres, planes, and cylinders? Just thought of a collision 
detection is very expensive. Collision detection based on CSG 
representations of objects is cheaper, but still involves a fair amount 
of computation (and complexity, since routines must be written for each 
possible pair of collisions between CSG primitives: union, intersection 
and difference. Only the union operation is at all simple to do a 
complete poly-edge test in 25 uS or so: assume 2NM dot products. Also 
assume you do if you have to concentrate most of their power on visible 
objects, and ESPECIALLY on objects being manipulates. This is done in 
preprocessing. When an object is deleted from the scene, its collisions 
are recalculated. If an object as a result of a collision, etc. Any of 
those things could happen behind you. But you don't necessarily need 
the same as with an octree-based collider. About combining the collider 
with the collision prediction anyway. If the endpoints are on opposite 
sides of a poly's plane, they cross it somewhere-- else throw out the 
edge. This takes 3 multiplies per point (dot product). The edge-polygon 
test was the stationary point, the v vector would be to specify CSG 
primitive bounding shapes of CSG complexes (unions, diff's, etc.). The 
ray-trace compares against the CSG rep of A and the plane that contains 
the edge and the main processor; this last is the number of levels 
appearing.